Amazon is renegotiating its contract with Anthropic and will pay significantly more to use Claude models under a new token-based pricing structure taking effect next year. The dispute has gone public, marking the most visible crack yet in a top-tier AI partnership — one where Amazon is both a major customer and a roughly $4 billion investor.
Key facts
- What: A renegotiated contract is expected to sharply raise Amazon's bill for Anthropic's AI, pushing Amazon toward OpenAI and its own models even though it's an Anthropic investor.
- When: 2026-06-29
- Primary source: read the source
Anthropic makes Claude, a family of leading AI models, and Amazon weaves Claude deep into its own products: its coding assistant Kiro, its workplace assistant Quick, and the shopping features in Alexa all lean on it. That dependence is exactly what makes a price change so consequential.
Anthropic is moving to a token-based pricing structure where cost scales with how much text the models read and write. For a company running Claude across many high-volume products, that math reportedly points to a much bigger bill. In response, Amazon is hedging hard. It has committed a reported $50 billion to OpenAI, Anthropic's chief rival, in an infrastructure-for-access arrangement, and it's leaning on its own in-house Nova models to cut dependence. The relationship has also turned personal: Amazon reportedly leaked a report suggesting Anthropic's Fable 5 model could "go rogue," around the same time Amazon was standing up its own cybersecurity AI agent. Amazon's official line denies any of this, with a spokesperson insisting the contract changes won't raise costs and describing an ongoing "multifaceted partnership."
The dynamic is straightforward: a customer and an investor, suddenly shopping around because the pricing model changed underneath it. The competition is shifting from pure capability — whose model is smartest — to economics: who can afford to run these models at scale, and who controls the pricing. "Diversify your model supplier" is quietly becoming corporate doctrine, and the striking part is that it's being adopted even by a company that owns a chunk of the supplier it's hedging against. Anthropic, for its part, has been growing fast and signing deals with Amazon's rivals, including a large cloud arrangement with Google, which only sharpens the tension.
The public denials and the "multifaceted partnership" language suggest the relationship survives this round, and some of the leaks read like negotiating leverage rather than a clean break. But the structural pressure underneath the rhetoric is real: token-based pricing changes the economics of building agentic AI products, and every large buyer is now asking whether being locked into a single model provider is a risk worth carrying. Watch whether other enterprises follow Amazon's lead into explicit multi-model hedging.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)