Amazon is exploring more affordable AI options as Anthropic transitions to a token-based pricing model.

      Amazon is considering OpenAI and other alternatives following a renegotiated contract that will transition Anthropic's billing to a per-token pricing model next year. According to The Information, Amazon is seeking less expensive options than Anthropic's Claude models due to a new pricing structure that could significantly raise the company's AI expenses. Although the new pricing doesn't take effect until next year, Amazon is already looking into alternatives, including OpenAI. The report emphasizes a growing divide between two companies that were previously close allies in the AI field.

      Amazon's reliance on Claude is significant, as its coding agent Kiro, workplace assistant Quick, and consumer application Alexa for Shopping all depend on Anthropic's models, according to The Information. The move to token-based billing would substantially increase this reliance's cost, especially after Amazon recently disbanded an internal leaderboard that encouraged employees to utilize AI tokens liberally.

      In its quest for more affordable models, Amazon is turning towards OpenAI, a company with which it has been building a closer relationship. Earlier this year, Amazon invested $50 billion in OpenAI, providing the AI lab with access to its cloud infrastructure in exchange for access to its models. This followed Amazon’s original $4 billion investment in Anthropic, which has since expanded to a potential $33 billion.

      Meanwhile, Anthropic has been broadening its own partnerships beyond Amazon. The company has committed to spending $200 billion on Google Cloud and chips over the next five years, as reported by The Information, effectively making Google a significant infrastructure partner alongside AWS. Amazon's recent $25 billion investment in Anthropic included a reciprocal commitment of over $100 billion in AWS expenditures, but the Google deal indicates that Anthropic is no longer reliant on a single cloud provider.

      Tensions escalated last month when the US government directed Anthropic to shut down its Fable 5 and Mythos 5 models following a security report that originated from Amazon. Andy Jassy allegedly informed government officials that Amazon researchers had utilized Fable 5 to gather information beneficial for cyberattacks. The timing of this situation raised concerns, especially as Amazon was preparing to introduce its own cybersecurity-focused AI agent aimed at identifying vulnerabilities.

      The contract dispute, the pivot towards OpenAI, and the Fable 5 incident collectively indicate that the relationship between Amazon and Anthropic has entered a more confrontational stage. Although Amazon remains one of Anthropic's largest investors and cloud clients, both companies now have motives to lessen their interdependence. For the broader AI sector, the fracturing of this prominent investor-model-provider partnership could reshape the competitive landscape.