Meta AI Capex 2025-2026: $72B Spent, $125-145B Guided, Stock Dropped 9%

Meta spent $72.2B on capex in 2025 and now guides to $125-145B for 2026 — nearly double, and more than 2024 and 2025 combined. That's the short answer. The longer answer is that Wall Street knocked roughly 9% off the stock the day Meta said this out loud.

Meta ai capex 2025 spending didn't stay where it started. Zuckerberg opened the year guiding to $60-65B, raised it twice, and closed 2025 at $72.2B. Then, on the April 29, 2026 Q1 earnings call, he raised 2026 guidance again — to $125-145B — even as revenue beat estimates. I pulled the actual numbers, the Hyperion buildout details, the peer comparison, and the analyst reaction that followed.

What Is Meta's AI Capex in 2025, and How Fast Is It Rising in 2026?

Meta's meta ai capex 2025 total landed at $72.2B, including finance lease principal payments, after a year of upward revisions from an original $60-65B outlook issued in January 2025. That number itself represented sharp growth over 2024. But the real story is what came next: alongside Q4 2025 results, Meta guided 2026 capex to $115-135B. Just one quarter later, on the April 29, 2026 call, that range moved again to $125-145B, driven by higher memory and component pricing plus added data center construction costs. The midpoint of the new guide is nearly double actual 2025 spend, and Fortune noted the raised figure alone exceeds 2024 and 2025 combined.

Total 2026 operating expenses, including that capex, are guided to $162-169B. Zuckerberg has described the approach bluntly: "I think it's the right strategy to aggressively front-load building capacity so that way we're prepared for the most optimistic cases," he said during 2025 capex commentary. Coverage of Meta's multi-year AI infrastructure ambitions has cited a cumulative commitment approaching $600B over three years across data centers, chips, and power.

Inside the Hyperion Buildout and Meta's AI Infrastructure Bet

The single largest disclosed project is Hyperion, Meta's AI supercluster data center in Louisiana, targeting roughly 5 gigawatts of AI compute capacity, with some power supplied by gas-fired plants Meta is financing directly. In October 2025, Meta structured a joint venture with Blue Owl Capital targeting up to $27B in total development costs, kept largely off Meta's own capex line and instead booked through "other investing cash flows," with Meta covering about 20% of remaining construction costs. Separate reporting puts Meta's all-in Hyperion spend, across every financing vehicle, at more than $200B over the project's life. Zuckerberg has said the first AI supercluster from this buildout is expected online in 2026.

This financial engineering matters for anyone trying to read Meta's actual AI exposure off the income statement: a meaningful share of the infrastructure commitment sits in joint ventures and lease structures rather than headline capex, which is part of why guidance keeps moving even after a quarter that already raised it. For more on how this compares across the industry's biggest spenders, our Big Tech Earnings tracker follows capex and margin trends across Microsoft, Google, Amazon, and Meta each quarter.

Meta AI Capex vs. Microsoft, Google, and Amazon in 2026

Meta's raised 2026 guidance is still the smallest of the four major hyperscalers by dollar amount, even as it's the fastest-growing on a percentage basis. Combined, the four are on pace for roughly $725B in 2026 capex, up about 77% from an estimated $410B in 2025.

Figures are 2025-2026 estimates blended from company earnings releases, Yahoo Finance, Tom's Hardware, and SEC 8-K filings, checked June 2026. Amazon and Microsoft 2025 figures are analyst estimates; Meta and Alphabet figures are company-guided or reported.

Why the Stock Fell on Meta's AI Capex 2025-2026 News

Meta shares dropped roughly 7-9% in the session following the April 29, 2026 call, despite revenue beating consensus by about 1.5% at $56.3B, up 33% year over year, and EPS of $7.31 excluding an $8B tax benefit, up 13.7% year over year. The sell-off wasn't about the quarter Meta just reported — it was about the one it just promised. JPMorgan's Doug Anmuth downgraded the stock to Neutral, flagging intensifying full-stack AI competition and a harder path to proving AI-capex ROI beyond the existing ads business. S&P Global Visible Alpha's Melissa Otto framed the drop as investors "questioning the real ROI" on the spend, while D.A. Davidson's Gil Luria said the results "met expectations but failed to impress investors."

Meta's counter-argument is that the AI spend is already showing up in the numbers: its AI-driven "value optimization" ad suite crossed a $20B annual revenue run rate in Q1 2026, more than doubling year over year, and the Lattice and GEM ranking models lifted ad conversion rates by over 6%. Ad impressions were up 19% year over year and price per ad up 12%. That's real monetization, but it's monetization inside the existing ads franchise — not yet a standalone AI product line that justifies a $600B multi-year infrastructure commitment on its own. Hargreaves Lansdown's Matt Britzman noted the market reaction "may be overblown," since part of the capex raise reflects higher memory prices industry-wide, not just Meta building more. If you track how this spend compares to valuation multiples across the AI stack, our AI Valuations dashboard follows funding and capex trends across labs and hyperscalers together.