Amazon, Microsoft, Alphabet, Meta Plan $725B AI Capex in 2026

The Financial Times' running tally of hyperscaler capex, compiled from Q1 earnings, now has Amazon, Microsoft, Alphabet and Meta collectively planning around $725 billion of capital expenditure in 2026. That is up 77% from last year's record of roughly $410 billion, and the bulk of the money is going into AI infrastructure: GPU clusters, custom silicon, and data center construction.

The breakdown from Q1 earnings is worth sitting with. Amazon is guiding to about $200 billion for the calendar year, Microsoft is tracking near $190 billion, Alphabet is targeting a $175 to $185 billion range, and Meta has guided $115 to $135 billion. That is four companies each committing sums close to a national capital budget, all pointed at the same three inputs at once: chips, power, and floor space.

Why it matters if you are not a hyperscaler CFO. The spending has already started to reshape free cash flow, and analysts are moving 2027 estimates well past a trillion dollars. Evercore and Bank of America now model combined 2027 capex above $1 trillion, with some pushing 2026 estimates up to a $800 to $900 billion range as guidance keeps drifting higher. The pricing power in that chain flows to whoever holds the constrained inputs, principally Nvidia and the custom-silicon partners, and increasingly the utilities and land close to substations.

The honest caveat is that these are guidance numbers, not booked spend, and the reporting does not tell you what share is already contracted versus optional, or what internal return hurdles any of the four are actually using. It also does not settle the bear case, that if AI revenue does not compound fast enough this cycle starts to look like an overbuild. One analyst quoted in the coverage called the bear thesis 'garbage', which is honest as a view but not as evidence.

The forward read is that AI capacity is likely to be less scarce for buyers than the last twelve months suggested, and the strategic question for anyone building on top of these platforms shifts from 'can I get GPUs' to 'am I on the right side of the pricing power that just moved upstream.'