VIP Deep Dive: Inside Amazon's $220B AI Infrastructure Build - What's Being Built, Where, and Why

Amazon has raised its 2026 AI infrastructure spending to $220 billion, the FT reported. The Seattle-based group had in April expected to spend $200 billion in 2026; the latest number is a $20 billion upward revision, and it is the largest single-year AI capex figure any hyperscaler has publicly disclosed.

The $220 Billion Build

The figure covers AWS data-center construction, networking, power infrastructure, GPUs and custom silicon. Amazon does not break out Trainium vs. Nvidia allocations, but the bulk of the build is concentrated in a small number of US regions - Virginia, Ohio, and Oregon - with new sites in Mississippi and Saudi Arabia coming online in 2027.

"$220 billion is the new baseline for hyperscaler AI capex," one analyst told the FT.

What It Buys

At an estimated $30,000-50,000 per GPU, $220 billion buys roughly 4-6 million GPUs across the next 18-24 months. Most of those are Nvidia Blackwell and Blackwell Ultra, but an increasing share are Amazon's own Trainium 3 silicon, which is now in production. Power, cooling, and shell construction typically add another 50-70% on top of GPU cost, which is why the headline number is so much larger than the GPU count alone would suggest.