Article · 2026-08-03 · 5 moments

Meta Earnings, Meta’s Timing Problems, The Financial Tail

Meta's earnings were a bit disappointing; future promises about AI products were more disconcerting. ✦ AI generated

GUEST Ben Thompson GUEST Mark Zuckerberg

Mechanism◆

Meta has a massive timing mismatch: it is double-paying for infrastructure — renting compute from third parties while building its own data centers — without a clear path to monetization, and its improving monetization story has lost luster.

Meta has committed nearly $700 billion in future AI infrastructure spending, is renting compute from third parties while also building its own data centers, and faces a core timing problem: it spent frantically to catch up in AI, but now must pay for both rented and owned capacity without a cloud business to monetize it.

transcript

Ben Thompson: Meta Platforms Inc. said it has already committed almost $700 billion in future spending, through long- and short-term agreements, related to artificial intelligence data centers, cloud computing and more. Meta has $349.3 billion of non-cancelable contractual commitments, mostly related to third-party cloud deals, servers and network infrastructure, it said in a regulatory filing Thursday. That is a conservative estimate, because for agreements with variable terms, 'we do not estimate the total obligation beyond minimum quantities,' Meta said. The company also has $347 billion in commitments for leases that have not yet started, and so are not yet reflected on its balance sheet. That includes $68 billion added in July alone, with payments starting in 2027 and 2028. The costs are in addition to active leases and consist of data centers, colocations and 'certain network infrastructure.' What confused several investors on the call was why Meta was also renting compute from third parties; the issue the company faces is one of timing. Last summer Zuckerberg realized — correctly, in my opinion — that Meta risked falling out of the AI race, and not only spent heavily on AI talent, but also had to scramble to get more compute for training. The problem is that actually building data centers takes time — multiple years — which Meta didn't have; thus the renting. Ultimately, however, Meta wants to own its own data centers, not just for training but also for inference, which means they need to spend to build now for years from now. This timing mismatch is the biggest issue Meta has when it comes to investors. The company right now is basically double-paying for infrastructure without a clear path to monetization (i.e. no public cloud business) and, to make matters worse, its 'improving monetization' story lost a bit of its luster.

extends · 1provides context · 1

Prediction

Meta has a third timing problem: Anthropic and OpenAI have a structural cost advantage in inference that is only increasing, and while Meta is investing in catching up, the frontier labs are already running ahead on both cost and data flywheels.

Thompson argues that Anthropic and OpenAI are accelerating and have a structural cost advantage in inference that is increasing, while Meta is making the kind of investments the frontier labs made in the past, and its data flywheel argument about billions of users echoes Microsoft's irrelevant Windows numbers argument from 15 years ago.

transcript

Ben Thompson: That leads to a third timing problem: right now Anthropic and OpenAI are accelerating and, as I argued a couple of weeks ago, almost certainly have a structural cost advantage in terms of inference that is only increasing. Meta is trying to sell investors on the future while making investments that the frontier labs made in the past. Moreover, those past investments aren't just paying off in terms of margin, but also data: Zuckerberg talked about the importance of data flywheels in the context of model improvements and the company's proven ability to scale out products to billions of people, but that had a whiff of Microsoft 15 years ago talking about a billion Windows computers in the face of the burgeoning smartphone segment. Trajectory and relevance matter more than sheer volume, and Meta does better on the latter than the former.

extends · 1provides context · 1supports · 1

Claim◆

Meta must build its own frontier AI models because relying on open-source models is not viable — they are not as strong as frontier models, and depending on other companies' actions is risky for a company like Meta.

Zuckerberg argued that Meta cannot rely on open-weight models because they are not as strong as frontier models, and depending on other companies' policies is too risky for a full-stack technology company like Meta that needs sovereignty over its own AI stack.

transcript

Mark Zuckerberg: I can take the open source question. Let's see. So basically the question is, do we think that because there are some open weight models that we can just rely on those. I mean right now the open source models are not as strong as the frontier models. So no is the basic answer. And then there's also just always the perpetual both policy debate and question around other companies' actions and whether that's actually a thing that a company like Meta can rely on. And I think that that's very tricky. So I think on both fronts, we believe we're going to be able to do better work, and we think that there's some risk in that reliance, I don't believe that that is the right thing to do. I think that we're a company that — if you look at Meta from — take a step back on this. A lot of people view the surface layer of we build some social media apps and we have an ad business. We are really a full stack technology company. We built our own data centers, our own infrastructure, our own chips, our own low-level software. When we got started — like my background in engineering, like I wrote a lot of the systems code. A lot of the reason why Facebook worked was because it actually — it just worked, right? Like it literally worked when other social networks did not work fast and efficiently. And I think we just have the ability to build things that can be more personalized, more optimized, more efficient. Some qualitative experiences are just not even possible for others to build because we go all the way down the stack. And it just seems to me pretty clear that having kind of sovereignty over building your own models is going to be an important part of that stack going forward which is why it is important for Meta.

Data

Meta's core ad business is still very strong, but the extraordinary impression growth and price-per-ad growth that investors needed to see to justify the AI capex both moderated this quarter.

Meta's historically strong ad business, where both impressions and price-per-ad grew together in 2023, saw that trend cool off this quarter, undermining the narrative that Meta's core business is expanding fast enough to justify its AI spending.

transcript

Ben Thompson: The most interesting lines on this chart are always impressions growth and price-per-ad growth, which historically move in opposite directions for what should be an obvious reason: more impressions growth means more supply, which given stable advertiser demand, results in decreased price growth; less impressions growth means less supply, which given stable advertiser demand, results in increased price growth. That means the most extraordinary results for the underlying business are when both impressions growth and price-per-ad growth are increasing, because demand growth is outpacing supply growth. This happened most notably in 2023 when Meta finally figured out ATT (improving advertiser demand) even as the company started to heavily monetize Reels (increasing ad supply). I was a bit concerned throughout 2025 that the company was juicing supply by increasing ad load; Meta characterized this as 'ad load optimization' and evidence that their spending was justified through increased monetization. And, last quarter, it all seemed to come together: impressions growth increased, and price-per-ad growth increased; if Meta could keep that trend up then perhaps investors would tolerate their capex spend simply because the core business was on a 2023-type of expansion. Unfortunately, while this quarter's results are still very good, they're not quite as extraordinary, and that's a problem when expenses increased 55% while revenue increased 28% — and remember, a lot of the company's capex hasn't started depreciating yet (and yes, that increase includes charges related to legal proceedings, but those might not be a one-off!). The company, more than ever, needs to convince investors about its AI spending on its own merits, and frankly, I came away from the call a bit alarmed.

extends · 1supports · 1

Claim

Meta's disappointing earnings and ballooning AI capex are creating a severe timing mismatch problem that the company must urgently address with investors.

Meta reported disappointing revenue guidance and its lowest free cash flow in years due to massive AI spending, with expenses up 55% while revenue grew only 28%, and much of the capex hasn't started depreciating yet.

transcript

Ben Thompson: Meta Platforms Inc. gave a disappointing quarterly revenue forecast, stepping up pressure on Chief Executive Officer Mark Zuckerberg to allay investor concerns that the company isn't swiftly benefiting from its massive outlay on artificial intelligence. The stock fell. The social media giant also reported the lowest free cash flow in years, a sign of ballooning expenses for AI bets, including data centers and smart glasses, which could amount to $145 billion this year. Meta shares slid about 8% to $539.03. In part because it doesn't yet have a cloud-computing business and its AI products have at times been considered less competitive than some other AI labs' work, Meta has faced recurring investor skepticism that it will recoup this spending. Meta announced several new AI-related business lines in recent months, including a consumer chatbot subscription and a pay-to-use AI model for developers, though those are in early stages. On the call Wednesday, Zuckerberg teased another potential business line: A cloud computing business where Meta would sell computing power to other companies. The CEO said that a 'substantial' amount of Meta's computing power currently goes toward training its own AI models, a necessity for being a leading AI lab. But he also said that Meta has a 'large number of offers' from companies interested in buying its computing power at a 'meaningful premium' over what Meta spent to acquire it. That has created an opportunity, he added, saying that Meta must now think through the tradeoff of selling the computing power it has for a profit versus continuing to use it for its own products and services. These calculations are happening at the same time that Meta is also buying computing power from independent data-center operators — so-called neoclouds — as well.

provides context · 1supports · 1

Highlight slides

Why Meta Must Build Its Own Frontier AI Models✦ from: Meta must build its own frontier AI models because relying on open-source models is not viable — they are not as strong as frontier models, and depending on other companies' actions is risky for a company like Meta.Meta's AI infrastructure timing mismatch✦ from: Meta has a massive timing mismatch: it is double-paying for infrastructure — renting compute from third parties while building its own data centers — without a clear path to monetization, and its improving monetization story has lost luster.Full-Stack Capability Enables Better AI✦ from: Meta must build its own frontier AI models because relying on open-source models is not viable — they are not as strong as frontier models, and depending on other companies' actions is risky for a company like Meta.Double-paying without a monetization path✦ from: Meta has a massive timing mismatch: it is double-paying for infrastructure — renting compute from third parties while building its own data centers — without a clear path to monetization, and its improving monetization story has lost luster.

Related episodes

Video · All-In Podcast

The Fight Over Open Source AI, Anthropic's $1.5B Payout, NYC Socialists: Evictions = Violence?

→

Audio · We Study Billionaires

TIP825: Meta, Adobe, Booking Holdings w/ Stig Brodersen, Tobias Carlisle & Hari Ramachandra

→

Audio · All-In Podcast

Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California's Broken Elections

→