Amazon's AI leader acknowledges that their models are behind those of OpenAI and Anthropic.

      TL;DR: Peter DeSantis, Amazon’s AI chief, acknowledged that the company's AI models “haven’t been at the very frontier” and expressed hope of catching up within a year. This admission coincides with Amazon's investment of $33 billion in Anthropic while simultaneously developing its own competing models. DeSantis remarked that Amazon's models have not met the demands of the largest workloads and aims to be part of the conversation regarding leading models in the next year. The effort will leverage custom silicon, proprietary data, and Amazon's extensive cloud infrastructure.

      Amazon is pursuing two AI strategies simultaneously. Its model marketplace, Bedrock, allows cloud customers to access models from various providers, generating revenue regardless of who leads in model performance. Nova2, Amazon’s in-house model launched in December, has around 50,000 customers but has yet to match the capabilities of Anthropic’s Claude or OpenAI’s GPT-5.5 for the most demanding workloads, hence DeSantis's honesty about their situation.

      In a significant move, Amazon has committed up to $33 billion to Anthropic, including a $25 billion deal from April that provides Anthropic access to considerable computing power on Amazon’s Trainium chips, with Anthropic agreeing to spend over $100 billion on AWS in the next decade. This partnership allows Amazon to benefit from both its equity stake in Anthropic and the cloud revenue generated by Anthropic’s operations, contributing to a $16.8 billion gain in Amazon's Q1 2026 earnings, despite a 95% decline in free cash flow.

      The competitive landscape is intense, with Google also committing up to $40 billion to Anthropic, making it a highly sought-after startup in Silicon Valley. Additionally, Amazon CEO Andy Jassy reportedly initiated the US government investigation that recently impacted Anthropic’s Fable 5 and Mythos 5, raising questions about how Amazon balances being Anthropic's major investor while also competing with them.

      DeSantis’s strategy for narrowing the gap relies on custom AI chips, proprietary training data from Amazon’s retail and logistics operations, and the engineering capability of a team focused on frontier models, silicon design, and quantum research. Amazon's Trainium chips currently support most inference workloads for Bedrock. The upcoming Trainium3 is expected to deliver four times the performance of its predecessor. Furthermore, Jeff Bezos’s separate AI initiative, Project Prometheus, is aiming to raise up to $10 billion, indicating that Amazon's AI ambitions go beyond just cloud computing.

      Whether custom chips and proprietary data can close the multiple-year advantage of labs that have invested heavily in training frontier models remains uncertain. DeSantis has suggested this could happen within "the coming year," though no specific benchmarks were provided for assessment.