SmallCapSnipa posts $930B AI capex
- Social posts claiming a $930 billion hyperscaler AI capex wave are extrapolations, not company guidance, from 2026 spending plans at Amazon, Alphabet, Meta and peers. - The clearest disclosed figure is Amazon’s roughly $200 billion 2026 capex plan; CNBC estimated four megacaps together could approach $700 billion this year. - Analysts say power and construction, not just chips, now constrain AI build-outs. (datacenterknowledge.com)
The viral $930 billion AI capex number is not an announced budget from Big Tech. It is a social-media rollup built on extrapolations from disclosed spending plans and trend lines. (cnbc.com) (epoch.ai) What companies have actually said is smaller, but still huge. CNBC reported in February that Alphabet, Amazon, Meta and Microsoft were expected to spend close to $700 billion combined in 2026 on artificial intelligence infrastructure. (cnbc.com) Amazon is the biggest single disclosed number in that mix. In Andy Jassy’s 2025 shareholder letter, published in April 2026, he wrote that Amazon is investing “approximately $200 billion” in capex in 2026 and said the company has customer commitments behind that build-out. (aboutamazon.com) Alphabet’s public guidance is also well above last year’s level. CNBC reported the company sees as much as $185 billion in 2026 capex, after Alphabet’s investor materials showed its first-quarter 2026 earnings call was scheduled for April 29. (cnbc.com) (abc.xyz) Meta’s disclosed range is $115 billion to $135 billion for 2026, according to coverage of its fourth-quarter 2025 results. Microsoft has not put out a simple one-line annual capex target in the same way, but its January results showed Microsoft Cloud revenue of $51.5 billion and commercial backlog of $625 billion as it kept expanding AI capacity. (datacenterdynamics.com) (news.microsoft.com) That is why the social posts split into different totals. A roughly $690 billion or $700 billion figure tracks reported 2026 plans for the biggest hyperscalers, while a $930 billion figure usually comes from stretching those annual numbers across multiple years or adding more companies. (cnbc.com) (epoch.ai) The money is not only for Nvidia chips and servers. Companies are also paying for land, substations, fiber, cooling systems, and long-lead power supplies that let data centers run around the clock. (datacenterknowledge.com) (about.fb.com) Meta’s January power announcement shows how far that has spread beyond semiconductors. The company said agreements with Vistra, TerraPower and Oklo would support up to 6.6 gigawatts of new and existing clean energy by 2035 for grids serving its operations, including its Prometheus supercluster in Ohio. (about.fb.com) Researchers tracking filings rather than company sound bites also show the same direction. Epoch AI said combined capex at Alphabet, Amazon, Meta, Microsoft and Oracle had quadrupled since GPT-4’s release and could reach about $770 billion in 2026 if the trend continued. (epoch.ai) Moody’s, cited by Data Center Knowledge, put the six-company total near $700 billion this year and said electricity shortages and build times will keep AI capacity behind demand through 2027. (datacenterknowledge.com) So the clean read on the viral post is this: $930 billion is an aggressive synthesis, not a board-approved line item. The verified story is that hyperscalers are already guiding to hundreds of billions of dollars a year for AI infrastructure, with power and construction now as central as chips. (cnbc.com) (datacenterknowledge.com)
