Amazon Is Already Shopping for a Cheaper Claude

TL;DR — Amazon's renegotiated contract with Anthropic has shifted to token-based pricing, significantly raising Amazon's costs — and Amazon is now quietly exploring alternatives, including OpenAI. The biggest early bet on Anthropic is getting rocky, and it says something important about how AI partnerships actually work.

There are two versions of the Amazon-Anthropic relationship in the public mind. The first is the headline version: Amazon invested $4 billion in Anthropic, Claude is embedded in Amazon's key products, and the two companies are building the future together. The second version, now leaking out via The Information, is more complicated: a renegotiated deal shifts to token-based billing, Amazon's costs are rising significantly, and the company is internally exploring alternatives — including OpenAI, whose parent Microsoft is a direct competitor.

So what? Token-based pricing is how most AI models bill in the real world: you pay for what you use. The problem is that Amazon had been encouraging its own employees to consume Claude tokens aggressively — and now it faces paying market rates for that usage at scale. Kiro (Amazon's coding agent), Quick (its workplace AI), and Alexa for Shopping all run on Claude. When those products scale up, the bill scales up with them. The old deal insulated Amazon from that exposure. The new one doesn't.

This is what AI partnerships look like once the honeymoon ends. Early deals are often structured to grow adoption — favorable pricing, shared upside, soft commitments. Then the model gets embedded everywhere, usage explodes, and someone has to pay for the compute. That reckoning has arrived for Amazon.

Amazon's Claude exposure The pricing shift
Kiro (coding agent) Token-based billing from next year
Quick (workplace AI) Costs rising significantly at scale
Alexa for Shopping Amazon exploring OpenAI as alternative

The deeper tension is harder to dismiss. Gizmodo reports that Amazon filed a federal report on Anthropic's Fable 5 model — the same model that's been cited for matching China in cybersecurity capability — and that this contributed to government restrictions on the technology. Separately, Amazon is building its own cybersecurity-focused AI, competing directly with Fable 5's core strength. A company that's your biggest investor quietly working to restrict and replicate your best product is not exactly a partnership built on trust.

None of this means the Amazon-Anthropic relationship is over. Amazon has $4 billion of reasons to want Anthropic to succeed, and switching away from Claude would mean rebuilding products that already have millions of users. But "exploring alternatives" is a classic negotiating move, and Anthropic knows it. The new pricing takes effect next year, which means both sides have time to either re-negotiate or follow through on their alternatives.

For observers, this is a useful reminder about the difference between investment and loyalty in AI. Amazon's $4 billion in Anthropic bought influence and early access, not exclusivity. Anthropic has also deepened its Google partnership. Everyone is hedging everyone else. That's not a scandal — it's how the industry was always going to mature.

Bottom line: Amazon helping build Anthropic and then shopping for a cheaper alternative is not a contradiction — it's just capitalism catching up to the AI hype cycle.


Sources: The Information, Gizmodo, Seeking Alpha, Crypto Briefing — June 29, 2026

Tags: #AI #Amazon #Anthropic #Business #Claude