Amazon is reportedly considering OpenAI and its own Nova models for internal workloads after Anthropic raised the price of using Claude. According to The Information, the move centers on cutting costs rather than any notion of severing the seemingly positive partnership.
The reported reassessment is striking, because Amazon is far more than a Claude customer. It is one of Anthropic’s biggest backers and cloud partners. Back in April, Amazon said it would invest a further 5 billion dollars in Anthropic, with the option of adding up to 20 billion dollars more if certain commercial milestones are reached.
That money came on top of an earlier 8 billion dollars. Anthropic, for its part, committed to spending more than 100 billion dollars over ten years on AWS technologies, including Amazon’s own Trainium chips. So when the company that bankrolls you also weighs alternatives, it raises eyebrows.
Nova and OpenAI as fallback options
Amazon hasn’t gained much traction on their own models, or at least not externally where the effect is visible. The Nova family arrived in late 2024 for Amazon Bedrock, aimed at text, image, and video tasks, and pitched squarely around cost and latency.
Anthropic, meanwhile, has focused on anything but reining in AI usage. With tools like Cowork and Claude Tag expanding who uses AI and Claude Code remaining a sensation among developers (not to mention the enormous API use of Claude), Anthropic has expanded who uses AI and to what ends. Moves toward more restrictive usage limits and usage-based pricing are turning the screws on customers, however.
OpenAI, at any rate, has also become a more realistic option for AWS. Earlier this year, OpenAI brought its latest models and the Codex coding agent to Bedrock, following changes to its previously restrictive Microsoft cloud arrangement. This let AWS serve customers wanting a Claude alternative without moving workloads off its cloud.
In a sense, LLM coverage is flattening out across the hyperscalers, and prices are rising. This combination of factors will likely stress the supposed advantage that Anthropic and OpenAI specifically have over their rivals, be they the hyperscalers themselves or open-source alternatives.
Pressure, not a breakup
Anthropic also recently moved ahead of OpenAI in the IPO race. Releases like Fable 5/Mythos 5 (although currently mostly unavailable) and Claude Tag for Slack, an AI agent that operates as a shared team member, are clearly intended to boost valuation by showcasing the company’s scope across AI use cases. That momentum may explain the leverage behind higher pricing.
Whether Amazon will actually shift internal workloads to OpenAI or Nova remains unknown for now. The reported evaluation appears tied to commercial pressure rather than a damaged relationship between the two companies. In addition, AI usage within large enterprises, especially tech-savvy ones, cannot be reduced down to “Company A uses LLM X” as various models will be deployed in a multitude of ways, from SaaS integrations to coding harnesses and a mix of proprietary tooling with external LLMs.