Amazon lifts 2026 AI capex to $220B and still expects to be capacity-constrained
Amazon raised 2026 capex to $220B from about $200B — a memory-price increase rather than a bigger construction push — while AWS grew 37% to $42.2B, its fastest in 18 quarters. Demand is not the constraint: backlog sits at $496B, Andy Jassy expects capacity to stay short through 2027, and contracted demand extends into 2028. Anthropic and OpenAI committed multi-year, multi-gigawatt volumes to Trainium, giving frontier-scale training a credible path off Nvidia.
Source: datacenterknowledge.com ↗
We will still not have enough capacity to meet all the demand we have in 2026. I believe this dynamic will also be true in 2027, too.
Andy Jassy, Amazon CEO
Why this matters
- → Demand for AI compute now extends into 2028, but supply remains the bottleneck through 2027.
- → Memory costs, not construction, are driving the $20B increase—signaling where cost pressure is tightening.
- → Anthropic and OpenAI's multi-gigawatt Trainium commitments validate Amazon's path away from Nvidia dependence.
Gold rush logistics