Anthropic lands $100B AWS deal
- Anthropic and Amazon expanded their partnership last week with a 10-year commitment for Anthropic to spend more than $100 billion on AWS. - The deal secures up to 5 gigawatts of capacity for Claude, spans Trainium2 through Trainium4, and extends inference capacity into Europe and Asia. - It matters because Anthropic is now tying frontier-model growth to cloud partners’ custom chips, not just Nvidia’s GPU supply.
Cloud compute is the story here — and the stakes are simple. Frontier AI labs need absurd amounts of power, chips, and data-center capacity just to keep training and serving bigger models. The bottleneck has been getting enough of all three at a price that does not wreck the business. Last week, Anthropic made its answer very explicit: it said it will spend more than $100 billion over 10 years on AWS technologies, locking in up to 5 gigawatts of new capacity for Claude. (anthropic.com) ### What did Anthropic actually sign? This is not Amazon investing another $100 billion into Anthropic. It is Anthropic committing to buy more than $100 billion of AWS infrastructure and services over the next decade. The package covers training and inference, includes AWS Graviton CPUs and Trainium chips from Trainium2 through Trainium4, and leaves room for future Amazon silicon generations too. (anthropic.com) ### Why is 5 gigawatts such a big deal? Because 5GW is utility-scale power — the kind of number that makes this feel less like a software contract and more like an industrial build-out. Anthropic says that capacity is for training and running Claude, which means the agreement is about both making new models and serving existing ones at global scale. It also s(anthropic.com)ainium3 capacity expected later this year. (anthropic.com) ### Is Anthropic going all-in on Amazon chips? Mostly, but not entirely. Amazon remains Anthropic’s primary cloud provider and training partner, and Anthropic says the new agreement expands work on Project Rainier, the huge AWS cluster already used for Claude. But Anthropic also said three weeks ago that it trains and runs Claude across AWS Trainium, Google T(anthropic.com)y. (anthropic.com) ### What is Project Rainier? Basically, it is the physical expression of this strategy. Amazon said in October 2025 that Project Rainier was fully operational with nearly half a million Trainium2 chips, and that Anthropic was already using more than 1 million Trainium2 chips across its workloads. The new deal scales that logic up again — fewer spot purchases, more pre-arranged industrial capacity. (aboutamazon.com) ### Where does Google fit in? Google is still in the picture, but the cleanest public number tied to Google right now is not an Anthropic financing package. Google announced a separate $40 billion investment in Texas in November 2025 for cloud and AI infrastructure through 2027. Anthropic’s own recent statement says it is (aboutamazon.com)mary cloud provider. The rumored “Google package” may exist in reporting behind paywalls, but the public record is narrower than the rumor version. (blog.google) ### Why does this matter beyond Anthropic? Because AI labs are starting to look like heavy industry. The old question was who had the best model. The new one is who can lock in enough power, chips, and networking before everyone else does. If Anthropic can spread workloads across Trainium, TPUs, and GPUs, it gets l(blog.google)icon, switching gets harder. (anthropic.com) ### So what changed? The new thing is not that Anthropic works with Amazon — that has been true since 2023. The new thing is scale and duration. A 10-year, $100 billion-plus commitment turns a strategic partnership into a capacity reservation system for frontier AI. That is the real signal. (anthropic.com)ought a long runway for Claude — and in this market, guaranteed compute may matter almost as much as model quality.
