Anthropic and Amazon agree $100bn AI infrastructure deal
Source · Technology desk
— Summary
Anthropic has committed to spend more than $100bn on chips and computing power from Amazon, its biggest deal yet to secure AI capacity after running into supply constraints. The $380bn lab will receive up to 5 gigawatts (GW) of new capacity to train and run its Claude model over the next decade, with close to a fifth arriving this year.
Amazon will invest $5bn immediately into Anthropic at current valuation and up to $20bn more over time, on top of $8bn already invested since 2023. Anthropic's annualised revenue has surged from $9bn at the end of last year to more than $30bn, with coding tool Claude Code straining capacity and causing outages during peak hours. Earlier this month Anthropic signed with Google and Broadcom for roughly another 5 GW. At current prices $100bn buys only about 2 GW, but the cost of compute is expected to keep falling over the decade of the deal.
The agreement cements a pattern of circular investments between AI labs and cloud suppliers: the partner invests in the lab, the lab buys its chips. Amazon is building Project Rainier, a 2.4 GW campus in New Carlisle, Indiana, and wants to anchor Anthropic on its Trainium chip family as a rival to Nvidia's GPUs. Source: Financial Times, 20 April 2026, George Hammond and Rafe Rosner-Uddin.
The story in one line. Anthropic will spend more than $100bn with Amazon on chips and data-centre capacity to keep up with ballooning demand for Claude, while Amazon pumps another $5bn to $25bn into the $380bn lab.
Key numbers
$100bn+ committed by Anthropic to Amazon Web Services over 10 years
Up to 5 GW of new compute capacity, with about 1 GW delivered in 2026
$5bn Amazon investment immediately at current valuation, up to $20bn more over time
$8bn already invested by Amazon since 2023
$380bn Anthropic’s current valuation
$30bn+ Anthropic’s annualised revenue, up from $9bn at end-2025
2.4 GW scale of Amazon’s Project Rainier data-centre campus in New Carlisle, Indiana
~5 GW additional capacity from Anthropic’s April 2026 deals with Google and Broadcom
Why it matters
The deal highlights the scarcity of AI compute (semiconductors and power to run them). At current prices, $100bn only buys around 2 GW - so Anthropic is effectively betting the price of compute keeps falling over the decade. Amazon wants Anthropic as a lighthouse customer for its in-house Trainium chips to compete with Nvidia’s GPUs. The investment is also “circular”: the cloud provider funds the lab and then sells it chips, a pattern that now defines the OpenAI-Microsoft, OpenAI-Amazon and Anthropic-Amazon relationships alike.
Takeaway
Anthropic is locking in multi-year capacity across three cloud-and-silicon partners (Amazon, Google, Broadcom) to avoid a repeat of 2026’s outages. The $100bn number will attract headlines; the real signal is that the AI lab with the fastest revenue ramp outside OpenAI now believes demand will stay well ahead of supply for years.
Source: Financial Times, 20 April 2026, George Hammond and Rafe Rosner-Uddin.