Amazon admits its AI models lag behind OpenAI and Anthropic, but says it can catch up

TL;DRAmazon’s AI chief Peter DeSantis admitted the company’s models “haven’t been at the very frontier” and said he hopes to catch up within a year. The candour comes as Amazon simultaneously invests $33 billion in Anthropic while building its own rival models.

Amazon’s AI chief has said what the market long suspected. Peter DeSantis, the senior vice president who oversees the company’s AI models, custom chips, and quantum computing, told CNBC that Amazon’s models “haven’t been at the very frontier for the very largest, most demanding workloads.”

He added that he hopes Amazon will be “in the conversation about leading models in the coming year.” The admission frames a catch-up effort that will rely on custom silicon, proprietary data, and the scale of Amazon’s cloud infrastructure.

Amazon runs two AI plays simultaneously. Bedrock, its model marketplace, lets cloud customers access models from OpenAI, Anthropic, Meta, and Mistral through a single service, generating revenue regardless of whose model wins.

Nova2, Amazon’s in-house model released in December, has attracted roughly 50,000 customers. It has not matched the capabilities of Anthropic’s Claude or OpenAI’s GPT-5.5 on the most demanding enterprise and research workloads, hence DeSantis’s candour.

Amazon has committed up to $33 billion in Anthropic, including a $25 billion deal signed in April that gave Anthropic access to up to five gigawatts of compute on Amazon’s Trainium chips. In return, Anthropic pledged to spend more than $100 billion on AWS over the next decade.

The arrangement means Amazon profits from Anthropic’s success through both its equity stake and the cloud revenue Anthropic generates. Amazon’s Q1 2026 earnings were inflated by a $16.8 billion Anthropic-related gain, even as the company’s free cash flow fell 95%.

The competitive dynamics are pointed. Google has pledged up to $40 billion in Anthropic, making the AI lab the most courted startup in Silicon Valley.

Amazon CEO Andy Jassy also reportedly triggered the US government crackdown that shut down Anthropic’s Fable 5 and Mythos 5 last week. The episode raised questions about how Amazon balances its role as Anthropic’s largest investor with its position as a competitor building rival models.

DeSantis’s path to closing the gap centres on custom AI chips, proprietary training data drawn from Amazon’s retail and logistics operations, and the engineering scale of a team that now spans frontier models, silicon design, and quantum research. Amazon’s Trainium chips already power most of Bedrock’s inference workloads.

Trainium3