Amazon's AI Chief Admits Its Models Trail OpenAI and Anthropic — Promises to Close the Gap in 12 Months

Peter DeSantis, Amazon’s SVP of Universal Intelligence — the executive title Amazon created for its consolidated AI organization — publicly acknowledged this week that Amazon’s own foundation models are behind OpenAI and Anthropic. Then he committed to closing the gap within 12 months.

That is a significant statement. Amazon has invested more than $8 billion in Anthropic as a strategic cloud and infrastructure partner. AWS Bedrock hosts Claude alongside GPT-4 variants and Gemini through its managed model service. Amazon has been in the enviable position of profiting from the AI race while hedging: let the frontier labs compete, host everyone’s models on AWS, and sell the picks and shovels rather than enter the gunfight.

The problem is that this neutral-platform strategy has limits. Enterprise customers signing eight-figure AI contracts want to know what Amazon’s own AI can do, especially as Microsoft bundles Azure OpenAI tightly into its enterprise agreements and Google bakes Gemini into Workspace. Amazon’s Nova model family — its in-house foundation models — has underperformed on standard benchmarks relative to GPT-4o and Claude 3.5. DeSantis acknowledging that publicly removes any pretense that the gap is a matter of positioning.

The 12-month commitment reframes the stakes. If Amazon delivers a meaningfully competitive foundation model by mid-2027, it becomes a three-way race with real incumbents on each side. If it misses, the admission will age poorly and re-open questions about whether Amazon can compete at the frontier or is structurally dependent on Anthropic.

Amazon’s advantages going into this push are not trivial. It has the world’s largest cloud customer base. It has Trainium chips — its own AI training silicon — giving it some independence from NVIDIA supply constraints. And it has vast proprietary data from AWS, Alexa, Amazon Retail, and AWS customer workloads, none of which OpenAI or Google can access.

The announcement arrives at an inflection point for the cloud AI market. The hyperscalers that can demonstrate frontier-quality models have a structural advantage: they can offer training, inference, and integration in a single contract. The ones that can’t are middleware.

DeSantis isn’t conceding the race. He’s announcing that Amazon is entering it. The clock starts now.

Sources

• TechStartups