Meta forecasts weak revenue as AI spending soars to $145 billion

Meta's fourth-quarter results triggered an 8% stock decline to $539.03 after the company gave a disappointing revenue forecast and reported its lowest free cash flow in years. The core problem is scale of spending: expenses rose 55% while revenue increased only 28%, driven by a planned $145 billion outlay this year on AI data centers, cloud infrastructure, smart glasses, and related technology. The company is struggling to convince investors it will recoup this spending, especially because Meta lacks a cloud-computing business and its AI products have at times been rated less competitive than offerings from OpenAI and Anthropic.

Zuckerberg is pursuing multiple hedges. On the earnings call, he announced a potential new business line: selling computing power to other companies. He said Meta currently dedicates a "substantial" amount of its computing capacity to training its own AI models—a necessity for leading AI labs—but has received a "large number of offers" from companies willing to buy that power at a "meaningful premium" over Meta's acquisition cost. This creates a calculus: should Meta sell the compute for profit, or use it for its own products and services? Complicating that choice is the fact that Meta is simultaneously renting computing power from independent data-center operators ("neoclouds") to meet current demand while its own data-center buildout proceeds.

The underlying timing issue is severe. Last summer, when Zuckerberg realized Meta risked falling behind in the AI race, the company had no choice but to scramble for rented capacity; building data centers takes multiple years, a timeline Meta did not have. But this creates a commitment problem: Meta has already pledged almost $700 billion in future spending through long- and short-term agreements. Of this, $349.3 billion consists of non-cancelable contractual commitments mostly for third-party cloud deals, servers, and network infrastructure. An additional $347 billion is committed to future lease obligations not yet reflected on the balance sheet—including $68 billion added in July alone, with payments beginning in 2027 and 2028. The company is thus locked into years of infrastructure costs before its own facilities come online and before it has a clear path to profitable cloud services revenue.

When an analyst asked why Meta could not simply rely on open-source models, Zuckerberg responded that frontier models remain stronger than open-source alternatives and that depending on other companies' decisions is risky for a major platform. He then reframed Meta's strategy: the company is not merely a social-media-and-advertising business but a "full stack technology company" that builds its own data centers, chips, and low-level software. This vertical integration, he argued, enabled Facebook's early speed and efficiency and is now essential for AI sovereignty. The company announced several new AI business lines in recent months (consumer chatbot subscriptions, pay-to-use models for developers) and is developing personal agents for future products, but these remain in early stages. Zuckerberg also outlined an enterprise opportunity: selling APIs, business agents, and compute directly to large customers, plus building internal productivity and coding tools that Meta will offer to small and large businesses—a muscle the company has not historically exercised.

Yet the analyst commentary and financial picture suggest investor skepticism persists. Meta's traditional strength—growing both ad impressions and price-per-ad simultaneously—faltered this quarter: while both metrics rose (a rare feat driven by 2023 improvements in advertiser demand and Reels monetization), the growth was less extraordinary than the previous quarter, raising doubts about whether the company's capex spending is truly justified. Moreover, Anthropic and OpenAI are accelerating their own development and likely possess structural cost advantages in inference—the step where an AI produces an answer—that may only widen. Meta is asking investors to fund yesterday's investments in the hope of matching tomorrow's frontier, while the frontier itself moves faster.