Amazon Is Building a Custom Chip Empire — And Anthropic and Meta Are Its Biggest Validators

[AIcash]

Amazon's internal chip program has quietly become one of the most strategically important assets in the cloud wars, and two landmark deals announced this month have made that case loudly. Earlier in April, Anthropic agreed to spend $100 billion over 10 years running its Claude workloads on AWS — with a specific focus on Amazon's Trainium chips — while Amazon committed an additional $5 billion investment into the AI company in return. Now Meta has signed on to use millions of Graviton CPUs for its AI infrastructure, giving Amazon back-to-back validation from two of the most prominent AI players in the industry within the span of weeks.

The two deals cover different parts of the AI compute stack. Trainium, despite its name, handles both model training and inference, and Anthropic's commitment has locked up a significant portion of that capacity for years. The Graviton deal with Meta covers CPU-based agentic workloads — the kind of always-on, multi-step reasoning tasks that are becoming the dominant use case as AI moves from model training toward deployment at scale. Together, the two deals position AWS as the cloud provider of choice for the full AI lifecycle, from building and training models to running them in production. For Amazon's chip team, the pressure to keep delivering on that promise has never been higher — and for enterprises evaluating their own AI infrastructure, AWS now has two of the biggest names in AI as its most visible endorsements.

[kevin1]

"price-performance ratios" is CEO for "we're cheaper than Nvidia and we'd like you to notice"

[etfs]

AWS dropped the Meta deal announcement during Google Cloud Next. Andy Jassy sent that press release personally.