An in-depth discussion of Meta Platforms; Charity auction for lunch with Warren Buffett

Meta Platforms (META) has long been one of my favorite stocks...

I named it – along with Alphabet (GOOGL), Amazon (AMZN), and Berkshire Hathaway (BRK-B) – as a core holding in my old firm Empire Financial Research's first newsletter on April 17, 2019. Since then, through Friday's close, they're up 241%, 546%, 192%, and 126%, respectively – an average of 276%, nearly double the 155% gain for the S&P 500 Index.

Today, I want to share an in-depth discussion I had with two of my friends about Meta. Our conversation was triggered by this Wall Street Journal article that laid out the bear case for the stock, which says its discounted price is more of a warning than an opportunity:

Meta's stock has languished recently as concern grows over its spending on AI, leaving it unusually cheap. The company is trading near its lowest price-to-earnings multiple in three years. At around 18 times forward earnings, its valuation also represents a discount to other big-tech companies...

Looming large in the longer term is user growth. The company has a huge base of users – more than 3.5 billion people used Meta's Facebook, Instagram, WhatsApp and Messenger daily in the first quarter. Growth, though, isn't impressive: Users rose 4% year-over-year in the quarter but declined sequentially, something that had never happened since it began reporting its active-user metric in 2019.

Meta executives believe their AI-infused ad strategy gives them plenty of runway to grow revenue further. Without better user growth, though, it is only a matter of time before Meta reaches a natural limit to what AI can do for the ad business it relies on.

As both friends prefer to remain anonymous, I'll refer to them as "Friend A" and "Friend B" to avoid confusion.

Friend A is a highly successful private investor who still owns Meta shares he bought at less than $1. So he knows the company better than anyone. (He gave his thoughts on my analysis of Meta's latest earnings release on April 30.)

Here was his response to the WSJ article:

I think this is more clickbait than substantial. It doesn't matter that users are declining at the edge because these users are not monetized and will not impact revenue at all.

Furthermore, they have several other businesses, like Facebook Marketplace, WhatsApp, and Click-to-Search. So it's just a case of someone saying, "Oh, hey, I know nothing about this business, I'm just going to say they're just an ad business." But their ads are multifaceted, and it turns out they're the best advertising vehicle in the world. They actually have more advertising revenue than Google, so that's something to think about.

Regarding AI, he continued:

Overall, look, there's a risk (we've gone through this) that they won't get AI to work. However, this is one of the most profitable, high-margin businesses that's ever existed, and it can always return to its traditional roots. I'm not concerned.

Have you actually tried Meta AI lately? It's as good as or better than Gemini. It's actually amazing. At the end of this month, they're about to launch their new Avocado model, and if it follows this trend, there are going to be some surprises.

And to add to this, the writer of the WSJ article obviously doesn't believe that AI has any value other than selling capacity, which is what Google is doing. Facebook is going to have full-on AI models, and they are basically valuing the possibility of any other company or entity using it at zero.

And he concluded:

Once again, you either believe that [CEO Mark] Zuckerberg is going to do what he's done many times before and continue to grow this company in the direction of products that make sense and are fast following, or you think he's just going to get stuck where he is.

In the meantime, let the stock go down a lot more, which is great: I'll buy more of it. That is what I've been doing, and it's always worked.

That said, it can be a frustrating stock to own because it's never broken out of the love-hate cycle, and yet it's a great company. I think it's just because people don't understand it.

Friend B, a European investor, wrote to me that he wasn't persuaded by Friend A's bull case on Meta, which I shared in my May 1 e-mail. So I forwarded his e-mail to Friend A...

Their conversation got into the weeds – and, I'll admit, a bit of it goes over my head. But I'm sharing their entire exchange because it's a good example of the type of in-depth analysis and debate that leads to differentiated views on companies and stocks. And that's the key to successful investments.

Friend B wrote:

Your friend's read on Meta is solid, and the 33%/19%/12% decomposition on 4% user growth is genuinely hard to argue with. Two points worth passing back to him.

First, Meta's revenues are around 98% from ads. And Llama isn't sold per token, so the DeepSeek V4 price shock hits OpenAI and Anthropic, not Meta. That's a real structural shield.

Second – and this is where I'd push him to dig further – Meta depreciates servers over 5.5 years (extended from four to five in 2025, cutting depreciation and amortization by $2.3 billion in the first nine months alone).

Nvidia's CEO Jensen Huang said on the Dwarkesh podcast: "None of the bottlenecks last longer than two, three years... Hopper to Blackwell, 30 times to 50 times efficiency." The CEO of the supplier of 60% of data-center value is telling you the useful life cannot be 5.5 years. I calculated it at sub-three years (bottom-up component aggregation).

That gap matters because of what the new capital expenditures ("capex") is actually funding. Fellow analysts at Research Affiliates estimate that only one-third of hyperscaler capex is expansion, and two-thirds is silent replacement of gear installed less than three years ago (a full-stack refresh runs around 60% of the original data-center asset value). Meta discloses total capex. I'm looking for the maintenance/growth split but have not found it. It's a gap worth pressing on if you own Meta.

Doron Nissim's monograph is blunt on the mechanism: "By understanding amortization, firms can permanently increase perceived recurring income" – because the eventual impairment gets stripped out as unusual.

Friend A replied first by conceding some points:

Your friend is making a good argument, and I want to engage with it directly rather than defend reported numbers that don't deserve full defense.

He's right on the Llama point, and I'd concede it cleanly – Meta running open-weights models internally for ad ranking means the per-token price compression hits OpenAI and Anthropic on revenue while hitting Meta only on cost-of-goods. That actually strengthens the bull case rather than weakening it, because model commoditization works in Meta's favor on both sides.

On depreciation, he's largely right on the mechanics. One factual note: Per the WSJ on April 29, Meta's new extension is from six to seven years on non-AI servers specifically, driven by a memory-chip supply deficit, with AI servers continuing on the previously extended 5.5-year schedule. The headline is narrower than a blanket seven-year extension across the fleet.

But that nuance doesn't dispose of his critique – the cumulative effect of the depreciation extensions is real (four to 4.5, five to 5.5, and now six to seven across various asset classes). And Nissim's mechanism on how the gap eventually surfaces through non-recurring impairments is the right way to think about it.

He continued by rebutting several points, starting with depreciation:

Here's where I'd push back, and it's the part that matters most.

The depreciation critique on Meta isn't new. Hedge-fund manager Mohnish Pabrai flagged it in 2023, the earnings-quality analysts have been writing about it consistently since 2022, and Michael Burry's version is the loudest iteration of an argument that's been around for at least four years.

During those four years, Meta extended depreciation repeatedly, raised capex from $30 billion to $135 billion, and went from $88 to $630 per share. The bears who shorted or avoided the stock based on this argument missed one of the great runs in mega-cap history.

Why were they wrong? Because they correctly identified an accounting issue but incorrectly extrapolated it into a stock thesis. The accounting issue was real. It just didn't matter, because revenue growth substantially exceeded what was needed to outrun the depreciation drag. Meta's revenue more than doubled from 2020 to 2025. The depreciation tail was real, but the gross profit on the incremental revenue more than covered it.

This reframes the question your friend is implicitly asking. The depreciation issue isn't really the issue – it's a symptom. The actual question is whether AI capex generates sufficient revenue to make the depreciation question moot, the way the 2022 machine-learning ("ML") capex did when Apple's App Tracking Transparency ("ATT") looked existential.

If revenue grows fast enough, the depreciation is fine. If it doesn't, the depreciation becomes the visible mechanism by which failure shows up in the financials. But the depreciation isn't the cause – it's the canary in the coal mine.

So the bear case finally being right requires revenue growth to disappoint. The depreciation argument doesn't stand on its own. It stands or falls on the revenue thesis underneath it. And that's where I'm genuinely comfortable with the bull case.

He then addressed Meta's financials and strategies in depth:

The 33%/19%/12% decomposition is auction-level reality on a mature business and unaffected by depreciation policy. That's revenue, not earnings – real cash flowing through the auction at accelerating rates. The only plausible explanation is that AI/ML investment is already converting in the existing ad business, the same way it converted post-ATT. That's not a forward-looking promise – it's the current quarter.

The structural position on consumer AI agents is the part that gives me real conviction. Meta has 3.56 billion daily users, 10 million advertisers plugged into existing targeting and measurement infrastructure, and Click-to-WhatsApp already proving the agent-mediated commerce template works. The revenue model isn't speculative – it's the existing ad model extended to a new interaction surface that Meta already owns.

The fast-follower track record is essentially perfect. Stories was Snapchat. Reels was TikTok. Threads was Twitter. Marketplace was Craigslist. Meta has never been a consumer behavior pioneer and has never needed to be. Some company will validate which AI agent behaviors stick, and Meta will absorb them at superior scale. That's a much easier problem than being the pioneer, and it's the position Meta has executed consistently for over a decade.

Zuckerberg's capital allocation pattern is the final piece. He validates ML/AI solutions at smaller cohort scale, then scales capex once return on investment is proven. The metaverse is the one significant exception, and notably the part he has rationalized relative to AI. The 2026 capex raise should be read as scaling something that's already working, not as a speculative bet. It's the same shape as 2022, which the market also misread as value destruction and turned out to be the buying opportunity of the cycle.

There's one more piece that I think actually clinches the argument. Even if you accept the bear case in full – that the AI capex doesn't generate the expected returns and Zuckerberg eventually has to acknowledge it – the downside scenario isn't catastrophic. It's a Meta that pulls back capex the way Zuckerberg pulled back on Reality Labs in 2023, and reverts to being a $250 billion to $300 billion revenue ad business with 40%-plus operating margins generating $80 billion to $100 billion-plus of free cash flow annually. That's $30 to $40 of normalized earnings per share at current share count, which supports the current stock price comfortably even with no AI upside priced in.

The metaverse precedent is exactly the right reference here. Zuckerberg committed publicly, renamed the company, took years of market pain, and then when the timeline became clear, he pivoted hard. The "Year of Efficiency" reset the cost base, Reality Labs was rationalized, and the underlying ad business carried the company through without missing a beat.

The base rate on Zuckerberg pulling back from a bet that isn't generating returns within a reasonable time frame is genuinely high based on observed behavior. He's not stubborn about big bets – he's demonstrated the ability to acknowledge when something isn't working at the pace expected and reallocate accordingly.

That's the part that makes the asymmetry work. The bear case requires Zuckerberg to keep spending into a failing return profile for years, refusing to pull back even when the returns clearly aren't materializing. That's a much narrower window than just "AI capex doesn't work." If AI capex doesn't work and Zuckerberg recognizes it within two to three years, you still have an extraordinarily profitable ad business at the current scale that justifies the current price without any AI optionality.

The depreciation tail only catches up if Meta keeps spending into a bad return profile. If they stop, the existing business absorbs the in-flight depreciation while continuing to generate exceptional profits. That eliminates the catastrophic downside scenario from the bear case entirely.

He concluded with this "net synthesis":

Your friend has identified a real weakness in reported earnings construction, and a long case for Meta here needs to underwrite the investment on revenue acceleration and free cash flow rather than on GAAP operating margin. The depreciation flattering is real, and the gap will eventually close through impairment charges.

But the depreciation isn't the thesis – revenue is. And on revenue, the underlying business is at its strongest point in the company's history, the strategic position on consumer AI is structurally advantaged, and Zuckerberg's track record on both capex deployment and capex withdrawal when bets don't pan out is exceptional.

The bears have been wrong for four years because they were looking at the wrong variable. The right variable is revenue growth, and on that variable the evidence is strongly in Meta's favor – fast-follower position, distribution moat, accelerating ad business, and a validated capital-allocation pattern. If revenue grows, the depreciation question becomes irrelevant.

And if it doesn't, Zuckerberg pulls back the way he did with the metaverse, and the existing ad business absorbs the spend. The asymmetry is heavily in favor of the long case, and the case for revenue actually growing is, in my view, as strong as it's been at any point in the company's history.

Tell your friend thanks – this was a sharp note and exactly the kind of pushback that sharpens the analytical work. The accounting point lands – the conclusion that should follow from it is "press Meta on revenue durability," not "the stock breaks because of depreciation." On revenue durability, I'm comfortable with the bull case.

Friend B responded:

Big thank you, Whitney, for doing all of this. I'd be happy to meet your friend if he comes over to Europe.

I have been cautious not to risk predictions on Meta's stock. I expect they will experience some forms of turbulence at some point, but they are among those who can sail through the impairment storm and be fine on the other side, unlike Oracle and others. I agree that open weights rather improve the profits for Meta.

And there is a significant potential profitability improvement for Meta if they could move a larger share of the AI processing away from the Cloud and onto the users' devices, on the Edge platform. Alphabet, Apple, and Microsoft are going for this, because it's easier for them (leveraging Android, iOS, and Windows 11, respectively), and it should be more challenging for Meta. However, if Meta finds a way to do it in their apps, then it would crush the cost base on AI-related operations in the future.

How far Meta is going on the migration from Cloud to the Edge in the apps is an open question for me. Most likely five to 10 years out, they will have hybrid Cloud/Edge operations. The more they can run on the Edge, the less cost they bear because it will run on the users' own devices instead of on expensive and fast-depreciating assets.

We could even question if it would be worth cutting a slice of infrastructure capex to allocate on expanding and accelerating an Edge strategy in the Meta apps, because that can make a share of recurring infrastructure costs truly disappear. It's a real improvement track because people are already designing and shipping dedicated semiconductor chips for this.

Thank you, my friends, for sharing such an insightful and respectful conversation!

After reading through their exchange and giving it some thought, I like Meta's stock even more.

Best regards,

Whitney

P.S. I welcome your feedback – send me an e-mail by clicking here.

P.P.S. The annual charity auction for lunch with Warren Buffett is now live! This year, basketball superstar Stephen Curry and his wife Ayesha will also join the lunch.

Proceeds will benefit the Bay Area's Glide Foundation and Eat. Learn. Play., and bidding closes this Thursday, May 14 at 10:30 p.m. Eastern time.

One year, the winning bid was $19 million! I'll be curious to see what it goes for this year... Here's a link with more information: