Big tech's AI spending balloons to $725 billion this year

The aggregate figure obscures an important asymmetry: these four companies are not spending identically or for identical reasons. Microsoft and Google are defending existing cloud market share while simultaneously funding the model providers (OpenAI and DeepMind respectively) that their infrastructure is meant to serve, creating internal tension between capex commitments and the returns those commitments are supposed to generate.

Platformer's railroad-boom framing from May 1st is directly relevant here: if the infrastructure buildout analogy holds, the winners won't necessarily be the companies laying the most track, but those who control the chokepoints once the network matures. The Pentagon's classified AI deals with Google, Microsoft, Amazon, and others (covered the same day via The Verge) add a second revenue thesis to justify this spending, since government contracts provide more predictable returns than commercial inference demand alone. That defense angle also reframes the Anthropic exclusion from those deals as a potential competitive disadvantage precisely when infrastructure investment requires every revenue stream available.

Watch whether any of the four platforms revises capex guidance downward in Q3 2026 earnings calls. A reduction would signal that inference revenue is not scaling fast enough to justify the current commitment pace, which would test the railroad-boom thesis directly.