Inside Big Tech's $650B AI Buildout

Alphabet, Meta, Microsoft and Amazon are collectively forecast to spend on the order of $650 billion on artificial intelligence in 2026. Numbers that large stop being a line item and become an industrial programme: land, power contracts, cooling, networking fabric, and the silicon that sits at the end of it all.

The interesting part is not the headline figure but the mix. A growing share goes to the physical backbone rather than to model training itself — data centres, next-generation networking, and long-dated energy agreements that lock in supply years ahead of the compute that will consume it.

For teams building on top of these platforms, the practical consequence is pricing and availability. Capacity secured today shapes what inference costs next year, and which regions get it first. That is worth factoring into architecture decisions now rather than after a bill arrives.