Amazon Invests $25B in Anthropic in Landmark AI Infrastructure Deal

Amazon（AMZN） said on Monday that it would invest $5 billion in Anthropic right away, with up to an additional $20 billion tied to commercial goals. This would bring Amazon's total investment in the AI safety company to up to $33 billion, in addition to the $8 billion it has already put in. Anthropic has agreed to spend more than $100 billion over the next ten years on AWS technologies. To do this, it has secured up to 5 gigawatts of power using Amazon's own Trainium chips to train and run its Claude AI models.

In premarket trading on Tuesday, Amazon shares went up about 3%. The market got it right: this isn't just an investment in money. It is a structural bet that Amazon's custom silicon can compete with Nvidia at the cutting edge of AI training and that Claude will be the most popular business AI model running on AWS for the next ten years.

What $100 Billion in AWS Spending Actually Buys

Anthropic has made a promise that is truly amazing in its scope. AWS has promised to spend $100 billion over the next 10 years on tens of millions of Amazon Graviton CPU cores and the full Trainium plan, which includes Trainium2, Trainium3, Trainium4, and future generations.

Trainium3 capacity should go online this year, giving Anthropic quick access to next-generation training infrastructure without the supply problems that have made it hard for the industry as a whole to buy Nvidia GPUs.

The deal also adds more international inference options in Asia and Europe, which directly addresses Claude's growing number of international customers.

Using AWS infrastructure for this kind of geographic growth is much cheaper than forming relationships with separate data centers.

They are already working together on Project Rainier, which is one of the world's biggest AI computing systems. It is made up of almost 500,000 Trainium2 chips. That cluster is now being used to train and deploy new versions of Claude, and the expanded deal pretty much ensures that Anthropic's infrastructure will be based on Project Rainier's architecture for a long time to come.

"Anthropic's commitment to run its large language models on AWS Trainium for the next decade reflects the progress we've made together on custom silicon," Andy Jassy, CEO of Amazon, said.

In his April letter to shareholders, Jassy said that Amazon's custom chip business has doubled its annualized revenue run rate to over $20 billion. He also hinted that Amazon might sell chip racks to other companies in the future. The Anthropic deal backs up both the demand signal and the long-term strategy.

The Trainium vs. Nvidia Subtext

The most important thing about the deal is what it says about the competition in the custom silicon market. For large-scale AI training, Amazon's Trainium chips are the main alternative to Nvidia's H100 and B200 GPUs.

Anthropic's decision to dedicate 5 gigawatts of power to Trainium hardware over Nvidia hardware is the most reliable endorsement Trainium has gotten from a third party.

The direct effect on competition was summed up by Truist Securities expert Youssef Squali: "The expanded partnership shows that Trainium is gaining momentum and share of AI training and inference." 

Amazon gets both the infrastructure margin and the cloud workload revenue for every dollar of AI training that goes from Nvidia hardware to Trainium. This creates an extremely strong economic flywheel.

Anthropic is already shaping the next version of Trainium. As part of the deal, engineers from Anthropic and Amazon's Annapurna Labs will work together directly on next-generation chip design.

The two teams will talk "almost daily" about everything from low-level optimization work to high-level architectural decisions." Frontier model training needs are directly affecting the design of chips. That feedback process is exactly the kind of co-development benefit that speeds up progress in silicon.

The Claude Platform Integration: Removing Friction at Enterprise Scale

The deal changes more than just the infrastructure. It also changes how businesses can tap Claude. Customers of AWS can now access the full Anthropic-native Claude Platform through their current AWS accounts. They won't need to create any new credentials, contracts, or billing relationships. The platform works with current AWS tracking and access controls, so there are no more problems with buying things or following rules that stop businesses from using AI.

At the moment, more than 100,000 businesses use Amazon Bedrock to run Claude models on AWS. It is now easy for that installed base to get to all of Anthropic's products.

Customers have already seen the real-world effects: Lyft cut the average time it took to solve a customer service issue by 87% by using Claude through Bedrock; Pfizer saved 16,000 annual search hours and cut infrastructure costs by 55% by using Claude to get drug research documents.

Anthropic's CEO, Dario Amodei, explained the strategy: "Our users tell us that Claude is becoming more and more important to how they work, and we need to build the infrastructure to keep up with demand that is growing very quickly."

For Amazon (AMZN) investors, the deal achieves three strategic goals at once: it strengthens the most commercially important AI model relationship in the AWS ecosystem; it confirms Trainium's competitive position against Nvidia; and it creates a ten-year revenue anchor that lowers the risk of AWS's AI infrastructure capital investment. Analysts at Stifel called Amazon one of their "top picks" going into the April 29 Q1 earnings report. They expected strong results in e-commerce, AWS, and advertising.

The Anthropic expansion adds another good story to that earnings cycle, which supports the long-cycle AWS AI thesis. Watch the Trainium capacity utilization metrics and Claude API call counts in Jassy's Q1 commentary. Those are the first signs that will show investors if the $100 billion commitment is already accelerating revenue in a way that can be measured.