Amazon poured over $13 billion into Anthropic to secure a privileged seat at the AI table. Now it’s quietly shopping for other chairs.

The tech giant is evaluating alternative AI models after transitioning its Anthropic Claude integration on AWS to token-based billing in June 2026. The pricing shift has illuminated just how expensive frontier AI models are at enterprise scale, and Amazon appears to be hedging accordingly.

The pricing problem nobody saw coming

Claude models on AWS now bill through per-million-token rates or through Claude Consumption Units, priced at $0.01 each. Reports indicate some enterprise costs have climbed past $500 million monthly across the industry as token-based billing becomes the norm. That kind of number tends to get CFOs asking uncomfortable questions about return on investment.

The shift isn’t unique to Amazon and Anthropic. OpenAI and other competitors have also embraced token-based pricing structures.

A $13 billion relationship gets complicated

Amazon remains the single largest investor in Anthropic. The company added another $5 billion in April 2026, building on $8 billion in prior commitments. There’s reportedly potential for up to $20 billion in total future investment.

In June 2026, the US Commerce Department imposed export controls on Anthropic’s most advanced models. The restrictions reportedly stemmed from security concerns that Amazon executives themselves raised.

Amazon SVP Peter DeSantis added another wrinkle in mid-June 2026, stating that the company intends to compete directly with Anthropic and OpenAI’s frontier models within the next year.

What Amazon is building on its own

Amazon has been developing its own foundational AI models within AWS. The company has pledged up to 5 gigawatts of compute power through AWS by 2026. DeSantis’s comments about competing with frontier models suggest Amazon’s internal AI development has progressed far enough that leadership feels comfortable making public promises.

What this means for investors and enterprise buyers

Anthropic faces its own challenges. Export controls on its top models could limit international revenue at precisely the moment the company needs to justify the billions Amazon has invested. And if Amazon successfully develops competitive in-house models, Anthropic loses its most important distribution channel and its biggest customer simultaneously.

The geopolitical dimension adds another layer of risk. If export controls expand to cover additional AI models or additional countries, organizations that have standardized on a single model provider could find themselves unable to deploy in key markets.