$725B just hit Big Tech's AI cap table — Wall Street already picked the winner

Q1 2026 earnings night was supposed to be a referendum on whether Big Tech's AI capex was getting reckless. Investors got an answer they did not expect: the spend is bigger than anyone forecast, and the market suddenly only rewards the company that can show paying customers. The combined 2026 AI capex commitment from Alphabet, Amazon, Microsoft, and Meta is now $725 billion — up 77% from $410 billion last year. That is not a typo. That is the new floor.

The reason this matters is not the size of the number. It is the split. Alphabet finished its earnings call with shares up 6% after-hours and a 34% gain for the month — Google's best month since 2004. Meta finished with shares down roughly 8% in early Thursday trading, capping a quarter that included beating revenue, beating EPS, beating net income, and still getting punished. The signal is unmistakable. After two years of "spend whatever it takes," investors have begun separating AI capex into two buckets: spending that drives revenue this year, and spending that is still a bet.

The numbers tell the story cleanly. Google Cloud revenue grew 63% year over year to $20 billion, beating analyst estimates by about $5 billion. Cloud operating income tripled. The contracted Cloud backlog hit $460 billion — almost double the prior quarter. Search revenue was up 19%, with Sundar Pichai noting search queries reached an all-time high. Microsoft raised its 2026 capex guidance to $190 billion, citing $25 billion of "higher component prices" and a continuing Azure supply crunch. Meta's 2026 capex range moved to $125 to $145 billion, up from $115 to $135 billion. Amazon held its $200 billion plan unchanged.

The backstory is about who paid for what. The 77% capex jump is the cumulative result of decisions made through 2024 and 2025 — committed power purchase agreements, multi-year GPU and TPU orders, custom-silicon roadmaps, and the second wave of data-center buildouts that came online this year. Anthropic's $40 billion Google partnership announced last week is one piece of that, but only one. The bigger pattern is that the hyperscalers stopped negotiating prices and started negotiating supply.

The countervailing angle is Meta. Reality Labs lost $4.03 billion in the quarter on $402 million of revenue. That division has now lost $83.5 billion across 21 quarters — more than every dollar of profit Meta will earn in 2026 unless the AI spend pays off. JPMorgan downgraded the stock to neutral and explicitly cited the "challenging path" to AI ROI. The question Meta has not answered is what its $145 billion of capex turns into — Llama is open-weights, advertising AI tools are inside Reels, and the AGI bet is years out. The market priced an answer it did not get.

Three things to watch over the next 90 days. First, Microsoft's Azure capacity guidance — the supply crunch comments suggest some Q2 revenue is left on the table. Second, whether Meta releases a clearer monetization roadmap for its Llama infrastructure investment before the next earnings call. Third, the Anthropic-Google compute interlock; if Anthropic revenue accelerates against Google's TPU buildout, it becomes a self-reinforcing flywheel for both companies' Q2 reports.

The takeaway is the simplest one in years. The AI race no longer has a runner-up. It has companies that turned compute into revenue and companies still paying for the privilege. Google did the first. Meta did the second. The next two earnings cycles will tell us whether anyone else can change buckets.