Nebius plans a bit more than 1 gigawatt a year from 2027.

That’s about 10x. And I own both.

The bottleneck Elon pointed at isn’t chips. It’s turbine blade casting. And he says SpaceX already has the most advanced casting shop on earth, built for its rocket engines.

If that edge holds, SpaceX brings power online 12 to 18 months before everyone else. That’s a big if, and I’ll get to why.

So I built a scoreboard: SpaceX, Nebius, CoreWeave and IREN. Who grows fastest in 2027, which pair I’m sticking with, and what would make me change my mind.

Let’s find out.

First, I am more and more convinced that this is the most important quote (it’s from Elon) investors should bear in mind in the singularity: “In tech the winner is not the company that is ahead at any point in time; the one who moves faster wins.”

And it looks like SpaceX is moving faster than anyone else in the AI buildout.

Elon, in the Q3 call a few days ago, pointed out that the ultimate bottleneck is now the gas turbine blade casting:

So we’re casting the blades and vanes for the turbines. So SpaceX already has a very advanced castings capability that was developed for rocket engines. So we have, I think, actually the most advanced casting capability on earth because the demands for castings for rocket engines exceed the demands for everything and anything else you could possibly imagine.

So taking that incredible team that has, I think, objectively, the most advanced casting technology in the world and saying, hey, can you guys also make some blades and veins for terrestrial turbines is much easier than what they have done in the past. And when we talk to the turbine suppliers, they tell us that the blades and vanes are 12 to 18 months behind the rest of the turbine, meaning they can deliver everything except the blades and vanes 12 to 18 months before they can deliver the blades vein. So casting is this really difficult bottleneck. And if anything, it seems to be getting worse.

So that means that we can bring power online 12 to 18 months faster than anyone else.

Of course, few can replicate that quickly at meaningful scale. Incumbents already cast turbine blades, and SpaceX’s own blade output at that speed is still unproven. If the edge holds, it’s a moat as long as demand stays overwhelmingly higher than supply.

Nebius, on the other hand, despite moving slow in comparison to SpaceX, is not moving slow compared to neoclouds.

Indeed, if we compare it to IREN and CoreWeave by using their Capex as a proxy for the compute they will bring online, Nebius’s higher sequential acceleration and speed are clear.

I know Capex is not a perfect metric to measure it, but for the purpose of this high-level comparison it is likely good enough.

Ready for Multibaggers? Unlock all my in-depth analyses and updates, including coverage of $NBIS, $MELI, $LMND, $HIMS, $RKLB, $ODD, $ZETA and $SPCX, by visiting Lorenzo2cents. Want the latest articles delivered straight to your inbox? Subscribe here for FREE and stay ahead of the curve.

Subscribe now

Now, let’s assume for a moment that SpaceX, Nebius, CoreWeave and IREN are all equally good at building and monetizing compute. With this assumption in place, the incremental revenue per year should roughly equal the incremental compute multiplied by the price of compute. The price of compute (assuming demand holds) is a sequential positive multiplier (the trend is positive, it’s increasing), but it is the same for all the companies because of assumption number 1 (all as good).

So what really matters for relative revenue growth is the incremental compute divided by current compute.

Now, because of the initial hypothesis, in 2027 SpaceX will bring online 10 gigawatts, Nebius 1 gigawatt, and CoreWeave (whose absolute Capex is still higher than Nebius’s, though the quarterly gap is narrowing as Nebius accelerates faster) about 1.15 gigawatts based on company guidance. On Capex, IREN looks like the least incremental spender. By the way, IREN may benefit from the reconversion of some mining assets and so need less capital to bring compute online, but again, here we are measuring the acceleration and speed of the buildout, and incremental Capex still makes sense with IREN too.

So let’s see the incremental compute divided by current compute, all measured in gigawatts, for 2027.

These GW figures aren’t strictly like-for-like (nameplate, active power, connected capacity, …), but they are good enough for this high-level relative growth comparison.

Nebius should end 2026 with connected capacity around 800 megawatts to 1 gigawatt (company guidance), and plans to deploy more than 1 gigawatt per year starting in 2027, meaning that in 2027 it is increasing about 2x to 2.25x (from 0.8–1.0 to about 1.8–2.0+ gigawatts).

SpaceX should end 2026 with compute around 3 gigawatts (Elon’s estimate), meaning that in 2027 it is increasing slightly more than 4x (from 3 to 13 gigawatts).

CoreWeave should end 2026 with active power of more than 1.85 gigawatts (company guidance), and guided for more than 3 gigawatts by end of 2027, meaning that in 2027 it is increasing about 60% (from 1.85 to 3 gigawatts).

For IREN, I don’t feel comfortable making a GW prediction because of what I have said above regarding the existing assets and the Capex picture so far, but the most likely outcome is still fast acceleration (maybe like SpaceX or Nebius) in 2027.

Here’s the scoreboard in one place:

Another assumption I make is that the growth of compute from 2028 on, in absolute terms, won’t be much higher for these companies than in 2027, as there is a limit given by physical resources (you don’t scale as in software), meaning that percentage growth will keep diminishing until it gets irrelevant in a few years.

In other words, although there are certain economies of scale like expertise and supply chain, returns to economies of scale are decremental over time, until you reach a tipping point where you keep building the same amount of compute per year or less (resources like land or natural gas are finite), unless you find a new breakthrough in tech.

SpaceX may be the exception here, thanks to orbital datacenters, which in my view should become a thing moving toward 2030 and could allow it to keep growing at high rates for longer.

There are a couple of other advantages that SpaceX has compared to other hyperscalers, and even more so compared to neoclouds.

The first is that they are selling the capacity dedicated to third parties mostly on short-term contracts priced near spot, which means that they will capture the full benefit of rising compute prices.

In contrast, neoclouds rely more on long-term, fixed-price contracts with hyperscalers to secure the credit needed to finance the Capex. This is especially true for CoreWeave.

Nebius also uses long-term investment-grade contracts, mainly to finance the build, while trying to limit them as they aim to become a hyperscaler themselves.

The second advantage SpaceX has against neoclouds is that they have internal demand coming from Grok, Cursor and Grok Bot, which is potentially offsetting a possible tempering of third-party demand that, even if not in sight, could still happen eventually. This is an optionality that I expect Nebius to enjoy more and more too, for the same reason I explained: they are focusing on getting their own end customers.

And demand isn’t what’s binding right now. Power is: Elon himself warned this week that a lot of AI compute could end up stranded in half-finished data centers.

Conclusion

My conviction in SpaceX is only getting stronger, and I believe that, together with Nebius, SpaceX is the best play to take advantage of the AI buildout.

With the Q3 earnings they reiterated they are on track to deliver their guidance of $100 billion in ARR by the end of this year, and Elon said that implies annual recurring revenue by the end of ‘27 somewhere between $500 billion and $1 trillion, which makes sense if you consider the 4x+ increase in compute plus the expected increase in spot price.

Ultimately, whether the bottleneck is the gas turbine blade casting, chips or anything else, Elon will go after it and build what is needed himself.

My ranking (not considering valuations): SpaceX first, Nebius second. That’s the pair I’m sticking with. IREN and CoreWeave come after.

As always, here is the “Deep Dive To Date” (DDTD): stock performance since my first deep dive and when I bought SpaceX on August 21, 2026, at $112.

+32% DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Personal views, not financial advice. I can be wrong, and sometimes I will be. Do your own research.

Whoever moves faster wins. Right now, that’s SpaceX.

AI and robots will make products, content and services almost free. So the bottleneck moves. Your time is finite. The supply of stuff that wants your eyes is not.

That makes the places that capture and hold real human attention the best real estate in advertising. Let me show you where it’s going right now, including agents, and who I think comes out ahead.

Social and video feeds

This is still the biggest sink. DataReportal’s Digital 2026 (GWI survey data) says the average online adult spends 18 hours 36 minutes a week on social networks and video feeds combined. Well over 2.5 hours a day.

Meta’s apps hit 3.60 billion daily active people in Q2 2026, up 3%. Growth is slowing. Time spent is not: Instagram time grew double digits year over year, and Facebook video time rose 9%. Instagram is now at 2 billion daily actives.

$META, YouTube, TikTok and X own the feeds, the data and the distribution. News sites and open-web display lose here. They can’t match the loop.

Gaming

Mobile gaming is mature, but people don’t leave. Sensor Tower’s State of Mobile 2026 says users spent 5.3 trillion hours in apps in 2025, about 3.6 hours a day per user. Gaming in-app revenue was close to $82 billion, up only 1.3%. Downloads fell, time spent rose a bit. Same players, deeper engagement.

Winner: . Losers: utility apps and passive browsing, because they don’t build the same loop.

Long-form video and CTV

YouTube keeps taking time. Digital i measured average daily viewing per account at 99.1 minutes in 2025, up from 87.2 in 2024. That’s more than Netflix, which dropped from 100.5 to 93.4. YouTube’s share of viewing on the TV screen went from 28% to 35%. The living room is the next fight.

GOOGL and Netflix are in the best position, along with any CTV player that can prove performance. Wurl ($APP) has a foothold, but it’s early. Linear TV is the loser.

Messaging

WhatsApp passed 3 billion monthly users in 2025. Similarweb’s Android data shows about 89% daily stickiness. People open it all day, even if each session is short.

$META wins this one almost by default. Broadcast-style social, with no real two-way interaction, doesn’t.

Agents

Fastest growing thing I can find.

Sensor Tower says time in generative AI apps hit 48 billion hours in 2025, about 3.6x 2024. Its State of AI 2026 projects 36 billion hours for H1 2026 alone, against 17.2 billion a year earlier. ChatGPT told us it has 1.2 billion weekly users (Sept 29, 2026). It was 400 million in February 2025.

Agents take the boring work: research, comparison, filtering. That hurts search. Chartbeat saw Google Search referrals to its publisher network fall 40% year over year to July 2026, after a 22% drop the year before. SparkToro’s study found 68% of US desktop Google searches ended without a click. Careful with that one: it’s browser data, and it’s one study. Search is still huge. But for publishers the direction is clear.

So the agent interface becomes a new place to sell things.

Winners: whoever owns or plugs into the leading agents. OpenAI, $GOOGL, $META, Anthropic, SpaceX. Losers: pure search, and shopping sites that live off long manual browsing.

Where ad value ends up

Ads are worth most where attention is dense, measurable and hard to hand to an agent. That means algorithmic video feeds, gaming and the big video platforms. People will keep spending that time themselves.

And a new surface is forming inside the agent layer. Whoever gets preferred by agents, or finds a way into the human-agent decision loop, takes a growing slice of money that used to sit in search. Everyone else gets squeezed as the noise grows.

Two ways to monetize attention

One: own the real estate where ads show up. $META, $GOOGL, TikTok, X.

Two: be better than everyone at deciding which ad goes where and at what price. The first needs scale of attention. The second needs data and a model that turns it into predictions.

$APP sits in between, and its edge is the closed loop connecting the two.

What AppLovin actually does

AppLovin runs MAX, a mediation platform, and AXON, a prediction model. A game publisher drops in the SDK. That opens ad slots in the app and lets MAX run a real-time auction across demand sources: AppLovin’s own, $META, $GOOGL, $U and others. The publisher gets paid. AppLovin takes a cut and sees every bid, the clearing price and, through the SDK and advertiser reporting, what happens after the ad.

AXON bids on behalf of AppLovin’s advertisers. It predicts how likely this user, in this app, right now, is to convert for this advertiser. Better prediction means smarter bids and more inventory won at a profit.

The closed loop

An impression opens up in a game. MAX sends a bid request. AXON reads what it can see: app, device, time, recent session behavior, how similar users did in similar contexts, past auction results. It bids. If it wins, the ad runs. Later the advertiser reports whether the user installed or bought. That outcome goes back into the model.

Because MAX runs the auction, AppLovin sees more than its own bids. It sees the clearing price and that others bid, plus the result. A typical demand-side platform only knows if it won or lost. AppLovin sees the shape of the market for that impression and what came after. That extra signal compounds.

So the model gets better at telling good impressions from bad ones. Advertisers with better returns spend more. More spend, more auctions, more feedback. Publishers get better fill rates and eCPMs, so they keep the SDK. The loop feeds itself.

Here’s Adam Foroughi, cofounder and CEO, on the Q1 2026 call (May 6):

The team improves the model, advertisers see better returns and they put more budget into our system. It is a virtuous cycle, and it is working.

Behavior over identity

The data in this loop is mostly behavioral and contextual. It isn’t a big deterministic identity graph. After Apple’s App Tracking Transparency, persistent cross-app identifiers got much harder to use at scale. AXON doesn’t need to know it’s the same person across ten apps. It learns patterns: people who act like this, in apps like this, at times like this, convert at this rate. Auction history and conversion feedback are the labels. Identity helps when it’s there. It isn’t the main input.

That’s why AppLovin can say its economics improve as the models improve. The inventory is mediated, not owned. The edge is pricing it more accurately than other bidders, with a feedback loop most of them can’t copy.

And AppLovin is big in this market. Sensor Tower puts $APP and Google AdMob together at roughly 65% of global mobile game ad revenue (Jan 2025 to May 2026), with AppLovin alone at 36% and AdMob at 29%. On the mediation side, GameBiz Consulting found MAX in 73% of top-downloaded ad-supported games and 55% of top-grossing ones (March 2025 data).

Where the next growth comes from

On the Q2 call (August 2026), Foroughi said: “Over the longer term, as we continue improving gaming and expanding consumer, we believe this business can compound at roughly 30% annually.”

My read: the easy part is taking the same mobile real estate to consumer and e-commerce advertisers.

The inventory stays the same, mobile game slots through MAX. What changes is the demand. Gaming advertisers already know how to buy here. Bigger consumer brands need clearer measurement, first-party data integration and proof of ROAS before they scale. Partnerships with analytics tools like Triple Whale and Northbeam give AppLovin a targeted way in. Same real estate, better demand.

On the Q1 call he was clear this is early:

In e-commerce and now what we’re calling consumer, this product is really early. So it’s like AXON 2.0, call it, 10 quarters ago. We’re going through the phase of not only rolling out a model and understanding what the consumer needs, but also then on the other side, getting more data into the system.

As we add more advertisers, we get more data, then we can build a more sophisticated model that can process it and create better output. So we’ve been continuously doing that. Last quarter earnings, I mentioned we just had one new model that had just created an uplift. The one we just had a couple of weeks ago was quite substantial.

On Q2 he added that consumer advertiser spend set another record, 28% above Q4 2025, the seasonal peak.

Streaming is where AppLovin adds new inventory. It bought Wurl in 2022 and got a connected-TV supply and measurement layer. The plan is to run AXON-style bidding on CTV. Attention is high there, attribution has been weaker than in mobile gaming. And CTV doesn’t have the dense in-session signals games give MAX. Early, but it widens the surface.

Then Gist. AppLovin published its own social app in May 2026, quietly. The idea is first-party, deterministic signals that feed AXON and don’t depend on third-party apps. It’s the most direct try at owning both the real estate and the closed-loop data outside gaming.

AppLovin already dominates the number 2 attention market (gaming) and wants into number 1 (social) and number 3 (long-form video).

Conclusion

What AppLovin has built is close to a digital twin of how attention works while someone plays a game. A model that predicts what grabs a human in that moment.

Now they want the same for social and long-form video.

For video, the plan looks like the gaming one: they’re the “Airbnb” of the real estate. Hosts own the rooms, AppLovin matches and prices. For social they want to own the real estate itself, so they’re building a network.

I think Wurl will likely work over time. Gist? I doubt it. A social network from scratch is less about using data well, which is their core business, and more about an art that very few have mastered in the last twenty years. And if it did work, it would mean hundreds of millions of monthly users pulled out of walled gardens that have been thriving for a long time.

But they don’t need either bet to win. The macro dynamics I laid out at the top are enough for me to believe $APP can grow 20%+ for the foreseeable future. Management talks about roughly 30% a year, and Q3 guidance sits at 46% to 48% growth.

On valuation: $APP trades at a trailing P/E of roughly 21 to 24, and about 15 to 17 on forward earnings. For a business growing like that, I think it may be severely undervalued.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Playing a bit with the numbers, my take away is that the improvement is driven by:

• SpoiledChild weighting more in the revenue mix (above 40%) as IlMakiage revenue drops

• SpoiledChild not affected as bad as IlMakiage by the algo dislocation

• Management investing more in SpoiledChild Customer acquisition in 2026 H2 vs 2025 H2 (remember that most of the Customer acquisition for Oddity happens in Q1 normally)

• Methodiq doing better than expected, ramping up its business

This is why I say so, in order.

SpoiledChild had a good quarter and a strong year to date, 2026 overall, and it is on track to grow at least 35 percent this year and approach $350 million of net revenue in 2026

$350 million of net revenue is 43% of 2025 ODD total revenue, which so start to play with IlMakiage 1:1 (IlMakiage rev is declining in FY26).

We believe SpoiledChild is being impacted by the algorithm dislocation issues IL MAKIAGE is facing, but to a lesser degree, and this has allowed us to continue scaling the brand

So it makes sense they are pushing the gas pedal a bit more than usual in H2.

Management is guiding for “net revenue will decline approximately 5 percent year-over-year”

Management has shared in the past that repeat sales is around 70% of total sales, so we can assume that il Makiage is declining 20% to 30% YoY (few new customers) in Q3 while SpoiledChild partially compensate it with its 35% growth. What is missing to get to -5% is made by Methodiq, which the management said doing better than SpoiledChild in year 1 after launch, which according to my estimation bring us to Methodiq sales in 2026 H2 around 15 $M per quarter. This is all incremental on 2025, as Methodiq was launched at the end of the year.

Confirming this scenario, there are both IlMakiage and SpoileChild web visits, which just keep declining both YoY and sequentially, which suggest that customer acquisition is not solved yet.

Oddity products are great

My take away here is that Oddity is becoming better to farming current user base, especially with SpoileChild, and also the products are great and Customers love it. Here the reviews growth from 2025 to date (please note that the numbers in the graph are not 100% accurate because I normally collect the data at the time of earnings, anyway data are good enough for the purpose):

Also, at the same the rating from 2025 to 2026 improved for all brands, now sitting average to 4,5 over 5, a great score for this kind of consumer goods company.

This numbers indeed supports what management is saying.

SpoiledChild continues to deliver very strong customer service cohorts metrics like AOV and repeated scale. 12 months net revenue repeat rates for the brand are well in excess of 100 percent today.

Oddity going physical

Yes, first of all, we learned a lot. The past few months were very intense in terms of media buying world. I must say that we thought that we know a lot, but now after those months, we are very deep in the details and learning every day better how those algorithms work. We increased our efforts both to fix the problem, but to make the business more resilient moving forward, including more distribution and more channels. We have nothing yet to announce, but once we have, we will. We believe that the key of the business is data, and in order to continue to have that ability, we need to remain focused. A big portion of the business must remain D2C. That is our strength, and we need to continue to work with those ad partners. By the way, the way that we work now with the ad partner and their commitment and like it is unbelievable

This comment to me suggests that Oddity is going either to distribute their products through digital or phsycal retailers (E.g. Amazon or Sephora) or is opening their own physical stores.

I would bet on the second.

Oddity main focus should remain D2C of course, which is the whole thesis, but I think owning a few selected stores in strategic locations (New York, London, Milan, ...) will be a net positive strategy.

On the negative of course there is the cost of operating the stores, on the positive there is:

• improved brand awareness

• physical distribution platform to acquire new customers, so diversification from Meta channel (which otherwise would likely remain the biggest D2C acquisition channel)

At the end of the day, if we assume that a new customer acquired through a store will then keep buying online on the D2C channel (which is reasonable to be assumed because it is convenient), the physical store operation cost can be considered as CAC.

Once CAC from the stores < than CAC from META, Oddity wins.

I firmly believe that, as long as stores remain limited to a few hundreds globally, this may be a winning strategy.

This is of course my speculation, there isn’t anything confirmed by Oddity, but I bet Oddity is thinking about doing this, let’s not forget that they already have a limited physical presence (<5% of revenue is coming from stores).

As always, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since my initial deep dive on the July 16th 2024, when the price was $44.24.

-58% DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

If it is true that share price follows free cash flow, eventually, and it is true, to be successful you need to predict where the free cash flow is headed, because once the direction is clear to the herd it is too late to get alpha from an investment. Well, I believe that, in tech, free cash flow follows speed of execution, which is the ultimate leading indicator.

Now, to Duolingo. Long story short, I don’t think they are moving fast enough.

First of all, I haven’t just sold Duolingo, I fully ported it into HIMS. I want to point it out because in a concentrated portfolio like the one I am managing, the total return highly depends (even more than a diversified portfolio) on how you allocate your capital over the few positions.

So the reason I made this move is that I judged HIMS at this point had a much higher asymmetry than Duol. And Duol was anyway a small position, so at a certain point either I doubled down or I allocated that money differently, as for me it does not make sense to hold a few percent position.

This quote is from Q2:

…this work is making Duolingo a better product. We believe that’s driving increased word of mouth and improving our top-of-funnel growth

This is a good summary of Duolingo’s management strategy for the last 9 months. They think that by making the product better, they will improve their top of funnel meaningfully. I doubt it.

As I explained in my article Duolingo: The Owl Has a Distribution Problem published on May 12, 2026, Duolingo reached peak search interest (measured with Google trends) last year around the Dead Duo campaign. That campaign was genius: impossible to ignore, and very Duolingo. But that is exactly the point. You cannot schedule virality every quarter.

After Dead Duo, interest did not keep compounding. It got worse, and MaU and DaU growth together went down to the lowest point.

Unless your thesis is that in the same period of time the product got worse, this proves that the top of funnel at this stage of the company’s maturity is uncorrelated (or not meaningfully correlated) with any product’s marginal improvements. Which signals that most likely the management strategy is not going to work out.

This is indeed confirmed by the interest over time, which over the past year only got worse.

You could argue that the recent DaU leaked on August 18 pointed to an acceleration to 27% YoY Growth, which confirms the reacceleration in bookings the management guided through the end of the year. Well, I am not surprised, because Duolingo is actually improving the product and farming the current user base, meaning that they are turning some MaU or inactive Users into DaU, and it is indeed working.

But new-user growth (measured with MaU growth as a proxy) has slowed a lot, which means they’ll hit a cap for the DaU, eventually (DaU is a subset of MaU).

In fact, looking at the sequential growth of MaU and DaU in the past, MaU led DaU all the way down from Q1 2023 to now, as it is logical to be, and it is still in a downtrend, with no sign of significantly reaccelerating.

Over the past year, we’ve built a stronger team and a more disciplined operating model, tripling both the number of channels in our mix and our performance creative output globally. We’re still early, but we’re already seeing this show up in top-of-funnel growth, and we’re increasingly confident that performance marketing can become a meaningful and efficient driver of long-term growth.

While they are clearly working on some levers, like performance marketing, I expected them to move much faster than this, experimenting with some game changing improvements like:

• introducing multiplayer for language learning, to generate a network effect

• introducing new courses beyond languages, math, music and chess, to increase the top of funnel surface

Duolingo has basically no competition now, and I think they are dissipating an important advantage which won’t last forever.

I am sure performance marketing as well as influencer-led marketing in China will bring results, but I am not any longer confident that this company has the potential to 10x in a few years, which is what I am looking for.

Of course, I may reconsider my thesis again as I get new inputs.

And here is how the DUOL stock performed from my initial deep dive on the January 11th 2025, when the price was $318.15, to when I closed the position on the July 24th 2026 at $122.10.

-62%

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Annual contract value per MW repriced higher. Our core AI cloud business won deals averaging above $20 million in revenue per MW, driven by increasing prices for new-generation GPUs and more than 30% higher pricing on older-generation GPUs versus Q1.

• Deal sizes and new logos set records. We closed four landmark deals, which averaged more than $1 billion in TCV each.

• TCV of deals closed in Q2 grew nearly 4x quarter-over-quarter, with TCV from new customers growing more than 9x.

The new Astra GPT-6 model released by OpenAI looks like an early version of AGI (I know, it was nerfed a few days ago, but this does not change my point, actually it reinforces it confirming insatiable demand for the model and Open AI which cannot keep up).

It can basically do everything you can do on a PC.

In addition, on 8 September, OpenAI said an unreleased model stronger than public GPT-6 Astra produced a claimed proof of the 3D Navier–Stokes existence and smoothness problem, a Clay Millennium Prize question, unsolved by mathematicians since it was posed in 2000, and open in substance since the 19th-century equations.

For me it is now clear that we are entering the singularity: AI has surpassed human intelligence, at least according to some benchmark, and will soon be designing and training the next model autonomously, and the chips it will run on, making the acceleration even faster through recursive improvement.

Here a couple examples of how an early version of this is already happening:

Google’s AlphaChip (reinforcement learning for chip floorplanning) has produced layouts used in multiple generations of Google’s TPUs, including v5e, v5p, and Trillium (6th gen), plus Axion CPUs. It generates layouts in hours that used to take human teams weeks or months, and Google describes those layouts as superhuman or comparable. The system also gets better as it sees more chips — it pre-trains on earlier generations, then places more of the next generation. Those TPUs are what train and serve Gemini.

Anthropic’s June 2026 report: When AI Builds Itself:

• As of May 2026, more than 80% of code merged into Anthropic’s production codebase was authored by Claude (low single digits before Claude Code in Feb 2025).

• Typical engineers merged ~8× as much code per day in Q2 2026 vs 2024.

• On an internal training-code optimization test, Claude went from ~3× speedup (Opus 4, May 2025) to ~52× (Mythos Preview, April 2026). A skilled human in 4–8 hours typically reaches ~4×.

• On one open-ended AI-safety research problem, Claude agents closed 97% of the available performance gap; two humans working about a week closed 23%. Humans still chose the problem and wrote the rubric.

Also, Astra is not the latest model available, Open AI kept their best model internally for now, which means that closed frontier labs still look to have a considerable edge on open models, likely 6 to 9 months. That is bullish for the whole ecosystem, because it means the infrastructure buildout — heavily driven by the 2 frontier labs revenue — is likely sustainable. In other words, Anthropic and OpenAI are not going to zero any time soon.

For companies like Nebius, which thrives on open models, it means we are nowhere close to a plateau in capability improvement. Demand for AI will keep skyrocketing.

Demand could actually accelerate. As AI surpasses human intelligence, the cost of doing nothing becomes unsustainable. Investing in AI is the only alternative to going out of business for most companies.

The waiting strategy is a wrong strategy

As a Nebius investor I reflect often if the $IREN waiting strategy often cited by Iren bulls (not sure this is really the management strategy) - waiting to get contracts at higher prices — will pay off more in the long term.

But the tech game is actually about learning faster and steering better than competitors. Elon once said that in tech the winner is not the company that is ahead at any point in time; the one who moves faster wins. By that standard, Nebius is moving at light speed: executing, learning, and building a hyperscaler. Sure, Iren may sell its power at a higher price, but this way it remains a bare-metal supplier, successful as long as supply and demand don’t rebalance, which may take years or even a decade, but likely not able to build a durable moat.

Multiple reasons for that.

First, every customer you don’t get now is statistically a lost customer forever.

These are the characteristics of a hyperscaler cloud service:

• Very low churn: Hyperscale cloud providers generally experience low customer churn because their globally distributed infrastructure and deep service ecosystems make migration complex and costly.

• Long-term contracts: Providers use multi-year commitments (3 to 10 years) and consumption discounts to lock in enterprise workloads, heavily reducing short-term customer loss.

• “Stickiness”: Once an organization embeds its data pipelines, storage, and AI training and inference workloads into a specific hyperscale environment, moving to another provider creates friction and high switching costs.

Once someone else gets that customer onboard, it is unlikely you steal it later.

Further on this point, I would argue that those “champions” Customer able to embrace a native AI mindset soon enough will likely become the leaders in their industry and win a large share, if not most, of the market.

Think about Revolut, building its AI capabilities on Nebius and clearly on a trajectory to become one of the largest fintech players globally. These companies build a learning flywheel for their products. With AI that flywheel just spins faster, and for a long time they become almost uncatchable. Once Nebius has secured that kind of elite companies, a big part of future revenue growth is already locked in.

This is a clear example of it happening:

Higgsfield, one of Nebius’s first AI cloud customers, has expanded its usage on the platform by >20x since their first contract. This summer they premiered a 95-minute feature film made entirely with AI on Nebius infrastructure — an industry first that compressed a two-year process into weeks, at a fraction of the cost of traditional filmmaking.

Second, and connected to point one: get the flywheel in motion and accelerate it.

At our June 9 Inflection Event, we launched our Customer Advisory Board, with AMI, Black Forest Labs, Cloudflare, Cognition, Cohere, Core Automation, Higgsfield, Recraft, Revolut, and Rhoda. This forum brings CEOs, CTOs, founders, and industry leaders together to sharpen our customer feedback loop, and keep Nebius at the forefront of AI development.

You learn from customers using your product, and you make it better. Every day you stay out of the game is a gift to your competitors.

Third, once you are delivering consistently you are likely getting some loyalty from your suppliers, which is crucial in a constrained supply chain.

Nebius is indeed taking a balanced, common-sense approach by saving only some of the capacity for short-term, high-priced deals:

We could sell our entire 2027 capacity on these terms today. We are deliberately not doing so because we see higher value in retaining some capacity for immediate customer needs

And this is the contract mix that confirms Nebius is optimizing for becoming a hyperscaler rather than maximizing price per MW:

• Shorter-duration contracts, typically between three and six months, is a new effort, for customers with an acute, timebound need — priced at a significant premium. One such deal has been signed in Q3.

• Mid-term contracts — our core business — now with an average duration of one to three years with the world’s most ambitious AI companies.

• Long-term contracts with investment-grade customers, which help us finance our buildout faster and more efficiently. The $775 million secured facility we raised in July was on the back of one of these agreements — and with $40 billion in customer commitments, we will do more of this.

Asset light model is genius

We launched a model that allows partners to deploy Nebius’s full-stack AI cloud platform in their own AI data centers. Nebius contributes:
• Our full-stack AI cloud software,
• Our systems architecture and reference designs, and
• Our global go-to-market organization that brings the demand.

Under this model, partners get fully-owned AI infrastructure assets, designed to Nebius standards, and a fast route to serve the AI cloud market; Nebius converts partner-financed capacity into high-margin revenue with minimal capital outlay. This complements our owned and colocated portfolio while reducing the capital burden of adding capacity for customers.

This strategy is simply genius. It lets Nebius deliver its customer experience, hold the customer relationship, and build a moat based on software and vertical integration without the burden and the risk of the full infrastructure cost. It basically becomes an asset-light business model, able to scale much faster, without carrying all the buildout risk.

This model addresses the 2 constraints of our industry, which is capital and capacity. Partners finance, build and operate the facilities, whereas Nebius brings the full-stack platform and the demand. We provide value-added services that sit on top of our partners’ infrastructure, which delivers us with high-margin revenue and requires minimum balance sheet capital. This model has the potential to unlock new capacity for us in 2027 and beyond.

Arkady (the founder and CEO) explained.

If this works, I would bet Nebius becomes a $1T company in a few years, thanks to two dynamics at play:

• Revenue can keep growing at a very high rate for years, as the buildout constraint becomes lighter

• Valuation can get a higher multiple because the risk is lower

How do we know if it is working?

I’ll monitor adoption.

The first KPI is Nebius announcing the new model is starting to contribute meaningfully, for more than 100 megawatts. Once it happens, we know the business model is working.

And according to Dado Alonso, the CFO of Nebius, it looks like they are getting early traction:

Revenue was driven by capacity added in Q1, higher utilization and high-margin revenue from our new asset-light business model, Token Factory and our recent acquisitions

From there, it is all about acceleration.

The second KPI is what percentage of added capacity comes from the asset-light model. Once it crosses 50%, we know Nebius did it. From there on, it is all about watching free cash flow rise exponentially.

we expect our asset-light model (in 2027), along with high-value services such as agentic and inference solutions, to contribute an increasing share of revenue while supporting even higher margins

Nebius x Palantir

Palantir named Nebius its preferred sovereign AI infrastructure partner on 8 September 2026. The two companies announced a strategic partnership to give Palantir’s commercial customers access to Nebius’s AI-native cloud and inference capacity from inside Palantir’s software environment.

After an integration period, Nebius compute and inference endpoints will sit inside Palantir’s “enterprise perimeter.” Eligible customers will be able to:

• Run and adapt open models on Nebius infrastructure

• Train or fine-tune those models on their own proprietary data

• Keep control of compute, data, and the resulting models rather than renting capacity from a general-purpose public cloud or relying on closed models

Palantir said it chose Nebius because the platform was built for AI from the ground up (not adapted from general-purpose computing) and can be integrated directly into Palantir’s stack.

This is the most bullish news I could get, even more bullish than a billion-dollar deal with a hyperscaler.

These are the main reasons:

1- Palantir is the best possible distribution channel to sell to enterprises → Demand
2- Palantir is the best possible partner to sell to enterprises → Nebius getting the customer/partner feedback loop needed to develop the strongest enterprise-grade offering
3- Palantir selecting Nebius is the confirmation that Nebius has the best infrastructure and software stack for AI → My original thesis is on track, Nebius isn’t a neocloud, it’s rather becoming the AI hyperscaler.

By the way, I posted this on X on July 3, two months before the partnership was announced.

You can say you heard from me first ;)

SpaceX for Nebius

And this is my next call. Take a minute to read my post.

Conclusion

I couldn’t feel better about where we are with Nebius. All signals and news just confirm my original thesis, and if you project it on numbers, I would bet Nebius becomes a $1T company sooner than what everyone thinks.

After Nebius management will share their 2027 guidance, which I bet will be around $27B ARR, I will elaborate and share my 2030 target price.

As always, here’s the Deep Dive To Date (DDTD): Stock performance since my first deep dive and when I bought in on July 28, 2025, at $51.63.

3,3x DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Demand for Athena, the recently released voice-based OS, is confirmed to be strong:

Since Athena’s launch for enterprise customers 130 days ago, more than 40% of our super-scaled customers are already monthly users. And together, they’ve generated thousands of campaigns using Athena. And customers who comprehensively adopt our AI, using it for audience creation, activation and other means, contribute a disproportionate share of revenue. This shows up in several ways.

For those who have worked in a B2B software sales role, the extraordinary result of this will be clear. When you launch a new product, it normally takes months to get someone to buy in for a pilot, and it can take years before the related sales have a meaningful impact on revenue.

Nobody likes to be the first, unless the ROI is crystal clear and provable.

Zeta, in less than 6 months, has been able to get almost half of its biggest customers to try or use it.

Athena is having a great impact on engagement too:

Over the last 60 days, Athena’s voice users are interacting with the platform at a 500% higher level than non-Athena users.

This is great for two reasons. The first one is the obvious one: higher ROI, as more users are able, as explained by David several times, to use the platform better and deeper, getting more juice out of it. This ultimately turns into higher retention, more spending on utilization fees and eventually buying new features.

The less obvious one is that as the UI becomes less relevant, the value accrues more and more on the ROI which the system delivers, and for which Zeta is recognized as the leader in their domain.

This is also a further tailwind for Zeta’s M&A strategy: buying a new company and integrating it into the value proposition is easier and easier if the UI is not a thing.

Zeta Business Intelligence

But let’s go to the main point of my article, which is about Zeta Business Intelligence and its partnership with Palantir.

What was once called Business Intelligence is, in my opinion, where 90% of software value will accrue within the next 5 to 10 years.

Software, UI and UX, once the bulk of a software company’s value, will be worth almost nothing, as they can already be generated almost for free.

Ecosystems (customers, partners, distribution, ...) still matter a lot, I do agree on this. But let’s keep them aside for a moment.

What products should they be moving into retail? How do they use co-op dollars and marketing dollars to get better shelf space and how to create deeper and more meaningful relationships with the end customer? So it starts with the business intelligence around a retail use case, you then end up identifying meaningful opportunities for them to move inventory around geographically, which retailers they should be doing more with, maybe which retailers they should invest less with, and then we end up being able to market to the end user to drive them into the retailer on behalf of the product and vice versa.

These are the kind of questions which, if answered in an accurate, data-driven way, can really move the needle.

When a decision is strategic and not only tactical, the consequence of such a decision is spread across the organization and the supply chain, and the difference between making a good or a less good decision is measured in millions of dollars.

Zeta understood it for a long time, but what is changing now is that they are enlarging the pie, because it is becoming clear that the utility of ZETA’s marketing ontology is not limited to marketing. It can deliver tremendous value if paired with an ontology of the rest of the business, by answering questions which extend beyond the marketing domain but which require information from the Zeta domain to be answered.

This is David Steinberg giving an example during the call:

And if you look at this very large sports league that is using us, we were in the room with them and their CMO and they’re like, listen, how do we better negotiate our streaming rights with the two partnerships that are coming up for bid and how can we equate viewership and mind share for their—they own multiple products, multiple leagues that are owned by one holding corporation.

How do we help them to better show it’s not just who’s watching it in that moment, but it’s the halo effect that comes for the broadcaster by having them on? And we were able to put together a solution in hours that came back and they believe will result in millions of dollars of incremental revenue to them on the renegotiation of those streaming contracts. So it then becomes a flywheel because then we get part of the marketing for those streaming rights. So it really becomes a major flywheel in and around the company...

And what is the best way to accomplish an extended ontology, which is worth much more than the two single ontologies taken alone, if not joining Zeta’s ontology with the king of all ontologies?

And here we are, a partnership between Zeta and Palantir, which was announced in June.

As it should be clear to everyone by now, Palantir is becoming the new OS for businesses which aren’t joking, and Zeta understood it so clearly that it re-architected its Data Cloud on top of Palantir’s Foundry and adopted its ontology.

Our partnership with Palantir significantly expands Zeta’s enterprise opportunity. Palantir provides the ontology, governance, and enterprise AI infrastructure that large organizations require. Zeta contributes proprietary customer intelligence, proprietary data, identity, decisioning and activation. And we have reached two important milestones in this partnership already. Our data cloud was fully integrated with Foundry as of July 31. And we have already received multiple agreements for our initial combined sale with several other meaningful opportunities in flight.

said David Stainberg.

I believe that the ROI the two companies can provide to customers is not even comparable to a standard software company, and we are already starting to see the first proofs of it coming in. These are still two initial combined-sale engagements and test cases, so the sample is tiny:

The two deals we closed were at a 100% hit rate. Met with two, closed two.

A 100% hit rate even though on two opportunities is simply insane.

Conclusion

The surprises from Zeta are unlikely to end here. I expect more over the coming quarters.

On April 28 I opened a position in Zeta at the price of 18,26 $ per share, as I shared in real time in the Business Ontology community.

Hence, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since I opened my position.

+67% DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Here are the reasons why I think it just makes a lot of sense.

It is going to be the largest hyperscaler and it is cheap

Management is guiding to about $100B ARR (including Space, Connectivity and AI) by the end of the year, which at the price I bought is around 15x Price to ARR. For a company which can “easily” keep growing double digits for several years, it’s not a high multiple.

Elon is projecting compute capacity closer to 10 GW than 5 GW by the end of 2027. At pricing in the latest disclosed leases - about $50M per MW on the Anthropic/Colossus 1 deal, with Google paying $920M a month for a large GPU block — and assuming part of the capacity will be used internally (Grok training and inference), this can generate, together with the other businesses, $300B+ ARR by the end of 2027 (SemiAnalysis estimate, not company guidance). If that happens, at the price I bought, SpaceX stock will have a 5x Price to ARR multiple at the end of 2027. Pretty cheap.

Elon and the SpaceX team have already proved they can build and bring compute capacity online faster than anyone else, with 122 days to bring Colossus 1’s ~300MW online and Colossus 2 ramping even faster toward multi-hundred-MW and gigawatt-scale clusters. There are no reasons to believe they won’t keep delivering. The historical disadvantage compared to other hyperscalers was the smaller balance sheet, compared to the enormous AI capex bill.
Elon recently said they will build exclusively on Nvidia’s Vera Rubin stack, which I think may help with supply and partnership economics, with Nvidia financing part of the deal.

Tesla dilution

I own Tesla. It’s around 7% of the Lorenzo2cents portfolio at the time I write.

Tesla and SpaceX are more and more integrated and their destinies are tied. Think about Terafab, Grok for Tesla’s cars, SpaceX tech for Tesla Roadster, Tesla solar panels and batteries for SpaceX datacenters, Tesla’s Optimus for moon colonization, and more certainly to come.

I think they will merge, eventually. It would make any synergy easier, including for Elon’s governance.

One of the biggest concerns about this obvious merge is the potential dilution of Tesla shareholders, as Elon will want to maximize his control of the merged company and somehow will make a better deal for SpaceX rather than Tesla. Elon’s ownership of SpaceX is higher: he has voting control — roughly four-fifths of the votes through super-voting shares — while his Tesla stake is only around 15% of shares.

One way to mitigate this risk is to own both $SPCX and $TSLA, ideally in equal parts. This way, if you are going to be diluted on one side, you are the recipient of that benefit on the other side and the result is neutral. Of course, for this to make sense, you need to get SpaceX at a reasonable price before the merge becomes priced in, which I think I did.

Grok

Grok is basically valued at zero so far, and one could say fairly, as it has been far from frontier-model intelligence and far from capturing a meaningful market share among enterprises or consumers, with Anthropic and OpenAI hitting all the headlines and benchmarks, for good reasons.

I think this is going to change and Grok is now underestimated.

Grok 4.6, released on August 12, is back in the frontier conversation. It is competitive on some agent/coding and cost-performance comparisons, even if it is not clearly #1 versus Claude Opus 5 on the main public leaderboards. Grok 4.7 is expected next - Elon has pointed to roughly early September after the 4.6 launch - and he is promising another push toward the top of the frontier.

It will likely be leapfrogged by Anthropic after a few weeks or days, so likely xAI is still behind at least 3 months to the leading frontier lab.

But:

• Elon is now back in the race and he doesn’t like to be the second at anything, so I would not be surprised to see Grok taking the lead in a few months.

• xAI will likely have the largest compute available to any frontier lab for training new models, which is a big advantage

• Elon disclosed that xAI/SpaceX is putting decades of SpaceX engineering data into upcoming Grok training, which in my view can give it an unfair advantage against competition, positioning it to become the go-to model for specific engineering domains.

• I don’t see why, eventually, Tesla’s engineering data shouldn’t also be used to train Grok models, reinforcing its lead in the engineering domain

• X data were and will increasingly be an advantage for training new models as Grok catches up

• I expect Tesla cars and Optimus data will also be used to reinforce Grok’s understanding of the physical world

• Owned distribution channels for Grok, beyond Grok apps: Cursor, X, Tesla’s cars, Optimus (forward looking).

All in all, I just see it as likely for Grok to become one of the leading models, and with Anthropic and OpenAI already valued near or targeting the trillion-dollar tier, this is not priced in yet.

Space and Connectivity

The revenue generated by Space (launches and satellite services) and Connectivity (Starlink) is clearly already priced in, including the discounting of future growth.

But the value of these two assets resides elsewhere, beyond the revenue.

These two capabilities are highly synergic with everything else Elon is doing and are basically positioning him to have an unfair advantage in the long term. Here are some examples:

• Elon/Tesla are integrating Starlink into Cybercab (the purpose-built robotaxi). Tesla has already shown Starlink V5 built into the design, which should help connectivity and fleet operations even if autonomy itself still runs on the car. It will likely do the same with Optimus and then with all Tesla’s cars. Can competitors do the same? Sure, but for now they need to rely on Starlink to accomplish that. Even when competition catches up, they will still be paying a fee to whoever supplies the service.

• Orbital datacenters may be 5 or 10 years away, or more, before becoming anything meaningful, but yet, I think they will happen. I consider this an upside, but it alone could be worth more than any other company’s product or service in the world right now.

• SpaceX is the gatekeeper to space, which I believe will become, together with AI and longevity, the biggest industry in the world. Even when Rocket Lab, the only credible competitor (which by the way I own), successfully scales Neutron, the competitor to Falcon 9, SpaceX will still be shipping most of the mass to orbit with Starship and will be the only one able to enable specific use cases like orbital data centers.

Conclusion

I believe $SPCX can be a $10 trillion company in 5 to 10 years, and I believe its downside is limited at this valuation, while it also hedges my $TSLA position. This makes it a good fit for the Lorenzo2cents portfolio.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Before doing that, let me highlight one positive insight from the first release of the new AI features to some cohorts: AI is making the service more engaging and improving retention. This is excellent news because it derisks the whole thesis. One of the most critical factors for Hims’ success is increasing subscribers and scale. Once it has scale and organic growth, almost everything else follows as a consequence.

… we spoke around how with a more immersive experience through the next evolution of the app and some of the AI tools as well as physical tools like the scale of the consumers are getting, that enables more connectivity with their providers. And ultimately, we’re already seeing signs of lower cancellation rates for the cohorts of users that are receiving that. And as we look to continue invest in the experience, and we see opportunities as we gain more efficiencies to make price points more affordable, we see an already solid retention rate getting stronger and stronger.

On the same topic:

Patients are interacting and messaging with their care teams 3x as much with the new care operating plan and operating system that we’ve rolled out. And at the same time, the actual tasks for humans has dropped 50% when you’re talking about nonclinical tasks

Hims Is Planning to Become the Backbone of the Healthcare Industry

We are excited to bring the experience to more of our customers across the world, and we think we are moving closer to a future where the model we are building becomes the backbone for health platforms beyond Hims & Hers.

Mohamed ElShenawy, the CTO recently hired to build the Hims AI infrastructure, explained this during his AI update.

This is not a random comment that can be dismissed as buzzwords or hype to reinforce the AI thesis. This is Hims explaining in plain English what the company’s ultimate goal is as it works to accomplish its mission.

Becoming the backbone of the healthcare industry.

As I have envisioned since my first Hims write-up more than one year ago, and reiterated in my Q1 update, Hims is an AI company, and its value is in the closed loop. I will not describe it again; just read the paragraph “Hims is an AI company and providers are training the model” in my previous article.

Hims is actually pushing it even further, trying not only to create the only vertical ontology for healthcare, but also to create the layer where all healthcare actors will operate.

Andrew Dudum elaborated on it better later in the call:

…building the operating system and kind of this AI platform that can connect to pharmacists, a doctor, care coaches, agents all together in that environment, we thought was the best place to start

Having the best longevity and healthcare AI model in the world - which they will, if they can keep executing for a few years - is already a strong incentive, but it is not enough. Big Pharma is not moved by good intentions. It is moved by making more and more money, whatever the outcome for patients.

Andrew understood this, of course, and is trying to make them happy enough to join the system:

• Hims’ platform becomes a distribution layer for Big Pharma (see the Novo partnership). The bigger it gets, the higher the FOMO to get in.

• Hims does not compete with Big Pharma because it quit mass compounding.

The next obvious step, in my view, will be for Hims to sell anonymized data to Big Pharma and biotech companies for their R&D.

• What are the most common and painful problems suffered by patients?

• How much are patients willing to spend on them?

• What are the most likely solutions to such problems—APIs, proteins, molecules, and so on?

• What is the addressable market?

Such information is worth billions.

Here is Andrew telling you in plain English during the call, when addressing the Novo partnership:

I think there’s an increasing amount of collaboration around key strategic markets that are valuable to both businesses and thinking about how to collaborate into expanding the market and getting people more access I think there is increasing ability to share data with regard to what’s working for patients, adherence benefits, what we’re seeing with regard to side effects or how to mitigate that in different dosing regimens.

So I think there’s a really powerful flywheel here when you start getting the leading drug manufacturers actually closer to the data on the ground of the consumers in a way that can not only help when it comes to commercializing therapies but also thinking through how we bring new therapies to market. And so I think there’s a growing set of opportunities that I think both of the teams are very excited by.

The Freemium Model

I think this also sets up a really powerful freemium offering for Hims & Hers, where patients could just have the opportunity to come leverage these tools at Hims & Hers in a free manner to just get benefits of the 360 view of their health, a unified platform with an ecosystem, all powered by AI on the bottom layer of actual doctors, actual trained nutritionalists as well as agents and care pilots. And so when you really step back, like what is this that we’re building is currently available. This is what most would consider concierge care today. And it costs anywhere from $50,000 to $150,000 annually. I think in the simplest way, my vision is to make that same level of proactive, preventative, always-on care affordable to everybody globally for a price that is universally accessible.

Andrew is basically thinking of delivering a Duolingo-like model, where anyone can join for free, dump all their health problems into the app - every person on Earth has a lot of them - chat with AI about them, and actually get contextualized answers instead of only a recommendation to consult with a doctor.

You may be able to use the Hims app for free to track your diet progress and blood analysis, and get useful insights and recommendations.

In other words, a no-brainer.

From there to buying a product or service, the step is very short.

Combined with the already good marketing ROI Hims is delivering, this could be a total game changer for Hims’ top of the funnel.

Conclusion

As I correctly predicted two quarters ago, growth is reaccelerating. After a beat and raise in Q2, I expect another beat and raise in Q3.

Why do I say this? Because I noticed some patterns in how management presented the guidance. I will point you to my post on X for more details.

The only thing I did not like about Q2 was management not being fully transparent about guidance and subscriber numbers excluding Eucalyptus. This is not to diminish the importance of the Eucalyptus acquisition or to say that it was not a good thing. It was. But it makes it impossible to track organic growth and retention, which is crucial for a company like Hims that is betting the house on its ability to acquire new customers and retain them.

Besides that, I think we will have a good end to the year. Peptides are coming, the Novo partnership is bringing a lot of new business, and the international business is growing fast. We could not ask for better.

As always, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since my initial deep dive on the February 8th 2025, when the stock price was $42.54.

-36% DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

This is my opinion, of course, and I can be wrong, so do not take this as financial advice.

Developing a thesis about an investment, owning it and monitoring it is tougher than one might think. That’s why I have developed my method, which I call the Business Ontology Framework.

It helps me think from first principles and identify the relevant metrics to monitor that are, as much as possible, leading indicators of how the thesis is unfolding.

I am going to explain in this short write-up what I think matters and why. First, my thesis: Lemonade is going to become an autonomous company, and this will allow a winner-takes-all scenario.

I’ll keep it simple and short here, as you can find a lot more details by checking my past articles on the matter.

Follow me in my first-principles thinking here. To become an autonomous company and thrive in the insurance market, Lemonade needs to:

1. Make human employees more and more productive over time, let’s say at least 2 times more productive than the most productive competitor

2. Be able to acquire customers profitably for decades

3. Be able to retain customers

4. Keep risks under control

That’s it. Really, that’s it.

If Lemonade keeps doing these four things better than others, it will become a $100B company in less than ten years, and a $1T company in less than 15 years.

Everything else doesn’t really matter. It’s noise.

Let’s now look at how to assess if Lemonade is on track to deliver in Q2 and onward.

Make human employees more and more productive over time

There is really one simple metric here that tells us if Lemonade is on track, and it is

In Force Premium (IFP)/Employee

Here is how it printed in Q2 and its trend over the last quarters.

It must keep going up. Anything else doesn’t matter. Period.

You may see headcount or salaries increasing; it doesn’t really matter. In the long run, what matters to validate the autonomous company thesis is how productive employees are.

How much they are leveraged by technology.

Financials will inevitably follow, eventually.

Now, the IFP/employee trend is for sure a leading indicator, but if we want an even better one, it is probably the LAE ratio. Even though it is less complete, as it covers only part of the business, it is a useful proxy for claims-handling efficiency. Lemonade says the improvement is being driven by the growing use of its OS and AI across claims operations, so I see it as evidence that automation is increasingly supporting the business.

If Lemonade is achieving this trend in claims management, we can reasonably assume that something similar is happening in other areas of the business, such as risk management and customer acquisition.

This is how the LAE ratio trended over time and printed in Q2.

Of course, the lower the better. Currently, it is already one of the best in the industry, if not the best.

Be able to acquire customers profitably for decades

This is crucial, even more so for a D2C business, because if Lemonade were to lose the ability to acquire new customers profitably, it would be a survival issue. Lemonade would likely need to change its go-to-market, starting to rely on human agents, the same as its competitors, which of course would invalidate the thesis.

Just think about what happened to Oddity Tech ($ODD). I recommend every Lemonade investor learn about it, as it is a good case study to understand how important this ability is for Lemonade.

The best indicator of this ability is

Lifetime Value/Customer Acquisition Cost (LTV/CAC)

This value has consistently been held at approximately 3x, which is a healthy level, as reported by management.

I see two positive trends here, both driven by scale.

As Lemonade becomes bigger and operates in more geographies and with more products, its ability to cross-sell improves. This means the same customers will buy more products, which means a higher LTV.

If we look at the denominator, CAC, the same driver helps. More customers, more products and more geographies mean more data and more targets, and therefore a higher probability of finding highly rewarding and efficient campaigns to run. To understand this point, you can read how they deal with marketing campaigns here (from an old report):

Going deeper, the second quarter saw the launch of our 6th generation Lifetime Value machine learning model (”LTV6”). This is our most advanced, credible, nuanced and precise LTV prediction model to date (a synthesis of several models, to be precise).

For each customer we on board, LTV6 offers a prediction of their likelihood to churn, claim, and cross sell - and from these it derives a predicted lifetime value. This is the projected net stream of dollars, discounted for the time value of money, from each customer (i.e., all premiums, minus all claims, adjusted for net present value).

Once they reach statistical significance, each product, each market, and each campaign is likewise scored using LTV6. After adding the customer acquisition cost (CAC) into the formula, the resultant LTV/CAC, together with the CAC Payback Period, determine which product, market, and campaign receives the incremental dollar spend.

This is what Shai said about it during the last earnings call:

We seek to increase LTV through sustained momentum in cross-sells, which can drive gains in retention. And we seek to improve CAC efficiency through more granular AI-driven pricing, which can provide a tailwind to conversion rates. We continue to focus on these key drivers that we believe can drive sustainable profitable growth.

Being able to retain customers

This metric is strictly correlated with the previous one, as it is the other side of the coin. Indeed, every customer retained is worth the same as a customer acquired (or even more as it does not require CAC). In other words, 1 is worth 1: either you gain it or you lose it. And better retention has a positive impact on LTV and vice versa.

Annual Dollar Retention (ADR) is the king metric here.

It held stable sequentially at 85%,

…continuing to reflect the impact of our prior clean the book actions within our homeowners product line. And as a reminder, ADR is measured relative to prior year’s IFP. So, while those portfolio actions are now largely behind us, they will continue to impact the reported ADR metric for the next couple of quarters before rolling out of the comparison period.

Management is basically saying that we will see ADR improving in a couple of quarters.

Where can it go?

Its peak of 88% was reached in Q1 2024, and the drop was due to the “clean the book” actions, based on what we know from management. This makes me think it can go back to that level, but even higher given that Lemonade now has more products to sell (it added Car), which of course is an incentive for customers to remain loyal. So I wouldn’t be surprised to see an ADR of 90% or more in 2027.

ADR is a lagging indicator though, meaning that when you see it changing, it could be too late. As a leading indicator, I instead like to use customers reviews. The rationale is that if customers are satisfied, they will stay. If not, they will change.

As long as Lemonade is rated 4 or above, I feel pretty comfortable with the expected ADR.

Keeping risks under control

This is the essence of an insurer: being able to accurately predict the risk of having to pay a claim and run the business accordingly.

The loss ratio is the king metric here. But as the loss ratio has to be seen as a trade-off between bearing a reasonable risk and keeping prices attractive enough to grow, I like to represent it with my proprietary adjusted rule of 40:

The beauty of this graph speaks for itself.

Given that for Car, the last product introduced, the loss ratio is following a similar path, I consider this part of the equation solved for Lemonade.

Profitability

Although important, of course, being obsessed with profitability is a waste of time and focus. Reasoning from first principles, if Lemonade keeps delivering on the four points that I have outlined, it will eventually become the most profitable insurer in the world.

If you are a long-term investor, you may actually want Lemonade to be as unprofitable as possible for a long time, as long as the cash is spent to grow faster.

Anyway, for the fetishists of profitability, I have at least two pieces of good news for you too.

Lemonade has two further important levers for profitability: reinsurance and Synthetic Agents.

During the quarter, we also completed our annual reinsurance renewal as well as the extension of our synthetic agents program with important upgrades to each. As it relates to reinsurance, the renewed program modestly increases the share of premiums that we retain while meaningfully strengthening catastrophe protection, including named storm coverage that was largely absent under the expiring structure.

Reinsurance is a way to lower risk and use less capital. As Lemonade is already doing, it will likely reduce the reinsured portion as it scales and has more capital available. Or maybe it will prefer (which is what I hope) to reinvest the capital to grow faster and keep the reinsurance share higher. The point is that once it has a meaningful capital surplus every quarter, being more or less profitable by retaining a higher or lower portion of the premium will be just a strategic decision.

Even more compelling is the argument for the Synthetic Agent program.

The agreement related to our synthetic agent extension provides $0.25 billion in growth financing at roughly 9.8% cost and applies to growth spend in ‘27 and ‘28. This amounts to more than 6 percentage points improvement in our cost of capital, materially lowering expected interest expense on a go-forward basis.

Lemonade secured a future reduction of more than six percentage points during Q2, which will be reflected in the financials of 2027.

As above, once enough cash comes in every quarter, Lemonade may decide to quit the program and bring the interest paid on growth to zero.

Conclusion

I’ll say it again: if Lemonade delivers on the four outlined points, it will be the most profitable insurer in the world.

As always, here is the “Deep Dive To Date” (DDTD), which shows how the stock is performing since my initial deep dive on September 24th 2024, when the stock price was $17.23.

2.8x DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

From a RAISE Summit discussion, here is a quote from Nikita Vdovushkin (Product Director, Token Factory at Nebius):

And we do see that like there is like some kind of a maturity curve where people start from using closed-source models like OpenAI and Anthropic or whatever. ... When they start getting traction, the cost basically soar in the the sky and they want to have and they start to get basically just two things. The first one, they do understand their customers better and due to that they have more data on that. And the second thing, they do understand the economics and how the product works better. And they want to apply these two things to make everything like work. And obviously, for that they start to fine-tune, post-train, do something with the model. And yeah, like most of the models that run on top of Token Factory, they still, like, some versions of the open source, but basically they’re somehow customized by the customer.

This makes perfect sense. As a company, you start by testing and may even launch your AI product, or AI-powered product, with closed models: it is faster, convenient, more intelligent and it scales nicely at the beginning. But as you validate your product and make it ready for production, it doesn’t make much sense to keep using closed models. You want it to be as cheap as possible, you want control and, once you have valuable data, you want to fine-tune it. All of these needs are met by open models and a powerful and reliable token factory, which is what Nebius offers and has been investing in from the beginning.

If you have ever worked in a manufacturing company, you may find the following analogy useful. Think about R&D and Production. For the first, you want the smartest and most capable people, as you need to come up with the best possible product. But for production, you want the best trade-off between people who are smart and capable enough and the lowest possible cost. You are willing to spend whatever is needed on individual salaries to create the product, but once the product is created, you need to optimize your manufacturing process to scale in the best possible way, while maintaining decent product quality.

According to research I ran with Grok 4.5, global spending on manufacturing salaries is from 2 to 4 times greater than global spending on R&D salaries. This data doesn’t need to be exact; being directionally correct is enough to prove my point about where the biggest opportunity is.

This doesn’t necessarily mean that global spending on open models will be greater than global spending on closed models. In fact, we have seen the opposite trend over the last 6 months: while tokens on open models are gaining share of total token usage versus closed models, in terms of spending the opposite is happening, and closed models are gaining revenue share (note that this data comes from a Vercel report, and open models are likely underrepresented here because fully self-hosted and direct-provider traffic is missing, but that only strengthens the observed direction: even in this gateway-heavy slice of the market, open models are already rapidly gaining token share while closed models continue to capture the vast majority of spend).

This actually makes perfect sense and supports my point.

More and more tasks will be accomplished by open models. But as you always need to create the best product and do better than the competition in coming up with new products and services, you are incentivized to use the best frontier model for some more strategic tasks. You cannot afford not to use the best intelligence available if you don’t want to be outcompeted.
So companies (and states too—think of departments of defence) will always be willing to pay a premium for specific critical use cases. And as we are still super early in the AI journey, it makes total sense that companies are still increasing their spending on closed models.

Quality of revenue and switching costs

Two main factors make the revenue of open-model providers which will be able to build a moat around software tools (like Nebius) better in quality:

• A company using closed models to experiment, test, build a product, discover stuff or do science is actually quite likely to fail, especially if it is a start-up developing and launching a new product. In that case, its revenue is only temporary for the frontier lab. A company running its workload on Nebius has likely already validated and market-proven its product using closed models, meaning that the chances of that revenue being permanent and growing are much higher.

• Although frontier labs are trying to build stickier and stickier tools to allow their customers to build agents and harness the raw model for a specific use case, if (or I should say when) their frontier model is leapfrogged by a competitor lab, the incentive for their customers to change provider could become compelling, as the strongest value delivered is the intelligence. On the contrary, if you have built on Nebius Token Factory, you have probably invested a lot of effort and money to fine-tune a model for your specific use case, and the incentive to go back to closed models is basically zero.

The arguments in favour of open models, and in particular of a provider like Nebius that is building a moat on what I have just explained above, are compelling and proven at this point.

This is not financial advice, remember that I have a position in Nebius.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

But manufacturing and launching satellites is not the same as operating a global constellation service.

Rocket Lab is buying what cannot be easily built.

Iridium brings:

• scarce, globally coordinated spectrum;

• a constellation operating continuously across the entire planet;

• government, aviation and maritime certifications;

• more than 500 commercial and technology partners;

• billing, provisioning and customer-support infrastructure;

• decades of experience managing a mission-critical satellite network;

• recurring cash flow that can help finance the next constellation and reduce future dilution.

These capabilities would take Rocket Lab many years to reproduce—and some, particularly the spectrum and regulatory rights, might be impossible to replicate.

This is what Rocket Lab is really buying.

In this article, I’ll share my first thoughts and considerations following Rocket Lab’s latest acquisition.

The Business Opportunity for Iridium’s Constellation

These are the numbers for Iridium’s current business:

• Total Revenue (Q1 2026): $219.1 million (+2% YoY)

• Service Revenue (recurring, 72% of total): $158.0 million (+2% YoY)

And these are the lines of business that make up its revenue:

Commercial Service $130.4 million +2% ~60% Largest segment. IoT is the main growth driver (+5%). Voice & Data stable with pricing gains. Government Service $27.6 million +3% ~13% Stable, driven by EMSS contract with U.S. DoD. Equipment Sales $20.2 million -13% ~9% Down due to timing; expected flat for full year.

Engineering & Support Services $40.8 million +9% ~19% Growing, mostly U.S. government work.

Most of the current business comes from legacy areas, such as commercial voice and data and commercial broadband, which will likely shrink over time or grow only moderately.

The most promising area may be commercial IoT data. One might think that, with the explosion of drones and autonomous vehicles more broadly, this business could see strong growth. Every autonomous object will need to be tracked, and Iridium’s PNT (Positioning, Navigation and Timing) service is state of the art in terms of reliability. Thanks to its LEO positioning and Iridium’s unique spectrum, its signal is stronger, more robust and more reliable than GPS, although less accurate. It is normally used together with GPS in missions where PNT complements and protects GPS and other GNSS-reliant systems. In some situations, these systems can be put out of service by bad weather or hostile interference—in war and defense missions, for instance. In those cases, having a more reliable, although less accurate, positioning system becomes crucial.

Digging further into this topic, I discovered that even drones currently used in defense are not generally equipped with Iridium PNT or any equivalent service. When GPS is unavailable, they rely only on vision to navigate. This makes me less certain about the strategic importance and utility of Iridium PNT for most autonomous vehicles, and therefore less bullish about possible future hypergrowth in this line of business. However, the miniature application-specific integrated circuit (ASIC) Iridium is about to launch will allow partners to integrate its PNT services more cost-effectively, using less power and space.

That could make the service more compelling for autonomous vehicles.

The most promising business appears to be commercial IoT. Once Iridium launches its Direct-to-Device IoT connectivity later this year, every phone could potentially connect to Iridium satellites and use their services without needing additional hardware. It will also become easier and cheaper to integrate Iridium services into any IoT object, simply by using a small commercial modem. This will certainly boost IoT segment revenue. Until now, Iridium IoT required a dedicated receiver, making it complex and expensive—and, in some cases, simply not feasible.

Yet I am not convinced that this will be a game changer capable of turning Iridium into a hypergrowth story. For mobile phones, Iridium is a good solution for providing connectivity globally and in almost any weather condition. In that respect, it is more reliable than Starlink, which can struggle in bad weather despite offering broadband internet speeds, while Iridium can provide only the lower speeds needed for messages and phone calls. The question is: how large is this market? Even assuming that every phone maker in the world adopts it across its devices, the ARPU would probably remain tiny, although the service itself would still be compelling.

My conclusion—possibly wrong, since I have only just started analysing the topic and there is still a lot I do not know—is that current’s Iridium constellation is a stable, recurring-revenue business heavily weighted toward commercial IoT and government contracts. Growth is modest, in the low single digits, but reliable, with upside from new services such as D2D, PNT and aviation. It is not a high-growth company yet, but the Rocket Lab acquisition could accelerate its future expansion.

The biggest opportunity is still defense

If we look at who is launching things into space as a proxy of the space TAM share, there are four meaningful demand categories:

1. Communications constellations — largest volume, but overwhelmingly Starlink.

2. Defense and sovereign infrastructure — broadest and most dependable launch market.

3. Earth-observation intelligence — smaller satellites, recurring replenishment and growing defense demand.

4. Science, exploration and human spaceflight — valuable but government-funded and relatively slow-growing.

If we exclude Starlink, the most meaningful missions are funded directly or indirectly by governments.

For me, defense remains for Rocket Lab the clearest space hypergrowth opportunity for at least the next five years.

The US and its allies are moving from a few expensive satellites toward proliferated constellations for missile tracking, secure communications, intelligence, resilient navigation and space-domain awareness.

Rocket Lab was already moving in this direction. Its Space Development Agency contracts proved that it can compete for awards historically reserved for legacy aerospace primes.

Iridium reinforces the thesis.

The combined company can potentially offer a government the complete architecture:

Design the satellites. Build the payloads. Launch the constellation. Operate the network. Deliver the service.

That is a much stronger proposition than being only a launch provider or component supplier. Rocket Lab can take responsibility for the entire mission—and capture more of its economics.

Iridium also gives Rocket Lab something especially valuable in defense: operational credibility. It has already demonstrated that it can run a secure, resilient global constellation relied upon by governments and safety-critical industries.

Becoming a Platform: The AWS of Space

A customer planning a new constellation does not only need spacecraft. It needs network architecture, spectrum strategy, ground infrastructure, fleet management, continuous operations and distribution.

After Iridium, Rocket Lab can credibly sell a much broader outcome:

Rocket Lab will design, manufacture, launch and operate your constellation.

That positions Rocket Lab to compete as a prime contractor for commercial deals, while creating more opportunities to sell its own solar arrays, star trackers, reaction wheels, flight software, payloads and launch services.

Once a space company owns the complete stack, I would argue that it becomes a platform.

If you need to send assets into space and operate them there—especially as part of a constellation—you will probably standardize your payloads around Rocket Lab or SpaceX. Once you know how to design and build them for, say, Rocket Lab, you can standardize dimensions, processes, communication protocols and operations around the same provider.

This is what happened with cloud providers such as AWS. Once it became clear that running software in the cloud was cheaper, more reliable and more convenient than hosting it on-premise, software vendors and end-users began standardizing their operations around AWS, Azure or Google Cloud. In space, this dynamic could be even stronger because there is no real “on-premise” option. The choice is between rebuilding the entire stack from scratch every time or standardizing around one of the only two or three end-to-end space companies.

With this acquisition, Rocket Lab has everything required to become a platform—the same concept I described in my original deep dive. I think we are getting closer to seeing it become a reality.

The optionality is real—but not yet the thesis

Iridium mentioned several possible new services in its 2025 annual report, including aviation data through Aireon and Project Authentic, which aims to use Iridium’s PNT signal to authenticate the physical location of devices involved in sensitive transactions.

Project Authentic is interesting because it could move Iridium from backup navigation into cybersecurity. But it is still early: no named platform partners or commercial contracts have been disclosed.

I would treat these services as optionality, not as justification for paying $8 billion today.

Rocket Lab next constellation

As a simple way to see it, most launches today and in the future will be about one of three things:

1. Connecting things (internet, phones, IoT, drones)

2. Watching things (Earth, weather, military targets)

3. Enabling future ambitions (Moon bases, Mars, space industry)

SpaceX is capturing most of the connectivity business through Starlink, while IoT remains a smaller and slower-growing segment.

When it comes to future ambitions, I think Rocket Lab will play an important role in orbital data centers, even though this opportunity may not become viable or meaningful for at least another five years.

Looking at the short term—say, three to five years—I believe Rocket Lab could launch its own Earth-observation constellation. That seems achievable.

Over a ten-year period, Rocket Lab should undoubtedly aim to operate its own orbital data-center service.

What could Rocket Lab buy next?

Rocket Lab has already acquired Mynaric, adding laser terminals for high-bandwidth links between satellites.

The next logical gap is the other side of that network: space-to-Earth optical communications and optical ground infrastructure.

A future acquisition could add optical ground stations, atmospheric and cloud mitigation, network-routing software and hybrid laser/RF connectivity.

That would matter for defense networks, Earth observation and, eventually, space-based data centers. Rocket Lab would control the satellites, power systems, optical links, launch and ground connection.

Strategically, this is the missing capability I would watch.

My conclusion

Iridium’s existing constellation will not support Rocket Lab hypergrow by itself. I would be very surprised to see its business growing by more than 15-20% YoY over the next five years. Its existing business is profitable and defensible, but mostly mature. It will certainly strengthen Rocket Lab’s financial position by adding a meaningful, cash-generating revenue stream that could even make the company profitable as early as 2027, when Neutron CAPEX begins to slow down.

The acquisition works if Rocket Lab uses Iridium to become something much larger: an end-to-end prime contractor capable of building and operating complete defense and commercial space infrastructure.

In terms of a proprietary Rocket Lab constellation, beyond what Iridium already has in operation, I struggle to see anything generating meaningful revenue before five years from now.

That is the opportunity.

Rocket Lab is not simply buying a satellite operator. It is buying the final operating layer it needs to compete for the largest contracts in space.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

I have a position in Rocket Lab, and please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

In terms of timeline, I starterd in 2024 and I believe the $1 million target is achievable by 2032, assuming no major black swan events or asset bubbles burst. In such scenarios, it would naturally take longer.

Target and KPIs

Every meaningful goal requires a clear path and process, which is why the portfolio’s success is evaluated on a rolling five-year average basis using two key performance indicators (KPIs):

1. Compounding at a minimum of 30% year-over-year (YoY)

2. Outperforming the NASDAQ Composite benchmark by at least 10%

A consistent 30% YoY compounding rate won’t quite deliver $1 million by 2032, but it represents a realistic yet ambitious target—one that would likely surpass the performance of most institutional and retail investors alike.

L2C Portfolio Performance

Now, turning to the H1 2026 (First half of 2026) performance.

Following a stellar 2024, and a good 2025, here’s how H1 2026 stacked up:

• Lorenzo2cents Portfolio YTD: +19.0%

• NASDAQ Composite: +11.5%

• S&P 500: +9.1%

While the YTD performance remains strong compared to the benchmark and is well on track to meet my goals, I am a bit disappointed about the significant drawdowns experienced in late February and late March. During that period, my portfolio declined by an average of 25%, while the Nasdaq fell only 2% to 10%.

The underperformance was primarily driven by idiosyncratic events affecting two holdings: Oddity and Hims. Oddity suffered a severe setback in customer acquisition (which has not yet been resolved), causing its stock to drop ~70% since the start of the year. Hims was hit hard too, falling 55% after the Novo lawsuit.

Events like these are an inherent part of the portfolio’s risk profile. That said, the probability of two out of nine stocks experiencing major negative market-moving event at the same time is quite low — even though it unfortunately happened in this case.

Let’s now review the success KPIs since inception (January 1, 2024) through July 2, 2026:

1. L2C Portfolio CAGR: 58% — exceeding the minimum goal of 30%

2. L2C vs. NASDAQ Composite (IXIC) CAGR: 58% vs. 24,2% — surpassing the minimum goal of outperforming the market by 10% (overperformed by 33,8%)

L2C Portfolio Composition (as of July 2, 2026)

The following positions constitute the portfolio:

1. $LMND: 25.6% Allocation - Reinventing the insurance sector.

2. $NBIS: 22.5% Allocation - The Application Software’s Universal Substrate for Agentic Computation

3. $RKLB: 17.7% Allocation - The leading public space company.

4. $HIMS: 14.9% Allocation – Redefining healthcare by making it accessible and affordable.

5. $TSLA: 6.8% Allocation - The ultimate leader in transportation autonomy, sustainable energy, and humanoid robotics.

6. $MELI: 5.9% Allocation - The backbone of Latin America’s digital economy, poised to benefit from continent-level network effect.

7. $ZETA: 4,3% Allocation - The Marketing Vertical Ontology

8. $ODD: 3,2% Allocation - Revolutionizing the beauty industry.

9. $DUOL: 1.9% Allocation - Transforming education by making learning both fun and effective.

10. Liquidity: (3.6%)

Key Portfolio Moves

• Sold $META - While I still believe META is unlocking AI’s future with an unparalleled dataset, I decided to allocate the capital to more asymmetric bets.

• Opened a $ZETA position - Incredibly undervalued with an exceptional management team.

Top Performers and Top Losers (+-20% movements)

• $NBIS +157%

• $RKLB +44%

• $ODD -55%

• $DUOL -28%

Stay tuned

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my

YouTube Channel

I share my thoughts on

X

To read my deep analysis

Subscribe now

Conclusion

I’m quite pleased with H1 2026’s performance and anticipate an even more exhilarating second half of 2026, if the market doesn’t crush. The Lorenzo2Cents project is growing steadily and is poised to deliver even greater value to those willing to engage with it. Stay tuned!

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Revenue grew 49% year over year to $8.8 billion, the fastest pace in almost four years. GMV grew 42% to $19.0 billion. TPV grew 50% to $87.2 billion. Fintech monthly active users reached 82,9 million, up 29% year over year. The credit portfolio reached $14.6 billion, up 87% year over year.

And yet the stock sold off because operating margin fell to 6.9%, down 600 basis points year over year.

But the headline is not that MercadoLibre is suddenly less profitable. The headline is that MercadoLibre is choosing to invest aggressively because the opportunity in front of it is still enormous, and those investments are already producing faster growth, stronger network effects, and better scale economics.

Here is how Ariel Szarfsztejn, CEO of Mercado Libre, described competition in Brazil.

Brazil is one of the most attractive e-commerce markets in the world. So it’s natural that it’s getting more and more intense and as that has been the case for many years... competitive intensity is also having a positive impact in the market as a whole by bringing new consumers from the offline world into the online world, and we feel we are very much equipped to offer all those consumers the opportunity to buy in MercadoLibre. So the pie is increasing at a faster pace than it was before, and we are taking an even larger slice of that pie.

This sounds almost like a Jevons paradox.

In the original Jevons paradox, when something becomes more efficient, people do not necessarily consume less of it. They often consume more because the lower cost expands the total market.

Something similar is happening in Latin American e-commerce.

Competition, free shipping, better logistics, lower friction, better search, and better prices are not simply taking share from MercadoLibre. They are accelerating the migration from offline retail to online retail. And when more consumers move online, MercadoLibre is structurally advantaged because it has the largest marketplace, the strongest logistics network, the most trusted payments layer, and the best data loop in the region.

In other words, competition is not necessarily a zero-sum game yet.

Latin America is still too early.

Management reminded investors that the average American makes 41 online purchases per year, while the average Latin American makes just 7. MercadoLibre buyers average 11. That gap is the whole thesis.

If the market grows faster because consumers are being pulled online more aggressively, the largest platform can benefit even if the environment looks more competitive on the surface.

That is exactly what Q1 showed.

Brazil: The Flywheel Is Working

Brazil was the standout market again.

MercadoLibre lowered the free shipping threshold in Brazil in 2025, and the market initially interpreted it as a margin-negative competitive response. But the data keeps pointing in the opposite direction: it is a long-term investment that is expanding the market, increasing frequency, and improving the utilization of MercadoLibre’s infrastructure.

Management put it clearly:

By bringing more buyers into the ecosystem, we’re strengthening network effects with higher purchase frequency, broader assortment and a logistics network that becomes more efficient with every incremental package. As a result, Brazil delivered another standout quarter for commerce. GMV grew 38% year-over-year as items sold growth accelerated to 56%. This is more than double the quarterly growth rate prior to lowering the free shipping threshold.

Free shipping penetration reached a new record and unit economics continue to improve with cost per shipment down 17% year-over-year in local currency. In other words, higher demand is driving lower costs.

This is the core point.

More demand is not just more revenue. More demand means more packages through the same network. More packages improve route density, fulfillment utilization, line-haul efficiency, and last-mile economics. MercadoLibre already has the most advanced logistics infrastructure in the region, so incremental volume should improve the economics of the network over time.

This is why I think the market is misreading the margin compression.

If the lower free shipping threshold were simply a price cut with no strategic benefit, I would be worried. But if it brings more buyers into the ecosystem, increases frequency, expands category breadth, improves logistics density, and strengthens the marketplace, then the short-term margin hit is an investment.

And the investment is clearly working.

Brazil GMV grew 38% FX-neutral. Items sold grew 56%. Unique buyer growth accelerated to 32%, the fastest pace in five years. Unit shipping costs in Brazil fell 17% year over year in local currency.

The market sees lower margins. I see MercadoLibre pulling more of Latin America into its ecosystem.

The impact of AI Is Still Underappreciated

Another part of the quarter that I think is still underappreciated is AI.

Most investors do not think of MercadoLibre as an AI beneficiary. They think about marketplace, logistics, payments, credit, and maybe ads. But AI directly improves the economics of all of these layers.

Management said:

We deployed LLMs in search in commerce for the first time this quarter. And basically, that is live in Brazil, Mexico and Argentina. So now we are using this technology to better understand users’ intent, combining both knowledge on the user behind the query and better interpretation of the query itself. And the impact is basically very visible across the funnel. We have higher conversions as buyers find what they are looking for much faster.

We have better ad returns as our search also improves the quality of the results that our ad tech stack is generating. We have higher -- stronger engagement from our users as the discovery experience actually improves. And clearly, this is one of the contributors to the great performance we had this quarter.

This matters because MercadoLibre has one of the richest first-party datasets in Latin America: shopping intent, payment behavior, merchant data, credit performance, logistics data, search data, and advertising performance.

LLMs improve the search layer, but the impact does not stop there.

Better search means higher conversion. Higher conversion means more GMV. More GMV means more payment volume. More payment volume means more data. More data improves underwriting, ads targeting, and personalization. Better personalization increases engagement. Higher engagement increases frequency.

That is the ecosystem loop.

The low-hanging fruit is advertising. If AI improves search relevance and sponsored listing performance, ad ROI improves. When ad ROI improves, sellers can spend more. MercadoLibre said ads revenue grew 73% in USD and 63% FX-neutral in Q1.

MELI does not disclose Mercado Ads revenue as a separate line item, but based on prior disclosed ads penetration and Q1’26 growth, Mercado Ads was likely running at roughly $2B annualized run-rate.

This is still a small line relative to the full business, but it is a structurally high-margin opportunity. It also reinforces the marketplace, because better ads help sellers sell more and help buyers discover more relevant products.

But I think the bigger AI opportunity is not only ads. It is the improvement of the entire consumer experience. Search becomes more natural. Discovery becomes better. Credit decisions become more precise. Customer support becomes cheaper. Logistics workflows become more productive. Seller tools become more useful.

This is not an AI story in the hype sense. It is an AI story because MercadoLibre has scale, proprietary data, distribution, and real use cases.

The Margin Compression Is Not What It Looks Like

The main reason the stock is trading around 2.5x LTM revenue is that investors are worried about competition and margin pressure.

I understand the concern, but I think the conclusion is wrong.

MercadoLibre is not losing control of its business. It is deliberately choosing to invest because the returns on those investments are attractive.

Martin de los Santos, CFO at Mercado Libre, said it directly:

If we wanted to improve margins in the short term, it will be fairly easy for us to slow down certain investments, but we don’t think it’s the right way to go given the large opportunity that we have in front of us for both commerce and fintech.

This is not a management team trying to explain away a broken model. This is a management team saying: we can harvest, but harvesting would be the wrong capital allocation decision.

The biggest distortion is in Mercado Pago.

The credit book grew 87% year over year, much faster than consolidated revenue growth of 49%. That creates accounting margin pressure because MercadoLibre has to provision expected losses upfront when it originates new credit.

Management explained the mechanics:

The fact that our credit book grows at a faster pace than revenues, the credit book grows at 87% year-on-year and our revenues for MercadoLibre growth at 49%, that generates margin compression.

And the reason for that is because as we issue any new loan, we have to provision for the full amount of the expected loss of that loan. And when we accelerate growth, we need to provision more.

So 2/3 of the margin compression comes from that.

This is crucial.

If the credit book is growing profitably, then near-term provisions are not the same thing as structural margin deterioration. They are the accounting cost of growth.

MercadoLibre’s Q1 NIMAL was 17.8%, down from 22.7% a year ago, mainly because of the higher mix of lower-spread credit cards and some spread compression in Brazil. That decline is worth watching. But 17.8% NIMAL is still an extremely attractive return profile, especially if the credit book is expanding and asset quality remains controlled.

The 15-90 day NPL ratio was 8.0%, broadly stable year over year, according to the company. Management also said that each of its consumer and merchant loan portfolios remains highly profitable.

So I do not think the right conclusion is “Mercado Pago is destroying margins.”

The right conclusion is: Mercado Pago is growing so fast that the accounting cost of new credit is temporarily compressing margins, while the future earnings power of the credit book is being built.

FCF: Do Not Treat Loan Growth Like a Cost

This is also why I think the way we calculate free cash flow matters.

If we subtract all cash used to grow the loan book as if it were an operating cost, as per standard account practice and as often done by investors, we risk dramatically understating the value of MercadoLibre.

That cash is not being burned. It is being converted into financial assets that generate yield.

Using the standard free cash flow definition:

Free Cash Flow = Cash From Operations - Capex

MercadoLibre generated roughly $11.8 billion of LTM free cash flow through Q1 2026 (LTM):

That is the number I care about for the main valuation framework.

Yes, the credit book requires capital. Yes, provisions and funding costs matter. Yes, NPLs must be watched carefully. But deducting loan originations 1:1 from FCF can make the business look much less valuable than it is, because it treats a productive asset like a cost.

The right way to think about it, in my opinion, is to keep reported FCF as the main line and analyze credit-book growth separately: NIMAL, NPLs, funding costs, portfolio mix, and return on the capital deployed.

If those metrics deteriorate, then the thesis changes. But as of Q1, I do not see evidence that MercadoLibre is simply buying growth with bad credit.

I see a company investing aggressively into a credit opportunity where it has unique data, distribution, and ecosystem advantages.

Mercado Pago vs Nubank: The Business Ontology Test

One of the most important data points I am watching for my Business Ontology is the performance of Mercado Pago versus Nubank.

Nubank is probably the cleanest benchmark for Latin American fintech at scale. It is an incredible business. It has more than 135 million customers, a monthly activity rate of 83%, and a credit portfolio of $37.2 billion, up 40% year over year FXN.

But Mercado Pago is becoming harder to ignore.

Mercado Pago reached 82,9 million monthly active users, up 29% year over year.

Its credit portfolio reached $14.6 billion, up 87% year over year. AUM reached almost $20 billion, up 77% year over year.

The active-user comparison is also interesting. Nubank had more than 135 million customers in Q1 2026 and an 83% monthly activity rate, which implies roughly 112 million active customers. That is still larger than Mercado Pago, but Nubank’s active base grew from 98.7 million monthly active users in Q1 2025 to roughly 112 million in Q1 2026, or about 14%. Mercado Pago’s active user base grew 29% over the same period.

The profitability of credit also tells a more nuanced story. Nubank’s risk-adjusted NIM was 9.5% in Q1 2026, down from 10.5% in Q4 2025. Mercado Pago’s NIMAL was 17.8%, down from 22.7% a year ago. These are not perfectly identical metrics, so I would not overstate the comparison. But directionally, Mercado Pago is scaling credit faster, with a higher reported credit spread, while Nubank remains the larger and more mature fintech machine.

So Nubank is still much larger in credit portfolio and active customer scale. But Mercado Pago is growing active users and credit faster, and it is doing it from inside the MercadoLibre ecosystem.

This is exactly my thesis.

The combination of commerce + fintech should create advantages that a standalone digital bank does not fully replicate.

MercadoLibre knows what consumers buy, what merchants sell, how often they transact, what categories they participate in, how payments flow, how logistics behave, and how users interact across the platform. This gives Mercado Pago a different underwriting and distribution advantage.

A standalone fintech can have an excellent app and an excellent brand. Mercado Pago has that too, but it also has the marketplace.

MercadoLibre is not just a marketplace with a payment app. It is a commerce and financial infrastructure layer for Latin America.

What Could Re-rate the Stock?

I do not think a re-rate is imminent. The market is clearly skeptical, and investors usually do not reward margin compression until they can see stabilization.

But a few catalysts could change the narrative.

First, a stock split. Economically, it changes nothing. But MELI’s share price is high, and a split could renew retail investor interest, especially if the fundamental story is already improving.

Second, NIMAL stabilization. The decline from 22.7% to 17.8% is one of the concerns. If NIMAL stabilizes while the credit book keeps growing, the market may start seeing the credit investment as value creation rather than margin risk.

Third, margin stabilization. Management does not need to maximize margins today, but investors need evidence that margins are being compressed by choice, not by necessity.

Fourth, continued FCF growth. If free cash flow per share keeps rising while the company reinvests aggressively, the market will eventually have to reconcile the narrative of margin pressure with the reality of cash generation.

Fifth, capital rotation from AI stocks. MercadoLibre is not an obvious AI trade, but it is one of the few large-cap businesses where AI can improve an already dominant marketplace/fintech ecosystem. If investors start looking for non-obvious AI beneficiaries with real earnings and real data advantages, MELI fits that bucket.

L2C take aways and performance

MercadoLibre is being priced like a company whose margins are under attack.

I think it is a company choosing to reinvest at high rates because the opportunity is still underpenetrated, and because its ecosystem advantages are getting stronger.

Q1 2026 was not a clean quarter if you only care about near-term EPS. Operating income declined, net income declined, and margins compressed.

But if you care about long-term value creation, the quarter was extremely strong.

Revenue growth accelerated. Brazil commerce accelerated. Items sold exploded. Logistics unit costs improved. Ads grew rapidly. Fintech MAUs grew. AUM grew. The credit portfolio expanded. AI started to show real business impact. Free cash flow remains very strong on an LTM basis.

The market sees lower margins.

I see a dominant platform with a huge volume of high quality first-party data investing to make the market bigger, take a larger slice of it, and turn that scale into long-term cash flow.

At roughly 2.5x LTM revenue, the asymmetry is becoming more and more interesting.

I do not know when the market will re-rate MELI. It may take time. But even without multiple expansion, a business compounding revenue and cash flow at these rates can still create significant value.

And if the market eventually understands that the margin compression is mostly a reinvestment cycle rather than structural damage, the upside could be much larger.

I am willing to wait.

I will soon publish a video where I explain my updated Rocket Lab 4D valuation model and the assumptions behind it.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

As always, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since my initial deep dive on the May 18th 2025, when the stock price was $2585. As proved by my post on X, where I share real time updates, I opened a position a few days later.

- 37% DDTD

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Demand is clearly not the problem.

Execution is not the problem.

The company is doing almost everything bulls wanted to see: revenue is growing, backlog is expanding, Electron is still proving its cadence, HASTE is becoming increasingly relevant for defense, and Neutron is attracting customers before its first launch.

On the business side, this was probably one of the strongest updates Rocket Lab has ever delivered.

But that is not the only question that matters for investors.

The real question is different:

How much of the future is already priced into the stock?

And after updating my model, my answer is less comfortable than it used to be.

Rocket Lab remains an exceptional company. I still think it is one of the most strategically important public companies in the space economy. But at the current valuation, the stock no longer looks like the obvious asymmetric setup it was before.

The company may execute beautifully.

The stock may still do well.

But the margin of safety is much thinner.

Q1 was a very strong quarter

Rocket Lab reported record Q1 2026 revenue of $200.3 million, up 63.5% year over year. GAAP gross margin reached a record 38.2%, and non-GAAP gross margin was 43.0%.

Backlog reached $2.2 billion, up 20.2% sequentially and more than double year over year.

That backlog number matters more than a normal quarterly revenue print, because Rocket Lab is not a simple transactional business. It is becoming a long-duration execution machine across launch, spacecraft, components, defense programs, and eventually Neutron.

Backlog is not revenue.

But in space and defense, backlog is the closest thing we get to forward visibility.

And the direction is very clear.

The backlog-to-revenue ratio increased sequentially again. Using Q1 revenue of $200.3 million and backlog of $2.2 billion, Rocket Lab is sitting at roughly 11x quarterly revenue in backlog.

That means demand is not just growing in absolute terms; the forward pipeline is growing faster than current recognized revenue.

That is exactly what you want to see from a company that is supposed to accelerate over time.

Peter Beck summarized the demand picture very well:

With the 31 Electron and HASTE launches and 5 Neutron contracts combined, we booked more launches in the first 3 months of 2026 than we did for all of last year.

That sentence is the bull case in one line.

Rocket Lab is not begging the market for demand.

The market is asking Rocket Lab for capacity.

Launch demand is inflecting

The most important signal in Q1 was not just revenue.

It was bookings.

Rocket Lab signed 31 new Electron and HASTE contracts in Q1, plus five dedicated Neutron launches.

This matters for two reasons.

First, Electron is still alive and very relevant.

Many people underestimate Electron because it is a small rocket, but that misses the point. Electron has cadence, reliability, and customer trust. In launch, those things matter more than PowerPoint capacity.

Second, Neutron is getting commercial validation before first flight.

That is important.

Neutron is still the biggest execution risk in the Rocket Lab story. Until it flies successfully, it is still a promise. But customers are already reserving capacity through the end of the decade, which tells you there is real market appetite for another reliable medium-lift provider.

The industry does not want only SpaceX.

Governments do not want only SpaceX.

Commercial constellation operators do not want only SpaceX.

The world needs more launch capacity, and it needs it from operators that can actually deliver.

That is where Rocket Lab has a shot.

Neutron still looks on track

The other key update is that Neutron is still expected to launch later this year.

No new delay was signaled in the Q1 update.

This is important because a large part of the Rocket Lab thesis now depends on Neutron. Electron is a great business, but it is not enough to justify the full long-term vision by itself. Space Systems is growing and becoming strategically important, but Neutron is what can turn Rocket Lab from a strong niche player into a true end-to-end space infrastructure company.

Neutron changes the size of the opportunity.

It opens the medium-lift market.

It gives Rocket Lab more control over its own space systems deployment.

It creates the platform that could eventually enable the services layer Peter Beck has talked about for years.

And that last point is key.

Because if Rocket Lab ever becomes more than launch + components + spacecraft manufacturing, if it eventually owns and operates services from space, the upside can be much larger than what a normal aerospace model captures.

But timing matters.

Peter Beck has been clear that services from space come after Neutron is established. That makes sense. You need the transportation layer before you can efficiently deploy and replenish your own infrastructure.

So even in an optimistic scenario, meaningful revenue from a services business is probably still three to five years away.

That does not mean it has no value.

The valuation is rich.

This is the uncomfortable part.

When I update my model, the conclusion is not that Rocket Lab is overhyped as a company.

The conclusion is that the stock price is already discounting a huge amount of success.

At the current valuation, the market is not pricing Rocket Lab like a speculative small launch company anymore. It is already pricing in a version of the future where:

• Electron continues to grow and maintain strong cadence.

• Space Systems compounds for many years.

• Gross margins keep expanding.

• Neutron launches successfully.

• Neutron scales without major delays.

• Rocket Lab captures a meaningful share of medium-lift demand.

• Defense and national security demand remain strong.

• The company keeps using its balance sheet well for M&A and vertical integration.

That is a lot of good news already embedded in the stock.

To be clear, none of these assumptions are crazy.

Actually, many of them are exactly what I want Rocket Lab to do.

The issue is not that the story is fake.

The issue is that the market now believes a large part of the story.

And when the market already believes, the asymmetry changes.

A year ago, the market was still treating Rocket Lab like a risky space stock with a small rocket and a dream.

Today, the market is closer to pricing it like a future space prime with a successful medium-lift rocket, a growing defense business, expanding margins, and long-duration backlog.

That is a very different setup.

What is not priced in?

In my view, the part that is not fully priced in — or at least not priced in enough — is the long-term services layer.

This is the part of the thesis where Rocket Lab stops being only a provider of launch and space systems, and starts monetizing infrastructure in orbit.

That could be constellations.

It could be data.

It could be communications.

It could be defense-related services.

It could be something we cannot precisely model yet.

This is why I have always liked Rocket Lab. Peter Beck is not building a small rocket company. He is building an end-to-end space company.

Launch is the access layer.

Space Systems is the manufacturing layer.

Services would be the monetization layer.

If Rocket Lab gets there, the upside can still be very significant.

But this is also the most uncertain part of the thesis.

It is far away.

It depends on Neutron.

It depends on execution.

L2C take aways and performance

Rocket Lab is still a great company. But the stock is no longer an obvious asymmetric bet at this valuation.

I remain invested.

But I am watching closely, and I am ready to take action if the risk/reward keeps deteriorating versus other opportunities in the portfolio.

I will soon publish a video where I explain my updated Rocket Lab 4D valuation model and the assumptions behind it.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

As always, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since my initial deep dive. For RocketLab stock ($RKLB), the price on September 17th, 2024, when I published my analysis, was $7.19. As of this update, the price stands at $102, reflecting a

+14x DDTD

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

I won’t repeat the full thesis here. If you haven’t read it, start there.

This is just a short update after Q1.

Sales declined 26% year over year, slightly better than management’s previous expectation of about -30%, but directionally nothing changed. The damage is real.

Management also said they “remain hopeful” that the business will return to normalization in the second half. That word matters. “Hopeful” tells me this is still not under their control.

There was one encouraging data point: IL Makiage CPA improved in May, declining an estimated 28% from April after multiple months of increases with the same advertising partner.

That could be the first sign of stabilization. But it also shows how serious the problem was. CPA kept getting worse for months, even after management knew what was happening and started taking corrective action.

May could be the turning point. Or not.

The other important update is that Oddity has already shifted 40% of acquisition revenue out of Try Before You Buy and into the standard Buy model.

That is a big deal. TBYB was part of the IL Makiage edge. I don’t know exactly how much of the historical advantage came from it, but it clearly mattered to some degree. If they can move away from it with “no impact on unit economics”, great. But I want to see more proof.

The most concerning part of the call was how much power Meta has over Oddity, and how little accountability Meta has when something breaks.

Oddity was damaged by a Meta algorithm change. Meta is not accountable for that damage. And now Oddity still has to keep spending money on Meta to understand the problem, test fixes, and try to recover.

Without spending, they cannot identify what works. But because the efficiency of that spend is broken, they also cannot simply scale it back up.

That is a very uncomfortable position.

Management also made clear that there is no real alternative at the same scale. Yes, Oddity advertises on other platforms, but their largest ad partner is by far the biggest acquisition channel in U.S. beauty, with more than 50% market share.

In practice, Oddity has no real alternative to Meta for new customer acquisition. They are still completely dependent on it.

This is the core lesson for me.

Oddity built a very powerful D2C engine, but it was more fragile than I appreciated. The company is not broken, and the thesis is not dead, but the acquisition model depends too much on one platform.

Meta changed the rules. Oddity took the damage. Now Oddity has to spend money to figure out how to adapt.

That is not a position of strength.

IlMakiage.com traffic also continued to fall, which confirms why I wanted to monitor it in the first place. It is not perfect, but it looks like a decent proxy for the state of the recovery.

For now, I am leaving my position untouched.

I will keep monitoring web traffic, CPA commentary, and any sign that the business is normalizing. If I make any trade, I will update Business Ontology members first.

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

As always, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since my initial deep dive on the July 16th 2024, when the price was $44.24.

-77%% DDTD

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

I think that call was right.

Q1 revenue grew only 4% year over year, gross margin compressed, and GAAP net income flipped to a loss. The quarter that Hims just reported is backward-looking.

The more important information came from the guidance, the Novo Nordisk pivot, the Eucalyptus deal, and especially what management said about AI.

And I think very few people have understood where Hims is going.

Management guided Q2 revenue to $680–700 million and raised full-year 2026 revenue guidance to $2.8–3.0 billion.

Yemi Okupe, Hims’ CFO, basically told investors that demand in weight loss is exploding again:

Within weeks of this launch, we are on track to add north of 100,000 new subscribers per month within our weight loss specialty. Early signs point toward a high level of subscriber engagement with nearly 90% of these users downloading the app and the average subscriber of these products interacting with the provider 3 times in the first month.

In Q4 2025, subscriber growth was almost irrelevant. In Q1 2026, Hims added roughly 100k subscribers in the quarter, or around 33k per month average.

Now management is talking about north of 100k new weight-loss subscribers per month.

That is roughly 3x the Q1 pace.

And this is happening after the strategic shift toward a broader assortment of FDA-approved GLP-1s through Novo Nordisk.

Hims announced in March that Novo Nordisk’s FDA-approved GLP-1 medications were available through the platform, including Wegovy options and Ozempic for eligible patients where clinically appropriate.

This matters because the market narrative on Hims became too simplistic.

The bear case was: “The compounded GLP-1 opportunity was temporary, so the business is broken.”

My view is different.

The compounded GLP-1 opportunity was the ignition. The real asset is the consumer healthcare distribution layer.

Hims can take demand, route it through a regulated medical workflow, connect it to providers, manage the customer relationship, and increasingly personalize the journey.

That is not easy to replicate.

Andrew Dudum also said demand is now stronger than what they saw during major seasonal campaigns:

We see it in the scale of new customers coming in the door today, which is larger than, as Yemi said, any spikes we saw even during the most important seasonal campaigns such as New Year’s and Super Bowl campaigns.

That is why I do not think Q1 is the quarter to obsess over.

The important signal is what happens from Q2 onward.

Hims is sandbagging

Management guided Q2 revenue to $680–700 million.

At the midpoint, that is around 13.5% sequential growth from Q1.

At the high end, it is around 15% sequential growth.

That already tells you the business is reaccelerating.

Now add the second part of Yemi’s comment:

...with record-level users coming to the platform, the focus really is on continuing to expand share, expand our reach and expand the overall subscriber base. And so with that, I think we would expect to see accelerating growth via revenue over the back half of the year.

For a subscription business, I read “accelerating growth” primarily as sequential acceleration.

If Hims does $700 million in Q2 and hits the high end of full-year guidance at $3.0 billion, then Q3 and Q4 together need to contribute about $1.692 billion.

One simple version could be:

• Q1: $608 million actual

• Q2: $700 million high-end guidance

• Q3: $800 million

• Q4: $892 million

• FY 2026: $3.0 billion

The problem is that this would not really show clean acceleration into Q4. It would imply Q3 growing about 14% sequentially and Q4 growing about 11% sequentially.

So if management is serious that the back half should accelerate, and if monthly subscriber additions really are running north of 100k in weight loss alone, then the full-year guide looks conservative to me.

This is not a precise model. I am not pretending to know Q3 and Q4.

But the direction is clear.

Hims is guiding to a year where growth reaccelerates, while also telling us that the biggest category is seeing record-level demand. That combination does not sound like a business in structural decline.

It sounds like a business sandbagging a little.

And this excludes Eucalyptus.

Hims valuation in 2027

The Eucalyptus acquisition is also important.

Hims announced the deal in February. Eucalyptus is a consumer healthcare platform with operations in Australia, the UK, Germany, Japan, and Canada. Hims disclosed that Eucalyptus had an annual revenue run-rate north of $450 million and delivered triple-digit year-over-year ARR growth in every quarter of 2025, while operating within line of sight of profitability.

The transaction is expected to close around the middle of 2026, subject to approvals, and Hims explicitly excluded any Eucalyptus contribution from its Q2 and full-year 2026 guidance.

This means the reported guidance does not fully capture the new international scale.

If Hims exits 2026 near a $3 billion revenue run-rate and then layers in Eucalyptus, which was already north of $450 million ARR and growing very fast, the combined business could be much larger than what the current headline guidance implies, probably around $4 billion revenue run-rate in 2027.

At around a $6 billion market cap, it sounds a great deal to me.

No financial advice, of course. Do your own work.

Hims is an AI company and providers are training the model

Now we get to the part that I think almost everyone is missing.

Hims is an AI company.

I know this sounds like a cliché because every company now says it is an AI company.

But with Hims, the claim is not marketing.

It is structural.

In my first Hims deep dive in February 2025, I framed one of the key points as: a critical factor in leveraging the quantum leap in AI technology.

The idea was simple.

Healthcare AI is only as good as the feedback loop behind it.

You need consumer intent. You need symptoms. You need diagnosis. You need provider decisions. You need treatment plans. You need side effects. You need outcomes. You need longitudinal data. And you need all of that in one system.

Most healthcare companies do not have this.

Hims does.

Mohamed ElShenawy, Hims’ CTO, explained it clearly on the call:

We’ve embedded intelligence at every step of the care journey, while keeping independent providers responsible for all clinical decision-making. For example, we currently have an AI copilot live on the provider side of the platform that drafts contextual responses on behalf of our care coaches, which are then reviewed by the care coaches before being sent.

This is exactly the loop.

AI drafts. Humans review. Humans correct. The system learns.

And then he made it even more explicit:

We currently support tens of millions of customer touch points annually and independent providers view, correct and approve AI inputs that make up our platform design. This means we are generating clinician-verified training signals for our models. This is the highest quality label in AI any company can hope for, and it can’t be acquired.

That sentence is the whole thesis.

Clinician-verified training signals.

This is not generic web data. This is not scraped text. This is not a chatbot guessing from public medical information.

This is real healthcare workflow data, corrected by providers, attached to actual patients and outcomes.

That is very hard to buy.

It has to be earned through volume, trust, workflows, regulation, and time.

And Hims has been building it for years.

I want to be much more direct here.

Providers are training the Hims AI model.

Yes, today providers remain responsible for clinical decisions. They have to. Healthcare is regulated, sensitive, and high trust.

But look at the architecture.

Hims is not simply adding ChatGPT on top of a telehealth workflow. Hims is building the workflow so the human provider corrects, approves, rejects, escalates, and labels what the AI does. That is exactly the feedback loop you need if your goal is to train proprietary healthcare models.

Here is the important part: providers are training the system that could automate a large part of provider work over time: AI handles the repeatable part, guardrails decide what can be done automatically, escalation rules decide what needs a clinician, and the clinician keeps improving the system every time he or she intervenes.

This is why I think Hims will eventually train at least two kinds of proprietary models.

First, a model to predict disease risk or likely conditions from symptoms, clinical history, labs, wearables, medication response, side effects, and outcomes. This may not even be a classic LLM. It could be a more specialized model, maybe an encoder-style system, because the task is not “write me a nice paragraph.” The task is: given this patient profile, what is probably going on?

Second, a model to recommend the next best treatment path from symptoms, history, predicted disease, prior responses, contraindications, and outcomes. In other words: if this is the likely problem, what should the platform suggest, prescribe, monitor, or escalate?

There can still be a frontier model as the orchestrator. There can still be a 24/7 companion talking to the subscriber. But the real IP, in my opinion, is not the chat interface. The real IP is the clinical prediction and treatment-decision layer trained on Hims’ own loop.

Management is already pointing in this direction.

Labs AI explains biomarker results in the unique context of each customer, flags what matters, and knows when to recommend escalation to a provider.

The AI weight-loss companion will proactively support customers, reach out at the right moment, and prompt clinician engagement when needed.

Providers are in the loop more to review and validate what AI does or would do, rather then the other way around. That is the architecture of an AI-native healthcare company.

Hims is not selling prescriptions online.

Hims is building an intelligent care layer between the consumer, provider, medication, lab result, wearable, side effect, and outcome.

That is a completely different business.

Switching cost as a function of time

I have a personal example I want to share.

I built a system based on Hermes Agent, which I call Max, to track what affects my health.

Max lives on a Mac mini. He has access to my Obsidian notes and my Google Sheets database. I speak to him on Telegram. He knows my diet, my habits, my baseline routine, and every day he asks me if I have anything to declare that deviates from normal.

It can be food, training, symptoms, sleep, supplements, anything worth registering.

I send a voice note and he logs it.

Every six months, I do blood work. I send Max a picture of the results and he helps me record the markers in the database.

The goal is simple: correlate what I do with how my biomarkers change, and then improve the markers over time.

For example, from October 2025 to May 2026, my testosterone increased from 3.89 to 5.1, probably helped by keto/carnivore, more sun, and heavy lifting.

This is useful.

But 99.9% of people will never build something like this.

They do not want to manage notes, spreadsheets, automations, reminders, and lab data.

They want the outcome.

If Hims can make this kind of proactive health companion easy, cheap, regulated, and personalized, then the value proposition becomes much stronger than “buy this treatment online.”

It becomes: stay on the platform because the platform knows you better every month.

The longer you stay, the more data you accumulate.

The more data you accumulate, the more useful the recommendations become.

The more useful the recommendations become, the harder it is to leave.

This is switching cost as a function of time.

And it is also a network effect, because every patient interaction, provider correction, treatment journey, and outcome can improve the system.

That is why Hims could become much more defensible than people think.

Hims building a moat

Management basically described the same flywheel:

With this infrastructure, we are strengthening the closed-loop proprietary data flywheel that competitors cannot recreate. We are one of the only platforms in healthcare where consumer intake and diagnosis, treatment journey, provider decisions and eventual outcomes all live in a single stack, and we’ve designed our platform to get stronger as these insights grow.

This is exactly what I want to own.

Not just a digital clinic.

A full-stack healthcare data loop.

Intent → intake → diagnosis → treatment → titration → side effects → outcomes → better models → better care → more customers → more data.

That is the flywheel.

And because Hims is consumer-first, the company is not waiting for employers, insurers, hospitals, or legacy systems to move.

It goes directly to the person who wants to feel better.

This matters.

The traditional healthcare system is not designed around consumer delight. It is slow, fragmented, and reactive.

Hims is trying to make healthcare feel more like a high-frequency consumer product.

That is why I think the company can win share in markets like weight loss, sexual health, dermatology, mental health, longevity, labs, and eventually broader preventive care.

The more categories they add, the more complete the customer profile becomes.

The more complete the profile becomes, the more valuable the intelligent layer becomes.

This is why the Eucalyptus acquisition also matters strategically, not just financially.

Different geographies mean different populations, disease profiles, regulations, medications, and care patterns.

That can make the model better.

Hims said it directly:

We believe our global scale will act as a force multiplier with models reflecting the populations, disease profiles and care patterns of multiple geographies, fundamentally improving the intelligence we are building.

This is the part I think investors are underestimating.

The data advantage is not just US GLP-1 demand.

It is global, multi-category, longitudinal consumer health data inside a regulated care workflow.

Lorenzo2cents (L2C) take aways and performance

I have never been more excited about Hims.

The vision and the catalysts are finally coming together at the same time.

GLP-1 demand is reaccelerating. Novo gives Hims a cleaner regulatory path. Eucalyptus adds international scale. Guidance looks too low. The AI/data flywheel is no longer hidden in the background. Labs, wearables, biomarkers, companions, and provider-verified training signals are turning Hims into something much bigger than a telehealth app.

This is the strongest version of the Hims thesis I have seen so far.

My position is fully built at around 10.3% of my portfolio.

But if the market gives me another dislocation, I could still use options to take advantage of it.

I do not own Hims because of a temporary GLP-1 trade.

I own Hims because I think it can become the default consumer healthcare platform for people who want to feel good, perform, lose weight, optimize biomarkers, and live longer.

That is a massive market.

And if Hims is right on the AI layer, this can become a winner-takes-most business.

As always, here is the “Deep Dive To Date” (DDTD), that is how the stock is performing since my initial deep dive on the February 8th 2025, when the stock price was $42.54.

-39% DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

I share my thoughts on X.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

Marc Boroditsky, CRO at Nebius, put it clearly:

The referenced pipeline growth of 3.5x a quarter, which -- 3.5x quarter-over-quarter, which we’re very proud of, is for our AI cloud business, and it does not include any strategic hyperscaler deals like the Meta deal. It does include qualified opportunities across our core AI cloud and Token Factory products as well as across all of our key customer segments, including AI natives, software vendors and enterprises.

What we can share about conversion is that we have maintained our solid win rates at the same time as we’ve accelerated our sales cycles and increased our average selling prices. And you can see this with some of the strong wins that we have, such as Sword Health in healthcare, life sciences and Rhoda and 1X in physical AI and core automation in one of our AI-native model builders as well as Revolut and monday.com, new customer wins for Token Factory.

What we are doing is enabling our go-to-market teams to have a consultative conversation with our customers about their plans and for current and future workloads, including, as an example, what they’re thinking about with regard to Vera Rubin. We’re also focusing and scaling our go-to-market and success teams to help customers to realize their plans, which turns into durable revenue for us.

This is exactly what I wanted to see.

Nebius is not simply selling GPUs into a hot market. It is starting to convert GPU scarcity into a broader AI infrastructure relationship. That is a very different business.

The company’s ability to deliver both high-performance, large-scale clusters and smaller-scale, on-demand compute is becoming a real advantage. In Q1, that advantage translated into a sequential re-acceleration in revenue. The momentum was broad-based and compounding:

• Pipeline generated in the quarter reached a new record, increasing approximately 3.5x quarter over quarter.

• Pricing continued to rise for new-generation GPUs.

• Older-generation chips also saw strong pricing support.

• Average deal sizes increased across both new and existing customers, driven by pricing, GPU commitments, and longer contract durations.

• Demand widened across verticals, model builders, enterprises, software vendors, AI-native companies, and inference workloads.

The most important detail is that the 3.5x pipeline growth excludes strategic hyperscaler deals like Meta. In other words, this is not a single large customer distorting the picture. This is the core AI cloud and Token Factory business accelerating.

Contracted capacity already exceeds 3.5 GW, far surpassing the 3 GW goal Nebius had set for the end of the year. Because of that execution, management raised its contracted power guidance to more than 4 GW by year-end.

This matters because, in AI infrastructure, demand is not enough. Everyone has demand. The question is who can convert demand into capacity, capacity into revenue, and revenue into a deeper customer relationship.

Nebius is starting to show that it can.

The World’s Biggest Factory Will Produce Tokens

My core thesis is simple: the biggest factory of the next decade will not produce cars, steel, semiconductors, or oil.

It will produce tokens.

My thesis is that AI can dramatically expand economic output over the next few years. But I also believe that, relative to the size of the economy, there will be fewer and fewer winners. This is because the companies that successfully build AI-powered flywheels will often capture entire market segments.

Once the data/AI flywheel is built, working, and spinning, it becomes extremely difficult to stop. The company gains more users, which creates more data, which improves the product, which attracts more users, which creates more data, which improves the product again.

The flywheel spins faster and faster.

And then another force kicks in: talent.

To make the data/AI flywheel spin, you need exceptional talent. Talent is naturally limited. And the best talent will increasingly concentrate inside companies where the flywheel is already spinning, because those companies will have more ambition, more resources, better data, better compensation, and a higher probability of winning.

So the companies that reach the “spinning flywheel” stage early will attract more talent. The companies that do not reach that stage soon enough will be left behind.

This is why I believe the ingredients of an AI flywheel are:

1. Visionary and brave management.

2. Data.

3. Talent.

But the first factor is the trigger.

Visionary management starts the process. It chooses the right market, takes the hard decisions, collects the right data, attracts the right people, and pushes the organization through the painful transition before competitors understand what is happening.

Once the flywheel spins for a few years, the company becomes almost impossible to catch.

I also believe that most of these AI flywheels will be built by AI-native companies, not incumbents. Most incumbents carry too much legacy. Their culture, incentives, shareholders, technical debt, and profit pools make internal disruption almost impossible.

The exceptions will be founder-led incumbents where the founder still has enough power, hunger, and vision to take painful decisions against the will of shareholders focused on short-term profit and risk avoidance.

$META is one example. Mark Zuckerberg has been systematically throwing billions at new bets and restructuring the company to take full advantage of the AI revolution.

Alphabet is another example. It has been willing to put at risk its highly profitable Search advertising business in order to push the AI-based business.

But most winners will be AI native.

And many of those AI-native companies will build on Nebius Token Factory.

Why?

Because Nebius is trying to offer something more valuable than raw GPU access. It is building an environment where AI-native companies can build with ownership, access the tools they need, and generate tokens at the lowest possible cost.

That is the key.

In an AI-native economy, tokens are not just a technical output. Tokens are the cost of doing business. They are the new unit of production.

Routing data from OpenRouter suggests proprietary models still dominate usage, with proprietary non-Chinese models averaging around 70% of token volume in 2025. But the direction of travel is clear: open models are gaining share as performance improves. My view is that this mix can flip over the next two to three years.

For now, open models are not good enough for many use cases. But if they reach something close to GPT-5.5 / Opus 4.7 level, everything changes. I think that could happen sooner than most investors expect, but this remains a forward-looking bet, not a fact.

There will always be specific use cases where organizations prefer frontier closed models. But over time, that percentage should decline. Most workflows will not need the absolute best model in the world. They will need the best cost/performance model for that specific job, integrated into a reliable workflow, running at scale.

This is where Nebius Token Factory becomes interesting.

My view is that Token Factory has the potential to become the preferred environment where companies build their AI-powered workflows. From there, they can still call external frontier models for specific tasks when needed. But the main workflow, the infrastructure layer, the orchestration layer, and the economics sit inside Nebius.

That would make Token Factory a gatekeeper.

And as human labor is increasingly leveraged or replaced by AI, a larger and larger percentage of every company’s cost structure will be related to token generation.

That is why I believe Nebius Token Factory has the potential to become the world’s biggest factory.

Not because it will produce the most GPUs.

Because it can become one of the places where the new economy produces work.

Significant Token Factory customer wins in Q1 include:

• Revolut, which is developing a platform to simplify finance for businesses and consumers, was able to remove human intervention from 80% of support chats and handle 1.2M chat tickets per month on Token Factory.

• monday.com selected Token Factory to support its AI work platform that helps manage, orchestrate, and execute workflows.

These customers show why Nebius is not just another GPU reseller. The company is moving up the stack, closer to where the customer’s AI workload is actually built and scaled.

Roman Chernin explained the logic very well:

First of all, I want to say that we are super excited that these 2 incredible teams of talented people from Eigen and Clarifai will join us. And to deep dive into the rationale, let’s start from foundations.

Our view is that we should own the compute stack. That is where our vertical integration, our supply chain depth and our hardware engineering generate advantage and it’s also the layer that drives the bulk of our economics. But the compute stack, we built the full cloud solution and software plays the role of enabler. By the way, we partner where partnership is the right path. And as you see now, we use M&A selectively where it accelerates our road map, brings in proven developer adoption or as capabilities complementary to what we are building.

Acceleration is the key lens we apply to every potential transaction where we can find rare talent or proven adoption. This is, by the way, the example of Tavily that has incredible developer adoption that would, in general, take us meaningfully longer to build organically, acquisition is the fastest path. And we evaluate every potential deal against the clear criteria. Does it deepen customer engagement, increase lifetime value, unlock the new category of customers or use cases we can address and in general, strengthen our position as a full stack AI cloud.

This is exactly the right strategy.

Nebius does not need to become a traditional software company. It does not need software as a separate revenue stream. It needs software as an enabler of compute consumption, customer stickiness, workload expansion, and lifetime value.

Roman made this even clearer:

As I said, we look at the software as an enabler. So it’s not that we build the software to generate a separate revenue stream. The software first of all plays the role of unlocking the new capabilities for us, unlocking the new opportunities in the types of the workloads that are growing on the market and the types of customers that we can address.

Software changes the shape of the customer relationship. Every layer of software unlocks another group of users and customers. We want to meet customers where they need us and let them consume our vertically integrated solution in the way that they need, and it might be a different way for different types of customers.

Customers come to our platform for different needs. In essence, they all need to run AI at scale — which means they need compute. But for example, people who use our multi-tenant cloud, they -- to a big extent come for large training jobs. People who -- and these are like research-driven, data scientist-driven workloads. People who come to Token Factory, they build vertically integrated vertical AI products or apply AI in their enterprises. And they come for the tokens. And moving forward, we’ll see new ways to consume infrastructure at scale that will be the agentic -- end-to-end agentic workloads.

This is the point most investors still miss.

If you only look at Nebius as a GPU cloud, you compare it with every other company that has land, power, and Nvidia chips.

But if you understand Token Factory, the comparison changes.

Nebius is trying to become the vertically integrated infrastructure layer where AI-native companies generate tokens, build workflows, deploy agents, scale inference, and eventually run entire AI operating systems.

That is a much bigger opportunity than renting GPUs.

Why Nebius is well positioned to win the AI supply chain bottleneck race

The most common argument I hear from investors and finfluencers is that power supply is the new bottleneck for AI infrastructure. Therefore, companies like IREN, which already own a decent amount of connected power, will win the race against companies like Nebius, which supposedly will not build enough capacity on time to serve demand.

Supporters of this thesis even claim that companies like IREN are delaying deals on purpose, waiting for the moment when the supply/demand imbalance is at its peak, so they can extract better terms because they own the scarcest resource.

I recognize the bottleneck. It is real.

Power matters. Land matters. Copper matters. Grid connection matters. Speed of execution matters.

But I think this thesis is still too shallow.

It can work in the short term. In a very tight market, the owner of the scarce physical resource can capture a lot of value.

But over the medium to long term, let’s say beyond three years, the winner will be the company that delivers the highest value per token.

And to deliver the highest value per token, you need to deliver now.

A few months of delay can mean losing the race. In AI, timing is not a detail. Timing is strategy.

The AI race is not just about land and copper. That is the physical layer. It matters, but it is not the full game.

The AI race is about spinning the AI flywheel faster than anyone else in your domain.

We are already seeing this with models.

OpenAI has more compute available than Anthropic. Yet Anthropic has been gaining a lot of ground, especially in coding and enterprise. Why? Because Anthropic found a way to spin the AI flywheel faster in a specific domain.

It focused on coding and enterprise instead of trying to win every consumer use case at the same time. The iteration pace in that domain is faster. The feedback loops are clearer. The willingness to pay is higher. And once your LLM can improve the workflows of the people building the next LLM, the flywheel becomes extremely powerful.

At that point, the chance for a competitor to catch up becomes much lower, unless the leading company hits a technological limit soon enough for laggards to recover.

The obvious objection is:

What about compute? Anthropic is constrained. OpenAI will catch up.

But Anthropic just got helped by Elon.

And that is exactly my point.

If you have the talent, the demand, and the best product, you will find the resources.

Resources are attracted to the winning player.

As a supplier of power, land, compute, capital, or any constrained resource, who would you rather sell to? The company with the strongest product and fastest flywheel, or the company with idle capacity and no clear edge?

This is the same dynamic we see with talent. Talent flows to the winning player. Capital flows to the winning player. Suppliers flow to the winning player. Partners flow to the winning player.

This is why I do not believe the AI infrastructure race will be won simply by whoever has the most connected power today.

Connected power is valuable.

But connected power without a differentiated customer relationship becomes a commodity.

Actually, I think the market is framing the bottleneck in the wrong unit.

The scarce resource is not power in isolation.

The scarce resource is useful tokens per watt.

This is where software becomes more important than power.

SemiAnalysis InferenceX data shows how dramatic the difference can be. According to NVIDIA, recent SemiAnalysis InferenceX data shows that NVIDIA software optimizations and Blackwell Ultra GB300 NVL72 systems can deliver up to 50x higher throughput per megawatt and 35x lower token cost than Hopper for DeepSeek-R1.

Think about what that means.

If one operator owns 1 GW of connected power but converts that power into tokens inefficiently, and another operator owns less power but converts every megawatt into dramatically more useful inference, the second operator may generate far more economic output.

The market sees megawatts.

Customers see latency, reliability, model quality, throughput, cost per token, developer experience, and time to production.

Capital sees revenue per megawatt.

That is why the most important metric in AI infrastructure will not be only connected power. It will be revenue per megawatt.

Power is the input. Tokens are the output. Software determines how efficiently one becomes the other. That is why “I own power” is not enough as a long-term thesis. Power without software optimization is like owning oil without a refinery: valuable, but not the highest-value part of the chain.

The best AI infrastructure companies will not only secure electricity. They will continuously improve scheduling, orchestration, utilization, inference serving, model optimization, networking, cooling, workload routing, reliability, and customer workflow integration.

That is exactly where Nebius is trying to go with Token Factory.

The real question is: who can turn power into tokens, tokens into workflows, workflows into customer lock-in, and customer lock-in into a compounding data/software/infrastructure flywheel?

That is where Nebius is better positioned than the market understands.

Nebius is not just trying to build capacity. It is trying to build the AI supply chain layer where customers come not only for compute, but for tokens, tools, workflows, and eventually agentic infrastructure.

That is a much more durable position.

Why Nvidia Has Every Reason to Back Nebius

The second key piece of the thesis is Nvidia.

As hyperscalers increasingly diversify chips and invest more in proprietary silicon, they are becoming, de facto, competitors to Nvidia.

Google has TPUs. Amazon has Trainium and Inferentia. Microsoft is working on its own chips. Meta is developing more internal AI silicon. Tesla and xAI may eventually go deeper into custom silicon as well.

This does not mean Nvidia is weak. Nvidia’s moat is still extremely strong.

But it does mean this will not be a winner-takes-all chip market forever.

Nvidia needs to defend its ecosystem, keep customers locked into CUDA, maximize distribution, and avoid becoming overly dependent on hyperscalers that are increasingly incentivized to reduce Nvidia dependence over time.

This is why neoclouds matter.

Recent news about Google and Blackstone creating a new AI cloud company confirms the direction. For Google, it is not enough to use TPUs internally and inside Google Cloud. It is building an external, neocloud-like distribution surface for its chips.

The probability that Amazon eventually does something similar is high. Elon could follow in a few years. AMD could do the same. Chinese players may eventually enter the American and European markets as well.

The implication is clear: chip companies need distribution.

And Nvidia has every incentive to support strong independent distribution partners more than ever.

Neoclouds like Nebius, CoreWeave, and IREN will be key to that strategy. Over time, they can become one of Nvidia’s most important go-to-market surfaces, especially for AI-native companies, model builders, enterprises, and software vendors that do not want to depend entirely on hyperscalers.

But not all neoclouds are equal.

The winner will not simply be the one that buys the most GPUs. The winner will be the one that creates the most customer stickiness.

Stickiness is not built by selling bare metal.

That is why I think IREN’s long-term upside depends on whether it can move beyond scarce power and build real software/workload stickiness. If it remains mostly a capacity seller, I think it will be structurally disadvantaged versus Nebius and CoreWeave.

IREN now seems to be trying to pivot toward a Nebius-like model, more focused on software edge and more targeted to end users instead of only hyperscalers. But in my view, the burden of proof is now on IREN to show it can build that ecosystem before Nebius and CoreWeave compound their lead.

The real race is probably between CoreWeave and Nebius.

CoreWeave has strong execution, scale, and market credibility. But Nebius is more clearly focused on building a lasting moat around the Token Factory ecosystem.

That is what makes Nebius different.

It is not only trying to be a GPU landlord. It is trying to become the place where AI-native companies build, deploy, and scale their token production.

This is, in my view, why Jensen has been so vocal about Nebius lately.

He understands that Nebius can become one of Nvidia’s most important distribution channels. Not because Nebius is the biggest today, but because Nebius is building the kind of customer relationship Nvidia needs in the next phase of the AI infrastructure race.

Nvidia also understands the quantitative side better than anyone.

If SemiAnalysis data shows that software optimizations and new Nvidia systems can create up to 50x more throughput per megawatt and 35x lower token cost than the previous generation for a relevant inference workload, then Nvidia’s objective is obvious: put those systems in the hands of operators that can actually turn them into scaled token factories.

Not every owner of power can do that.

The best partner for Nvidia is not the company with the loudest announcement about megawatts. It is the company that can convert Nvidia’s full-stack platform into customer workloads, high utilization, developer adoption, and recurring token demand.

This is why Nebius matters strategically: it gives Nvidia a partner that can turn full-stack hardware and software into recurring token demand, not just one-off GPU capacity.

That is why Q1 mattered.

Not because one quarter changes the thesis.

But because the quarter showed that the thesis is starting to play out:

• Demand is accelerating.

• Pipeline is exploding.

• Contracted power is already ahead of plan.

• Token Factory is landing real customers.

• Software is being used as an enabler of compute consumption and customer lock-in.

• Management understands the strategic importance of owning the compute stack while moving high enough in the stack to shape customer behavior.

• Nvidia has every reason to support the strongest independent AI cloud distribution partners.

The market still wants to reduce Nebius to a simple AI infrastructure story.

I think that is wrong.

Nebius is not just building data centers.

Nebius is trying to build the token factory layer of the AI economy.

And if the economy really becomes increasingly AI-native, the biggest factory in the world will not be measured in square meters, machines, or human workers.

It will be measured in tokens.

Lorenzo2cents (L2C) take aways and performance

My core takeaway from this quarter is simple:

1. Demand is outpacing supply.

2. Nebius is scaling capacity fast.

3. AI-native customers are becoming enterprises.

4. Nebius is building a software moat.

5. And open-source models are an underrated tailwind because they accelerate tokenization and commoditize the model layer, pushing value into delivery.

If Nebius becomes the best “token factory” for robust performance at the cheapest price, then it is not “a neocloud”.

It is an emerging hyperscaler.

And that is exactly why I’m here.

As always, here’s the Deep Dive To Date (DDTD): Stock performance since my first deep dive and when I bought in on July 28, 2025, at $51.63.

4x DDTD

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

I still believe that.

The app is getting better. Speaking practice is becoming more central. AI is helping them create content at a speed that would have been impossible a few years ago. The long-term thesis, in my view, is still the same: Duolingo is not just a language app. It is trying to become the learning ontology. The place where AI is not just a chatbot, but a teacher.

That was the core of my last $DUOL update: Duolingo’s AI Pivot: The Teacher Model Thesis.

But Q1 2026 also exposed the other side of the story.

Duolingo is becoming a better teacher, but it may have a problem finding enough new students.

And for a company valued on years of user growth, that matters.

Q1 2026: good numbers, but the mix matters

The headline numbers are not bad.

In Q1 2026, Duolingo reported:

• DAUs of 56.5 million, up 21% YoY.

• MAUs of 137.8 million, up only 6% YoY.

• Paid subscribers of 12.5 million, up 21% YoY.

• Revenue of $292.0 million, up 27% YoY.

• Total bookings of $308.5 million, up 14% YoY.

• Free cash flow of $147.8 million, a 50.6% FCF margin.

So this is not a broken business.

Actually, the opposite. Duolingo is still growing revenue at 27%, printing free cash flow, and improving monetization. Paid subscriber penetration increased to 9.4% from 8,9% in Q1 2025 and 9,3% in Q4 2025.

The company is clearly getting better at farming the existing base.

The problem is the top of the funnel.

MAUs grew only 6% YoY and 3,5% QoQ, while DAUs grew 21% YoY and 7% QoQ.

Being MAU a leading indicator for DAU, and being the former in deceleration, the latter will likely follow soon.

That means engagement is improving, but the pool of users is not expanding nearly as fast as before.

DAU/MAU went from 35.8% to 41.0% YoY.

That is impressive product execution.

But it also tells you where the growth is coming from: retention, habit, conversion, monetization.

Not enough from new user acquisition.

The bull case is still alive: AI is making Duolingo a better product

The best part of the quarter was product velocity.

Duolingo is attacking the biggest weakness of the app: speaking.

Management wrote in the shareholder letter:

Speaking practice, historically the biggest gap for learners on Duolingo, is now a core part of the product.

They added that Video Call remains one of their most impactful paid features and that, over the last year, they have “more than doubled the average number of words spoken per user in Video Call.”

This matters because Duolingo’s historical weakness was obvious.

It was great at making you show up every day. It was great at teaching vocabulary and grammar in small chunks. It was not great at making you speak.

That gap is now closing.

They introduced spoken tokens, flashcards, and Speaking Adventures. Management described spoken tokens as a change that “enables almost every exercise in Duolingo to become speaking practice.”

This is exactly what I want to see.

If Duolingo becomes the place where a beginner can go from zero to real conversation, the product becomes much more valuable. Not because it adds another feature, but because it moves closer to the real job to be done: teach me until I can actually use the language.

AI is also changing the content creation curve.

This was the strongest quote of the quarter:

In Q1 2026 alone, we published 20,500 course units across our language courses, up from 7,100 per quarter in 2025 and 1,800 per quarter in 2024.

Management added that automating more of the content creation process lets them push changes across many courses at once and improve quality faster.

This is the AI tailwind.

Most consumer AI stories are vague. Duolingo’s is very concrete: more content, faster iteration, more speaking, more personalization, lower content bottlenecks.

This supports my teacher model thesis.

A generic LLM can translate, explain grammar, or roleplay a conversation. But Duolingo has the structured curriculum, the user behavior data, the retention loops, the A/B testing engine, and the feedback loop on what actually helps people learn.

In my view, that is the moat.

The raw model is not the product. The learning system around the model is the product.

But the funnel is the problem

Here is the part I do not like.

On the call, they were asked directly about MAU growth and top of funnel. Luis von Ahn answered:

The reality is that top of funnel has been about flat for certainly for this quarter, and we would like to accelerate it.

He then explained that the historical growth driver has been word of mouth:

Again, the main thing that has been responsible for top of funnel historically has been word-of-mouth. And word-of-mouth is this interesting thing that is beautiful because it is free, but we do not have that much control over it in terms of being able to measure it the same way that we can measure retention.

That is the key issue.

Duolingo has world-class retention mechanics. They can measure them. They can A/B test them. They can improve them.

But the main top-of-funnel engine is word of mouth, and management openly admits they do not have the same level of control there.

The Google Trends screenshot makes the same point visually.

Duolingo reached peak search interest last year around the Dead Duo campaign. That campaign was genius: almost free, impossible to ignore, and very Duolingo. But that is exactly the point. You cannot schedule virality every quarter.

After Dead Duo, interest did not keep compounding. It got worse. That does not mean the brand is dead. It means word of mouth is not a controllable machine. It is a powerful engine when it works, but it is not the same as paid distribution, network effects, or owned social graphs.

This is not fatal.

But it is a different type of risk.

If MAU growth keeps slowing, DAU growth will eventually follow. DAU/MAU cannot expand forever. At some point, the ratio saturates. When that happens, the business needs new users again.

This is why I am more cautious after Q1.

The market is not wrong to ask whether Duolingo’s growth is structurally slowing. The question is whether this is a temporary transition year, while management invests in teaching better and user growth, or whether the core organic acquisition engine is reaching saturation in the most mature markets.

The D2C lesson: distribution is everything

The more I follow companies like Duolingo, Oddity, Hims and Lemonade, the more I think the same lesson keeps coming back.

D2C is brutally hard.

When it works, it looks magical. You own the customer. You own the data. You own the product loop. You can compound without paying rent to a platform forever.

But distribution is still the bottleneck.

Meta sits on a goldmine because it owns attention and distribution. Most D2C companies have to fight for it every day.

Duolingo’s advantage was that it escaped the usual D2C trap. It did not need to spend heavily on performance marketing because the product itself created word of mouth. The owl became the distribution channel.

That was a superpower.

Now it may also be the constraint.

Luis was very honest about performance marketing:

Performance marketing at Duolingo has been this interesting thing that because our free version is so good, it has not been easy for us to do profitable performance marketing because what happens is we acquire people and then they are super happy as free users.

He added:

At the moment, in some geographies, we have profitable performance marketing, but in many geographies, we do not.

This is almost an Innovator’s Dilemma.

Duolingo’s free tier is the reason the company became huge. It creates trust, habit, brand love, and scale. But the same free tier makes paid acquisition harder because many acquired users land in the free product and stay there.

If they make the free tier worse, they damage the product and the brand.

If they keep it too good, performance marketing ROI is limited.

So the business has to thread a very narrow needle: keep the free product excellent enough to drive word of mouth, while making the paid product valuable enough to justify acquisition spend.

That is not easy.

Asia and China may be the answer

There is one important counterpoint.

The problem is not equally bad everywhere.

Luis said Asia is the fastest-growing region, and China is particularly interesting. On the call, he said:

One of the things that is interesting about Asia is in a number of large Asian markets we can do profitable performance marketing

He also said: “We are able to acquire profitably” in China, where the users are English learners.

That matters.

If Duolingo can use performance marketing profitably in underpenetrated Asian markets, then the top-of-funnel issue may be more of a mature-market problem than a global problem.

The China brand partnerships are also interesting.

Luis mentioned Luckin Coffee and a coming McDonald’s China partnership, and said:

We have had incredible brand partnerships in China.

But he also made the more important point:

China is not just growing fast because of the great partnerships. I think it is kind of the other way around. I think the great partnerships are coming in part because we are growing fast and because we are seen as a very cool brand.

This is the bull version of the story.

Duolingo may not need to solve paid acquisition everywhere. It may need to professionalize it where it works, use partnerships where the brand has pull, and keep improving the product so word of mouth reaccelerates.

That is possible.

But it is not yet proven.

The other answer: make learning multiplayer

There is another path I would like to see more aggressively.

Duolingo should take the chess playbook and apply it to languages, math, music, and every course where it makes sense.

In Q3 2025, management wrote that chess had become Duolingo’s fastest-growing subject. In the same shareholder letter, Luis also said they had introduced player-vs-player on the chess course.

That is the most interesting user acquisition clue, in my view.

Multiplayer changes the growth equation.

A solo lesson is retention.

A multiplayer lesson can become distribution.

If I challenge a friend to a chess puzzle, a vocabulary duel, a speaking battle, a grammar race, or a math challenge, I am not just using the product. I am pulling someone else into the product.

That is how word of mouth becomes less random.

Today, Duolingo’s social graph is still relatively weak compared to what it could be. Streaks, leagues and leaderboards are powerful, but they mostly make the individual user come back. Multiplayer can make users bring other users back.

This is where the product can create its own acquisition loop.

Imagine language learning with real-time speaking battles. One user learning Spanish, another learning English. Short, structured, gamified conversations where both users improve and both users have a reason to invite friends.

That would be much closer to a true learning network.

And it fits Duolingo perfectly because the company already knows how to make hard things feel like a game.

The 2026 strategy sounds like “improve the product and pray”

I am exaggerating a little, but not much.

Management’s strategy for user growth is basically:

1. Teach better.

2. Improve speaking.

3. Make the free tier more engaging.

4. Invest more in marketing, especially in underpenetrated regions.

5. Hope better learning outcomes create more word of mouth.

This can work.

Actually, it is probably the right strategy for a product like Duolingo. You do not want management to ruin the product with aggressive monetization. You do not want them to turn the owl into a CAC machine.

But as an investor, I also have to be honest: this is less controllable than retention optimization.

Luis said it himself. Word of mouth is beautiful because it is free, but they do not have the same measurement and control they have with retention.

That is the risk.

L2C take aways and performance

I am still bullish on Duolingo long term.

The product is improving. AI is a real tailwind. The company is highly profitable. The balance sheet is strong. The learning ontology thesis is intact.

But Q1 made the main risk clearer.

The question is no longer: “Can Duolingo use AI to make the product better?”

I think the answer is yes.

The question is: “Can a better product restart top-of-funnel growth?”

That is much harder.

If the answer is yes, then $DUOL remains one of the most interesting consumer AI companies in the market. It would mean the company can use AI to improve teaching quality, increase engagement, expand monetization, and eventually reaccelerate user growth.

If the answer is no, then Duolingo may still be a very good business, but the multiple has to come down because the market will stop underwriting hypergrowth.

For now, I see Q1 as a yellow flag, not a red flag.

The business is not deteriorating. It is evolving.

Duolingo is becoming a much better teacher.

Now it needs to prove it can still find new students.

Looking at my Business Ontology visual, while Duolingo is performing well on the first two pillars (“Engagement” and “Paid Subscription Penetration”), it needs to improve a lot on User Growth.

As always, here is the “Deep Dive To Date” (DDTD), that is how the DUOL stock is performing since my initial deep dive on the January 11th 2025, when the price was $318.15.

-67% DDTD

I remain long .

If you want access to:

• My portfolio

• all my trades in real time

• my price targets and how I calculate them

• Business Ontology content

Join the Business Ontology Community

My price targets will be shared in the Telegram Group and explained on my Youtube channel.

Please note that:

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

So what becomes scarce?

The domain map.

The workflow.

The data.

The ability to turn AI into action inside a specific industry.

That is what I call a vertical ontology.

$LMND is building it in insurance.

Customers, policies, claims, risk, fraud, pricing, and service interactions. Insurance is a perfect domain for an ontology because the business is basically a machine for pricing uncertainty. The better the object graph, the better the underwriting. The better the claims loop, the better the data. The better the data, the better the model.

That is the flywheel.

I have been very vocal about this, but let me say it again.

Lemonade is becoming a vertical ontology because it was built around the idea of becoming an autonomous organization. Its IT systems and business processes were designed for that from day 1. And it runs mostly a D2C model, which gives it direct access to first party customer data. That data, plugged into those systems, makes Lemonade a vertical ontology that keeps learning in a reinforcing cycle.

No incumbent will ever achieve it. I think it is almost impossible because of the innovator’s dilemma, which I have already explained multiple times.

Now, on to Q1 earnings, which were stellar. Probably the best in the company’s history.

• Topline growth acceleration: At $1.33 billion, In Force Premium grew 32% – extending their streak of IFP growth rate acceleration to ten consecutive quarters.

• Revenue grew faster still, by 71% to $258 million, reflecting the impact of recent reinsurance transition and higher premium retention.

• Gross Profit: Increased 159% YoY to $100 million

• Q4 2026 EBITDA positive guidance confirmed.

So it is time for an update of my Business Ontology for Lemonade. If you want to know more about what a business ontology is, let me know in the comments and I will write more about it.

Business Ontology Framework by L2C

The Business Ontology is a framework I built after tearing apart several tech companies from the ground up. I break them down to their basic parts, then piece together a real thesis on what drives them. Think of it as a map of a company’s soul. It is a tight set of core indicators, tailored to each business, that shows whether it is heading in the right direction, no matter what the stock price says. These are not basic stats like P/E ratios or revenue bumps you grab from Yahoo Finance. They are deeper, sharper, and linked directly to the thesis I have built on how the company creates value and competes in its market.

The framework boils down to two big pieces:

1. The Business Ontology—This checks whether the company is worth buying into (or holding).

2. The 4D Valuation Model—This gives you a 4D roadmap to guide your own decisions on how much to allocate.

Now, let’s dive into Lemonade and this quarter’s update.

Business Ontology

Here’s a quick visual on the Business Ontology for Lemonade.

Annual Dollar Retention

Annual Dollar Retention has been stable at 85% for a few quarters, after trending down from 88% in Q1 2024.

Management said the decline is primarily due to the continuing impact of their “clean the book efforts” in the home business. It should start rising again soon. There is no reason to doubt it after looking at the leading indicators:

App Store and Trustpilot rankings are above 4, which is extraordinarily strong compared to peers.

Happy Customers keep using the product.

IFP Per Employee

IFP per Employee keeps rising and is about to overtake ROOT Insurance. For context,

Lemonade Premium per Customer is $424. ROOT Premium per Policy is $1423.

I could not find a perfectly comparable metric, but if you compare apples to apples the difference must be even more staggering (since a customer can have more policies).

The point is that ROOT should be much more efficient than Lemonade because any acquired or managed policy is worth on average 3 or 4 times more than Lemonade.

This is only a benchmark, but the takeaway is that Lemonade’s AI engine is firing on all cylinders, and my original thesis is slowly coming true.

The LAE ratio confirms the same point. At 6%, it is among the lowest in the industry.

Loss ratio

Loss ratio is at a healthy level: 62%. I do not expect it to go much lower than this, because Lemonade, as said multiple times, is optimizing the business to maximize gross profit, not to minimize loss ratio. The end goal is to make more money, not to have the lowest loss ratio in the industry.

To get a feel for how the company manages the trade-off between loss ratio and growth (lower loss ratio usually means higher prices, which usually means lower growth, and vice versa), let’s check my proprietary metric: Adjusted Rule of 40.

The higher, the better. Lemonade printed 69 in Q1, which is an excellent result that, if maintained, would drive outstanding gross profit growth.

As management expects the car business to drive most growth in the near future, let’s see how the same metric is performing for the car segment.

The car business is growing healthy and strong, as shown by the positive trend in this metric. It prints even better than the overall business, even though it is not perfectly comparable. The car business is still tiny in absolute terms, so it benefits from “easy” hyper-growth that inflates the metric.

If you want access to the source of all the data I track, including what I used to update my business ontology, subscribe to Lorenzo2cents.

If you are already subscribed and you still do not have access to the Business Ontology Knowledge base, join the Business Ontology Community (Telegram channel) here.

Join the Business Ontology Community

4D Valuation Model

If you would like access to my 4D valuation model, including my price target, let me know in the comments of this post or in the Telegram channel. I will share it only if there is interest from enough people.

Lorenzo2cents (L2C) take aways and performance

My thesis on Lemonade—growth with near-zero marginal costs—is fully unfolding. As of now, there is not a single reason to doubt that 30% IFP growth for the foreseeable future, as guided by management, is achievable.

Even more, Lemonade is positioning itself to conquer a totally new market: autonomous agents and objects, which will be even bigger than the current insurance industry. If Lemonade captures a 5% market share of the US Property & Casualty (P&C), Auto Insurance, and autonomous objects market in 10 years, it will likely be a $1 trillion company.

As always, here is the “Deep Dive To Date” (DDTD), which shows how the stock is performing since my initial deep dive on September 24th 2024, when the stock price was $17.23.

3.3x DDTD

Lemonade (LMND) remains my highest-conviction holding and largest portfolio position.

Unlike other names (e.g., HIMS), Lemonade faces no acute, company-specific legal or regulatory headwinds that would derail its trajectory. Standard industry risks (e.g., compliance, reinsurance, AI model scrutiny) exist, but nothing idiosyncratic suggests near-term impairment.

The position is fully built. I have no plans to sell or add, except that I may use options or other trades to take advantage of market dislocations from real value, as I did in the recent drop. If you want to follow my moves in real time, join the Business Ontology Community on Telegram.

Join the Business Ontology Community

There you will also find the link to access my portfolio.

Here is an example of a message I sent on Monday, when I swapped my Meta shares for ZETA Global shares.

I expect Lemonade to be the primary driver of portfolio returns over the next 5+ years.

I can be and will be (hopefully not often) WRONG. This is just my personal strategy—NOT FINANCIAL ADVICE. I don’t know your financial or life situation well enough to give any recommendations. Please do your own due diligence and research. Don’t be LAZY.

Be the architect of your own destiny.

Ciao

Lorenzo

• Writing copy gets cheaper.

• Designing landing pages gets cheaper.

• Building software gets cheaper.

• Producing content (video, images, ads) gets cheaper.

• Soon, with humanoid robots, even physical goods and services will get cheaper to produce and deliver.

When production becomes commoditized, differentiation shifts to distribution.

In practice, that means a higher % of enterprise revenue will go to:

• demand generation

• retention and lifecycle marketing

• personalization and automation

• measurement and attribution (or better: outcome-based ROI)

Marketing turns into the main “battlefield” because every product starts to look similar, faster.

I bet that the following downtrend is going to reverse soon.

As the ad market consolidates into “walled gardens” (Meta, Google/YouTube, Amazon, TikTok), a larger share of every incremental marketing dollar gets spent inside closed ecosystems where the platform owns the identity, the inventory, and—crucially—the measurement.

In that world, the “tax” advertisers pay is not only higher CPMs; it’s also the inefficiency created by siloed data and last-click-style reporting. That’s where Zeta becomes a commission to pay on walled gardens: a layer that sits on top, resolves identity, and optimizes spend allocation based on outcomes—not on whatever each garden claims. Zeta’s data cloud is built around a durable person-level graph (hundreds of millions of opted-in people, with thousands of attributes per profile) and its first-party pixel footprint across massive web surfaces, which gives it an independent signal set to judge incrementality and ROI across channels.

The Meta partnership is a concrete example of why this matters: David Steinberg has said Zeta has “a matching between the Zeta ID and the Meta ID,” which lets Zeta connect exposures in Meta to downstream actions and then orchestrate the next best message/channel (e.g., follow with CTV, email, or another paid impression) based on the highest expected ROI. Put differently, if walled gardens are where the budget must live, Zeta is the vertical “marketing ontology” that makes that budget work harder—because it understands who the person is (identity), what to do next (propensity), and where to spend (allocation) in one closed-loop system.

If you have not read it, check my ZETA deep dive part 1:

If marketing is the battlefield, ROAS is the weapon

In a world where everyone can build, the winners are the companies that can sell.

For an enterprise buyer, the “best marketing cloud” is not the one with the most features. It’s the one that:

• gets the highest ROAS (return on ad spend)

• makes ROAS more predictable

• scales ROAS across channels

• proves it financially (CFO-grade reporting)

That’s why the category is structurally winner-takes-most.

If Vendor A gives you 2x ROAS vs Vendor B, you don’t “diversify vendors”. You reallocate budget.

In the last earnings call, David Steinberg (CEO) framed it very directly:

• The marketing ecosystem grew about ~10%, while Zeta grew ~30% — and he explicitly said that this is Zeta “taking meaningful market share”.

• On wallet share: he said Zeta’s 603 global enterprise clients spent about $100B in marketing last year and Zeta captured ~1.3% of that spend (~$1.3B revenue). (Math check: 1.3% × $100B = $1.3B.) He also shared a long-term goal of 10% wallet share.

• Retention backs it up: net revenue retention hit a record 120% in 2025 (up from 114% in 2024).

• He also explained the basic mechanism: “the higher we drive the return on investment, the greater the percentage of budget that our clients move to us.”

This is the “budget gravity” thesis:

1. deliver superior ROAS

2. get more wallet share

3. invest more in product + data

4. deliver even better ROAS

Flywheel.

Zeta is stealing share to Salesforce, for comparison.

If Zeta grows ~30% while the overall ecosystem grows ~10%, somebody has to be losing share.

Salesforce is a natural place to look because it’s one of the biggest incumbents in enterprise marketing.

One public data point that supports the narrative:

• Salesforce’s Marketing + Commerce segment declined ~1% YoY in Q4 FY26 (constant currency), according to Salesforce’s quarterly breakdown.

Why I think Salesforce (and other incumbents) can’t easily “catch up”:

• Legacy architecture + latency: many suites were assembled via acquisitions; stitching data + channels together often adds delay and complexity. In a ROAS world, milliseconds matter.

• Different center of gravity: incumbents are optimized around being the system of record (CRM). Zeta is optimized around being the outcome engine (identity + activation + measurement). Those are different products, cultures, and data assumptions.

• Signal disadvantage: Zeta’s data cloud/pixel footprint means more third-party-like signals feeding optimization loops. Incumbents can buy AI, but it’s harder to buy continuously-updating proprietary signals at that scale.

• Enterprise switching dynamics: when a platform shows provably higher ROI, budgets migrate faster than core CRMs do — so challengers can win share even if the incumbent “stays installed”.

This is not “proof” by itself — but it’s exactly the type of footprint you expect when:

• legacy stacks slow down

• budgets get reallocated toward higher-ROI platforms

• enterprises separate “CRM system of record” from “marketing outcome engine”

Zeta is trying to be that outcome engine.

The data moat is the product

Zeta’s moat is not “AI”. AI will commoditize.

Zeta’s moat is data + identity + signals at scale, and the ability to activate those signals across channels.

Steinberg gave the most important numbers in one quote:

“We have 552 million active people who are opted in our data cloud today with an average of 5,000 to 7,000 data elements per person. Our first-party tracking pixel sits on 1 trillion pages of content and ingest trillions and trillions of proprietary marketing signals…”

Those numbers matter because they describe:

• scale (552M people)

• depth (5k–7k data points per person)

• freshness (signals ingesting continuously)

• distribution surface (pixel on 1T pages)

That is the raw material to train models, build propensities, measure incrementality, and optimize campaigns in near real-time.

Why this matters now more than ever?

Steinberg also made an important observation about the current shift in traffic:

• Google now answers 60%+ of queries on its own platform (down from the old world where “90%+” of queries were sent out to publishers).[1]

The conclusion is clear:

When “free clicks” disappear, marketing becomes harder and more expensive.

And when marketing becomes more expensive, the value of a platform that can:

• find the right audience

• personalize the message

• measure outcomes

• reallocate budget dynamically

…goes up.

The “ROAS compounding” piece: why Athena matters

Steinberg used an analogy I love:

Zeta built an “F-22 fighter jet”, but clients were flying it like a “Cessna”. Athena becomes the copilot that lets customers use more of the machine.

This matters because it points to a second order effect:

• If Athena makes customers more effective, their ROAS goes up

• If ROAS goes up, their budget moves to Zeta

• If budget moves to Zeta, Zeta gets more signals

• If Zeta gets more signals, the models get better

It’s not “AI feature”. It’s distribution + data + ROI loop.

Key risks

• Regulation / privacy: Zeta’s advantage is signals + identity (including its pixel footprint). Any tightening around consent, tracking, data sharing, or cross-site identifiers could reduce signal quality or raise compliance cost.

• Data-source scrutiny: even if data is “opt-in”, the market is increasingly skeptical about where signals come from and how durable the permissioning really is.

• Steinberg reputation / past controversies: the CEO’s history (and periodic short-seller narratives) can raise the bar for trust and keep the valuation multiple suppressed even when fundamentals are strong.

My take

The Zeta thesis is not “marketing grows”.

It’s: marketing becomes the profit engine in a commoditized world.

And in that world, the company that can prove higher ROAS will win share, even against the biggest incumbents.

Zeta is explicitly telling you this is already happening — and the growth delta (30% vs ~10% ecosystem) plus the early cracks in legacy segments support it.

Potential catalysts from here:

• Athena going GA (announced March 24, 2026) could increase adoption inside existing accounts and drive faster wallet-share expansion.

• Next earnings is imminent (Q1 2026 on April 30, 2026). Any beat/raise + continued strong NRR can change the narrative quickly.

• Dilution is coming down: dilution used to be much higher (~15% in FY2024), then management guided 4%–6% for FY2025 and is targeting 3%–4% net dilution in FY2026.

  • In the latest disclosures, management said they ended 2025 at the low end of their range with total net dilution of 4.3% (or 2.2% excluding Marigold) and improved stock-based comp from 19% of revenue in 2024 to 14% in 2025.

• Valuation reset: the stock still trades like a “question mark”, not like a compounding 30% grower — which is exactly why the upside (and the risk) can be asymmetric.

Business Ontology Framework by L2C

The Business Ontology is a framework I built after tearing apart several tech companies from the ground up—breaking them down to their basic parts and piecing together a real thesis on what drives them. Think of it as a map of a company’s soul. It’s a tight set of core indicators—tailored to each business—that show if it’s heading the right way, no matter what the stock price says. These aren’t your basic stats like P/E ratios or revenue bumps you grab from Yahoo Finance. They’re deeper, sharper, and linked straight to the thesis I’ve cooked up on how the company makes value and fights in its market.

The framework boils down to two big pieces:

1. The Business Ontology—This checks if the company’s worth buying into or hanging onto.

2. The 4D Valuation Model—This gives you a 4D roadmap to guide your own calls on how much to put in (allocation).

Now, let’s dive into it.

Read more