I first pitched Meta in January, and I didn’t expect to be back this soon. But with everything that goes on in the world of AI and hyperscalers, nine months feels like five years. Meta has not only announced a new AI, but also one of the most capable personal AI agents out there, ambitions for a cloud business, raised capex guidance several times, and settled the biggest lawsuit in its history in record time.

The market didn’t like the capex raises and the lawsuits, so Meta traded about 25% below all-time highs when I pitched it in January. And although it recovered strongly in the last few weeks, it remains one of the cheapest of the Mag7 companies. More importantly, it’s the company with the most optionality.

So there’s a lot for us to get into!

— Daniel

I'm actually writing this from Shawn's living room, having traveled down for a couple of days from New York City, where we met with many members of our Mastermind Community at our Intrinsic Value Conference.

It was a fantastic weekend, and I immensely enjoyed meeting members in person. We have these in-person meet-ups at least twice a year – one of them in Omaha for Berkshire Weekend.

Online, we host weekly calls with Kyle, Shawn, and me, along with outside fund managers, well-known authors, acclaimed individual investors, and our members themselves.

If you want to build relationships with other value investors, exchange and test ideas, and discuss what's happening in the markets, I'm confident the Mastermind Community would be a great fit for you.

You can apply through the link below. From the applications we choose, we'll set up onboarding calls hosted by Shawn and me.

A huge topic for Meta right now is capex and Zuckerberg's enterprise and AI vision. The market wasn’t really buying it, and there are some good reasons for that. Zuckerberg hasn't been very successful with his latest ventures outside the family of apps (Facebook, Instagram, and WhatsApp). So, besides talking through what changed in the last few months for Meta and how we think about it, I also wanna take some time today and try to explain why I believe Zuckerberg is so focused on expanding the business anyway.

After all, he could just be happy with having the biggest and most profitable ads business in the world and print money at a scale that rivals most countries on this planet. I actually looked it up; Meta generates more revenue than the GDP of about 140 countries. That’s the size of business we are talking about.

To understand some of Zuck's latest ambitions, we need to go back to the late 2000s, when he first tried to turn Facebook into a platform. Back then, Facebook was a typical desktop app, and besides the social media business, it had many third-party apps, payments, and games. At the 2012 IPO, games and apps made up around 20% of the overall business.

Then the world moved to mobile, and Apple started making the rules by turning its App Store into the primary platform on iPhones. One of those rules was that you don't get to run your own platform within Apple's walls. So Facebook doubled down on the ads business model instead.

In hindsight, this was the best thing that could've happened to Meta. And yet, Zuckerberg didn't seem to be too happy with that position. And the privacy update Apple pushed through years later, which cost Meta a whopping $10 billion, showed his concerns weren't unreasonable.

His second attempt, after the original games-and-apps phase, started with buying Oculus for $2 billion in 2014. It took years until we were shown why Zuckerberg wanted to own Oculus, when he started building the Metaverse. This might have been the first time the market saw how serious he is about this. He spent about $100 billion on a complete flop that was not only technically laughable, but also had no customers. As it turned out, nobody needed the Metaverse.

And here we are again, in 2026, when AI seems like the right opportunity for Zuckerberg to start attempt number three at building his own platform. We'll get into all the details of what this attempt actually looks like and how likely it is to succeed. Upfront, though, I think Meta is in a much better starting position this time. On the other hand, $100 billion for the Metaverse seems laughable compared to today’s AI capex. This is by far the biggest bet Meta has ever made.

Before we get into the weeds of the AI bet, let us quickly address the recent lawsuit against Meta. It was the largest in Meta's history and one of the largest in American history.

It all started in New Mexico, where Meta was sued in 2023 for endangering children. Meta was found liable at trial earlier this year and ordered to pay $375 million. In a second phase in August, the judge ruled that Meta had created a "public nuisance," a concept historically used against polluters and opioid distributors, and ordered another $570 million into an abatement fund. Meta is appealing, but that's where we stand right now.

The billion dollars is not the problem. Meta earns that in less than a week. The problem is that Meta also had to change its product in New Mexico. Like-counts are hidden for users under 18 unless a parent approves, push notifications to minors are paused between 10 p.m. and 7 a.m., and usage is capped at 90 hours a month. The state's attorney general recommended this as a blueprint for other states and countries when going after Meta.

And he was heard. In Oakland, 29 states went to court against Meta about two months ago. The headline number for potential fines was $1.4 trillion – pretty much exactly Meta's market cap at the time. Now, to be fair, Meta calculated this number itself. They came up with it by treating every under-13 account as its own violation of COPPA, the federal children's privacy law, and multiplying by the maximum penalty.

What would've actually been argued in court was closer to $200 billion, and based on what U.S. courts have historically awarded, I would have expected $20 to $30 billion if Meta had lost. But it never went that far.

In a surprise development, the states settled early with Meta, agreeing that Meta would apply similar changes to those agreed upon in New Mexico and pay about $12 billion in fines. That number could rise to about $17-18 billion if competitors like YouTube and TikTok agree to similar changes and contribute payments of their own.

This might not be the end of it, since hundreds of lawsuits are still pending, but it was still a success for Meta. The fine and the changes to the product are manageable, and by calling for competitors to join "the battle to protect kids," Meta even has some moral high ground. Very much the opposite of where it stood at the beginning of the trial.

And even if we assume more lawsuits follow and Meta ends up paying something like $2 billion a year for the next decade, that's negligible if the fear of much harsher regulation is off the table.

Beyond that, the sentiment around tech and especially social media has changed so much in the last 10 years that Meta is better off working on its image to get out of the crosshairs of regulators and courts eventually.

Even if it means leaving some money on the table today. And the best way to do that isn't through marketing campaigns, but by making sure its product becomes more of a win-win for society. I don't believe social media can be successful only when its content is divisive and polarizing.

Meta's stock has recovered significantly since we recorded the episode. Unfortunately, we had to record it a bit early so we don't run out of episodes for you while we're hosting our Intrinsic Value Conference in New York. When I covered Meta, the stock was in the low-500s – down 25%.

If you looked at the front page of the New York Times, the reasons for that would've been the lawsuits and the AI capex. But the stock didn't recover meaningfully after the settlement, and AI capex was also a thing when the stock traded at over $700 earlier this year.

I believe the reason was the lack of visible payoff through AI. It's funny because it was the exact opposite at the beginning of 2026. Back then, Meta was supposed to be the only Mag7 with an immediate AI payoff. The reason was the ad engine. Its ranking and targeting models were getting better, and in Q1, ad impressions and ad prices went up together.

That's quite rare, given how ad pricing works on Meta's apps. Meta uses a so-called auctioning model. When Meta opens up new ad inventory, like Reels in 2022, supply jumps, and that usually means prices fall. That looks scary but has usually been the best moment in the cycle, because cheap inventory pulls in advertisers, and a few quarters later, both lines grow again.

Impressions grow either because Meta shows more ads per minute or because you spend more minutes in the app. The second is clearly healthier, and according to Meta, time spent on Instagram and Facebook grew about 10%. On price, higher isn't automatically bad for advertisers. If a lead costs 10% more and spends 10% more, nothing has changed for them. That's what better AI targeting is supposed to deliver.

That’s why, in Q1, prices and impressions grew at the same time. In Q2, however, that trend did not persist, which made the market question how far AI can take the ads engine and whether the hundreds of billions of dollars in capex would be worth it.

At the time of writing, though, the market has found something else to focus on, which is why the stock is back up to the mid-700s. Meta's newly released Muse personal agent has been No. 1 on the App Store for the last few days and is growing at a pace that only ChatGPT saw after its release.

Perhaps some of you remember me talking about agentic commerce last year in my PayPal episode. PayPal didn't age too well, but the agentic vision might be coming now. It's interesting because just a few weeks ago, I told members of our Mastermind community that I seemed too bullish on it, since there had been no progress in an entire year. Well, now we are much closer.

It opens up a lot of interesting questions. Amazon, for example, has already decided to refuse access to Muse. Not a major surprise after Amazon had similar problems with Perplexity before. But many other companies immediately started partnerships with Muse, including Shopify. I wouldn't be surprised if this is a major moment for agentic capabilities used by the "average" consumer, not tech geeks who bought a Mac Mini to play with OpenClaw (not that there’s anything wrong with that).

All that said, I'm not sure how much staying power the Muse agent has. It is incredible, don't get me wrong. And as someone who bought Meta stock last month near the annual bottom, I'm more than happy about the jump in the share price. But we have seen how fast things change in the AI world. Who knows if and when OpenAI, Anthropic, Google, or someone else will come up with a similar agent.

Speaking of agents that could revolutionize how Meta works, this could be a major thing for WhatsApp too. When we first covered Meta, we talked about the business account opportunity and how it is primarily feasible in countries with low labor costs, because you have to pay someone to respond to messages all day, which only makes sense in countries like India or some other Southeast Asian countries.

If, however, you can have an AI agent do that work, you don't need to pay anyone anymore, and the economics get much more attractive even in countries like Germany or the U.S. That still doesn't solve my personal hesitations, since WhatsApp is for personal chats to me, and I don't want business accounts in between them, but I might be underestimating how quickly I could adjust to that.

After all, it's working in other parts of the world. Zuckerberg pointed out that Thailand and Vietnam rank around sixth and seventh among Meta's countries by revenue, while by GDP, they're somewhere in the 30s. Apparently, about 2% of Thailand's GDP flows through commerce on Meta's messaging apps. Which is an incredible number — 2% of Germany's GDP would mean we are talking about more than $100 billion a year.

Another area AI might revolutionize is Meta's hardware business. I recently watched Sam Altman's interview on David Senra's show, where he said that he expected AI to have had a much more disruptive impact already and that the iPhone moment hasn't yet arrived for AI. I think Zuckerberg believes AR glasses are that moment.

Meta is certainly at the forefront of the AR glasses race, despite some embarrassing product announcements in the past. Its wristband appears to be the way to go, while Apple's Vision Pro approach failed.

And AI should be the next step – control the glasses via voice, and have your Muse agent integrated into your daily life as closely as possible. Despite seeing that vision (no pun intended) and being biased toward the product, since I wear glasses every second I'm awake, I still wonder what the incremental value would be compared to an iPhone.

I don't think people will be willing to pay $1,000+ to have a virtual map integrated into their field of view or get some other gimmicks directly integrated. It might just be that the phone is the perfect mix between being quick to grab, capable, but not intrusive or overkill. The just-presented Muse Charm didn’t blow me away either. I’m exaggerating, but it feels like a phone without all the capabilities of a phone.

But I'm very open to being wrong here and seeing how the charm and AR glasses take over the world.

What I'm slightly worried about, given Meta’s history of unsuccessfully trying to expand its business, is how much money will be invested in this project in the meantime. Also, while patents should protect Meta's technology, Apple seems best positioned to sell hardware like that at scale. They are never the first, but they tend to be the most dominant once a market exists.

There's another part of AI innovation that gets me excited about Meta, though. And it goes back to the ads business. Imagine Instagram with an almost infinite inventory of ads/opportunities to sell. Let's say you see a photo of a friend or a celebrity wearing a sweater you like. Today, you would need to find out where that sweater is from, search it up, and order it. And Meta didn't partake in any of that transaction.

Now imagine asking the AI in the app where it's from. It links you to the shop, you buy it in two taps, and the seller pays Meta for the conversion. That would be an entirely new way to use posts as an advertising tool – direct access to the consumer. Brands could also pay Meta to appear in these AI responses.

Since this isn't ad space in the traditional sense, it shouldn't materially affect ad prices either. In theory, the same works through glasses, where anything you look at is one question away from being a product. I just visited Washington, DC, with Shawn, and as we walked through the different neighborhoods, we checked every second house we saw on Zillow. What if we no longer had to grab our phones for that, and it just popped up? I know, it sounds a bit dystopian to have it pop up right in your field of view, but you can’t deny that it’s handy.

We've spent a lot of time talking about the capex and AI investments, so let's actually get into the numbers. Capex will be $130 to $145 billion this year, and analysts expect close to $200 billion for 2027. In January, Zuckerberg announced that Meta wants to build tens of gigawatts of data center capacity this decade, where a single gigawatt is roughly the output of a nuclear reactor.

He even brought in Dina Powell McCormick, a former government official, as President to work on government partnerships. When your capex plan needs its own diplomat, things are getting serious.

And Meta likely needs her. A member of our Mastermind Community who used to work in the White House and now runs a consulting firm that companies like Meta turn to for political connections told me that Meta isn't very good at understanding how Washington works. That's not particularly a surprise given its history with the government and the courts.

But let's get to what the enterprise vision actually is. Zuckerberg described the enterprise opportunity as the sum of selling compute, API services, productivity services, and business agents. That sounds a bit vague, and it's not necessarily consistent with what Zuckerberg said in the last few years. He repeatedly argued against a cloud business because a GPU earned more inside Meta's own products than rented out. But I guess times have changed.

With the recent spending, Meta needs a way out if it overspends. And given the current imbalance between compute supply and demand, selling excess compute is a sort of margin of safety.

The other reading is that Meta is serious about its cloud ambitions. It is building data centers, it has custom AI chips, some of the best engineering and coding talent, and it now has a high-quality LLM that comes close to the frontier labs in terms of capabilities. I suspect it won't start with the Fortune 500 companies, but with small businesses that already operate on Facebook, WhatsApp, and Instagram. Selling them agents inside apps they use every day is a go-to-market strategy none of the big clouds can copy.

Despite all this optionality, what gives me the most pause with Meta is that it can't just roll back investments this time if things don't work out. When the Metaverse failed, Zuckerberg declared a "year of efficiency" and cut costs. That exit doesn't exist this time. Meta has close to $700 billion in off-balance-sheet commitments, up 800%(!) year over year.

A good example of why these investments can't simply be rolled back is how the data center financing works. Take Louisiana. Meta has partnered with the leading alternative asset management firm Blue Owl on that one. They set up a separate company whose only job is to own that data center. Blue Owl reportedly holds around 80%. That company borrows the construction money from investors like BlackRock and PIMCO, roughly $27 billion in bonds, and Meta signs a long-term lease and guarantees a minimum asset value at the end.

This way, Meta looks less indebted, and the risk is somewhat shared with financing partners. But the most important reason, I believe, is that it opens up a different pool of money. Asset managers, insurance companies, and private credit funds are desperate for long-dated, bond-like exposure to AI infrastructure. This is how pipelines and power plants have been funded for decades, which tells you a lot about how the Mag7 are now financed.

It's very hard to say what will work for Meta and what won't. And the sentiment changes all the time. With Muse, the entire AI narrative, and thus the story of the entire company, has changed materially. I wouldn't be surprised if the hype around Meta's agent turns out to be less sustainable than some people think. On the other hand, as a shareholder, I'm very excited about the potential of agents and Meta's position in the market.

So where does all of that leave us? The stock has rallied about 30% since we covered it on the show. I like that for my personal portfolio, but it also means that the margin of safety has come down. That said, and I really don't want to overstate the Muse agent's success, Meta has made a big step in the right direction with that launch. Perhaps it will get commoditized quickly, perhaps it's a one-hit wonder, and perhaps agents won't spread much further into the mainstream than they have right now. Nevertheless, Meta has proven its AI capabilities are now on par with the major labs – or at least much closer than in prior years.

But this price increase also makes the model a bit less attractive. Now, you might argue that Meta has opened up a major new business opportunity through Muse, but that kind of optionality is hard to put into a spreadsheet. I do not yet feel comfortable underwriting major revenue or profit growth because of Muse. Zuckerberg doesn't plan to charge for it, but will eventually make money by taking a fee from sales generated through Muse.

That implies that Muse actually stays on top of the agent game and that the business model turns out to be successful. That, in turn, also depends on other companies accepting Muse and connecting to it. We see most companies doing that, but companies that dominate their niche – like Amazon – have an incentive to try and stop Muse from being successful. Point being, I like the optionality, but I don't feel comfortable underwriting more growth for it at this point.

All that to say, take the model with a grain of salt. It's certainly on the conservative side, given the many upside options Meta has up its sleeve. In my base case, top-line growth slows to the high teens over the next few years. Meta could well keep growing at 20% a year, but after the recent slowdown – if you can even call it that – I feel more comfortable a notch below that.

The biggest impact will be on the margins. I have the operating margin go as low as 32% in 2027, which is 10 percentage points lower than in 2025, due to depreciation from the AI investments coming through. But I do have the margin recover to 38% by the end of the decade.

Capex might be the most important item in this model. I have capex for this year between $135 and $140 billion, but about $200 billion for each of the next three years. With the recent success of Muse, it gets more and more likely that this is the level we will see for the foreseeable future.

The consequence is that Meta generates basically no free cash flow until the end of 2028. That said, with the margin picking up again at the end of the decade, Meta could still comfortably generate about $100 billion of FCF in 2030. Whether that will actually be the case in 2030 is up for debate, but the earnings power is definitely there.

For the valuation, I've chosen to go with three versions this time, just to give you some context on how you can think about valuation. We've gotten many questions over time about when to use which multiple and what it implies.

So the first approach is a classic, but perhaps not the most suitable for Meta today – you just have a multiple on 2030 FCF. The problem is that in 2030, Meta is still spending around $190 billion on capex in my base case, so free cash flow is still depressed, and the multiple would essentially punish that "for eternity." There is no "after build-out" value. You could go with this approach if you think Meta will remain a business that spends hundreds of billions on capex every year and never returns to a less capital-intensive model. In that case, the model gives me a fair value of about $610 per share.

The second one fixes that by assuming that, by 2030, capex has come down to a "normal" level – roughly 1.5 times depreciation, which is what a mature company needs to replace and modestly grow its assets. That gets me to about $760. The depreciation factor does a lot of heavy lifting here, though, and it has varied quite a lot in the past.

And then last but not least, you can put a P/E multiple on 2030 earnings. By then, depreciation from the buildout runs fully through the income statement, so earnings already carry the bill, so there’s no need to assume anything about capex normalizing. That’s the version I personally prefer for Meta. Before a margin of safety discount, this gives me a price per share of $840.

If you add a 10% margin of safety to this, the expected return would be in the high single digits, which is not too inspiring. That said, there's the optionality of Muse, the cloud business, agents in WhatsApp, and so forth that isn't captured by this model. I will certainly hold on to my personal stake in Meta. At the same time, Shawn and Kyle aren't big fans of the stock, given their views on the company's ethics as well as its capital allocation decisions. Add to that that Meta no longer seems massively undervalued, and we decided not to add it to the Intrinsic Value Portfolio.

To listen to our episode on Meta, check out our podcast here.

Updates on our Intrinsic Value Portfolio below 👇

To discuss stocks daily with Shawn, Kyle, Daniel, and all Intrinsic Value Mastermind Members, apply here.

"Building is my love language”

— Mark Zuckerberg’s T-Shirt Print at this Week’s Meta Connect

📺 WATCH: The Meta Connect 2026 Keynote

🎧 LISTEN: William Green’s Interview with Rob Vinall on Richer, Wiser, Happier

📖 READ: The Wall Street Journal on Meta’s New AI Agent

You can also read our archive of past Intrinsic Value breakdowns, in case you’ve missed any, here — we’ve covered companies ranging from Alphabet to FICO, Transdigm, Lululemon, PayPal, DoorDash, Crocs, LVMH, Uber, and more!

Enjoy reading this newsletter? Forward it to a friend.

Was this newsletter forwarded to you? Sign up here.

Apply to join our Intrinsic Value Mastermind Community of investors.

Learn how to join us in New York City for the 2026 Intrinsic Value Conference!

Read our full archive of Intrinsic Value Breakdowns here.

Keep an eye on your inbox for our newsletters on Sundays. If you have any feedback for us, simply respond to this email or message [email protected].