After pouring more than $58 billion into two rival model makers, Amazon Web Services CEO finally responds

Big bets, blunt defence

Amazon Web Services (AWS) has quietly funneled the kind of capital normally seen in nation-scale bets: roughly $58 billion to two leading AI model companies — about $8 billion into Anthropic and a $50 billion commitment to OpenAI — commitments that AWS CEO Matt Garman addressed publicly for the first time at the HumanX conference in San Francisco. Garman’s message was simple: this is not a conflict but “coopetition” — the same partner-plus-competitor model AWS says it has run since 2006. Short version: AWS argues it had no choice if it wanted to avoid being outflanked by Microsoft Azure’s deep tie-up with OpenAI.

Why AWS says it had to act

The stakes are strategic. Microsoft’s Azure, via its longstanding relationship with OpenAI, has been eroding traditional cloud share by bundling top-tier models with platform services. OpenAI’s loosening of exclusivity with Microsoft created a narrow window for AWS to secure access to GPT-class models and an exclusive third‑party distribution right for OpenAI’s Frontier enterprise platform, it has been reported that, while Anthropic already runs deeply on AWS infrastructure. For AWS the play was defensive: lock in model variety for millions of corporate customers and avoid being a one‑model platform.

Risks and friction beneath the rhetoric

Garman’s narrative of neutral model routing masks three hard constraints. First, political risk: in February 2026 the U.S. administration restricted Anthropic’s Claude for federal use, and AWS had to migrate affected workloads — policy can trump technical redundancy. Second, model makers are building their own compute and hiring infrastructure veterans; Anthropic’s recent hires and OpenAI’s increasing Trainium deployments signal a “de‑cloudification” trend that could erode AWS’s leverage. Third, it has been reported that Microsoft is undertaking a legal review of the AWS–OpenAI deal and may pursue litigation, raising the prospect of regulatory and contractual headaches that cash alone cannot solve.

Can capital buy a moat?

The numbers are staggering: AWS’s AI-related capital expenditure is running toward roughly $200 billion this year, and its AI revenue run‑rate has climbed — but shareholders are nervous. AWS frames the $58 billion-plus as rational insurance: losing model parity would be far costlier than serving competing model vendors. But money buys access more easily than it buys long‑term alignment. If model firms continue to build their own stacks or policy shifts curtail vendor access, the trillion‑dollar infrastructure bet may turn out to be a temporary patch rather than a sustainable fortress. Who wins then — the cloud provider, the model company, or the customer? The answer is still very much up in the cloud.