Amazon locked in a new deal with Anthropic that will switch its billing for Claude models to a per-token system next year. That change could send AWS bills through the roof, so Amazon’s already scouting rivals like OpenAI. Behind the scenes, tools such as the coding assistant Kiro, the workplace helper Quick and even Alexa Shopping rely heavily on Claude. With token billing, those tools suddenly look a lot more expensive—especially since Amazon recently axed an internal leaderboard that once rewarded employees for burning through as many AI calls as possible. In response, Amazon has steered more business toward OpenAI. Earlier this year it committed $50 billion to OpenAI in exchange for broad access to OpenAI’s models on AWS. By contrast, its investment in Anthropic started at $4 billion and has swelled to a possible $33 billion. Anthropic hasn’t put all its eggs in one basket, though. It just agreed to spend $200 billion on Google Cloud and chips over five years, a move that makes Google a second major infrastructure partner alongside AWS. Relations between the two have soured further after Amazon triggered a U.S. government order to shut down Anthropic’s Fable 5 and Mythos 5 models. Andy Jassy told officials Amazon researchers used Fable 5 to uncover data that could aid cyberattacks. That allegation came just as Amazon was about to launch its own security-focused AI agent, raising eyebrows about timing and motive. All this points to a more adversarial phase in what was once a cozy investor-provider relationship. Both companies remain financially tied—Amazon is Anthropic’s top cloud customer and one of its biggest backers—but now each has strong reasons to hedge its bets. The fallout will force rivals and customers across the AI industry to reassess where they place their workloads and investments.