Zuckerberg bets Meta can win AI ‘own-the-stack’ despite collapsing free cash flow and rising data center costs

Meta is leaning harder into artificial intelligence spending even as its cash generation deteriorates, according to commentary reported by Yahoo Finance and TheStreet. The discussion centers on a central tension for the social media giant: data center costs are rising at the same time that free cash flow is falling sharply.

The report’s framing is blunt. It says Meta’s free cash flow fell 91% and that data center costs keep climbing. That matters because free cash flow is a key measure of how much cash a company can generate after covering day-to-day spending and capital expenditures, including major infrastructure buildouts like data centers.

In the same account, Mark Zuckerberg is presented as insisting that “the math still works,” but with a condition. The argument, as described in the report, is that Meta can only justify the AI buildout if it effectively owns the whole stack, meaning it controls critical parts of the technology required to train and run models rather than relying as heavily on outside suppliers for the core compute and software layers.

The report also implicitly ties AI strategy to infrastructure risk. When a company expands compute capacity quickly, it tends to front-load costs while monetization can lag, making margins and cash flow vulnerable in the near term. In Meta’s case, the reported combination of collapsing free cash flow and increasing data center costs suggests the company is paying the upfront price for capacity that the business expects to use for AI-driven products and services.

Meta’s broader AI push has typically been linked to improving ranking and recommendations, building AI assistants and content experiences, and supporting internal engineering at scale. While the company’s newsroom regularly highlights applied AI work and infrastructure efforts, the specific “own the whole stack” claim in this latest discussion emphasizes control over both hardware and software decisions as a route to cost discipline and performance.

Investors and analysts often evaluate whether an AI infrastructure expansion can eventually lower unit costs or unlock new revenue streams. The report’s message is that Zuckerberg believes Meta can reach that inflection point, but the cash flow data cited in the discussion underscores how expensive the transition is while models and systems are still scaling.

What is not clear from the market commentary is the level of detail Meta disclosed around the timing of returns. The report does not, in the material provided here, lay out specific targets such as expected cost per inference, capacity utilization milestones, or a quantified bridge from data center spend to future product monetization.