Amazon Is Betting Another $25 Billion on Anthropic

Amazon announced on Monday that it will invest up to $25 billion more in Anthropic, building on the $8 billion it committed between 2023 and 2025. The deal puts $5 billion in immediately; the remaining $20 billion is conditional on commercial milestones. If all of it lands, Amazon's total stake in Anthropic reaches $33 billion.

The investment comes with an infrastructure commitment that may matter as much as the dollars. Anthropic has agreed to spend more than $100 billion on Amazon Web Services over the next decade and has committed to running up to five gigawatts of AI compute on AWS infrastructure, including the company's Trainium3 chips, which are expected to arrive in volume later this year.

That arrangement turns what looks like a venture bet into something closer to a captive supply relationship. Amazon gets a guaranteed customer for its cloud and chip infrastructure. Anthropic gets the capital and compute it says it needs to stay competitive at the frontier. Both sides benefit from the other's success in ways that go well beyond a typical investment deal.

Google has also invested several billion in Anthropic, making the AI lab unusual in having secured major commitments from two of the three biggest cloud providers simultaneously. Neither company holds a controlling stake, but the combination of financial and commercial ties makes Anthropic's compute situation more secure than most AI labs its size.

The deal was first reported by CNBC on April 20, with Bloomberg confirming the structure shortly after. Amazon has not disclosed what equity stake, if any, the new tranche represents.

Why infrastructure is the real story

Frontier AI training runs are expensive in ways that compound quickly. The cost of training the largest models has roughly doubled every year since 2020, according to Stanford's 2026 AI Index. At that pace, the gap between well-capitalized labs and everyone else grows faster than most outside observers account for.

Anthropic CEO Dario Amodei has said publicly that compute access is the binding constraint on AI development. Securing five gigawatts of committed capacity with a defined timeline is not just a financial arrangement. It is a statement about where Anthropic expects to be in three to five years.

The Trainium3 detail is worth noting. Amazon's AI chip program has lagged Nvidia in public attention, but it has been improving steadily. Anthropic's commitment to run models on Trainium3 gives Amazon a flagship customer for chips it needs to prove at scale. That relationship, more than the headline dollar figure, may be what both parties care about most.

This deal follows Jeff Bezos's separate $10 billion commitment to Project Prometheus announced earlier this week. The two investments are independent but tell a consistent story: the largest pools of capital in technology are moving toward AI infrastructure at a pace that is difficult to fully absorb.