Meta raises AI capex forecast

Meta just made the clearest possible statement about how it sees the next phase of the AI race. It is not trimming, waiting, or trying to finesse its wa(investor.atmeta.com), up from the $115 billion-$135 billion range it gave in January, even after already spending $19.84 billion in the first quarter alone. (investor.atmeta.com) That matters because capex is the physical side of AI — chips, servers, networking, and data centers. Not c(fool.com)maller. (finance.yahoo.com) ### Why did Meta raise the number? Because its AI plan is turning into a full-stack infrastructure project. Meta has spent the last few months lining up the pieces: a 1GW data-center campus in Lebanon, Indiana, a new AI-optimized site in Tulsa, long-term su(investor.atmeta.com)er capex guide stops looking like a surprise and starts looking like the bill coming due. (about.fb.com) ### So is the busine(finance.yahoo.com)1% operating margin even while costs and expenses rose 35%. Family daily active people reached 3.56 billion in March. That is the key backdrop here: Meta is not raising spending from a position of weakness. It is doing it while the ad machine is still throwing off huge cash flow. (investor.atmeta.com) ### What does AI have to do with the ad machine? A lot. Meta’s cor(about.fb.com)more chances to show relevant ads. Management said Muse Spark — the first model from Meta Superintelligence Labs — drove double-digit percentage increases in Meta AI sessions per user after launch. It also pointed to gains in ad conversion from upgrades to ranking and ad systems. Basically, Meta is arguing that infrastructure spend is already showing up as engagement and monetization improvements, not just science-project bragging rights. (fool.com)e, but they hate open-ended AI invoices. Shares fell in after-hours trading after the report even though revenue and earnings beat expectations. The market reaction tells you the concern is not whether Meta can spend — it can — but whether the returns stay visible enough to justify a capex number that now tops out at $145 billion. (finance.yahoo.com) ### Is this just about buying more Nvidia chips? Not anymore. Meta is trying to avoid dependence on any single supplie(fool.com)MTIA chips are already deployed in the hundreds of thousands for inference workloads, and newer generations are meant to expand into ranking, recommendations, and generative-AI inference. That matters because the long-term win is not just more compute — it is cheaper compute per useful result. (about.fb.com) lower latency, more sessions, better ad conversion, higher revenue per user, maybe eventually new AI products with their own business model. The company has the scale to make this bet. The harder part is making the payoff legible quarter by quarter. (fool.com) ### Bottom line? Meta is no longer treating AI as an add-on to its social apps. It is rebuilding the company’s industrial base around it. The business is strong e(about.fb.com)udged less on whether it can build AI and more on whether all that concrete, silicon, and power turns into measurable product gains. (investor.atmeta.com)