Amazon’s top AI executive said publicly on Wednesday that the company has fallen behind OpenAI and Anthropic at the frontier of AI model development – and that he expects Amazon to be in the conversation about leading models within “the coming year.” Peter DeSantis, the senior vice president who oversees Amazon’s semiconductor, AI, and quantum computing operations, told reporters: “I think it’s a fair narrative that our models haven’t been at the very frontier for the very largest, most demanding workloads.” The statement is unusual for its directness. YourNewsClub reads DeSantis’s public acknowledgement of the capability gap as a more significant signal than the aspirational timeline – a company that controls the world’s largest cloud infrastructure and has $25 billion invested in Anthropic is openly admitting its own models are not at the frontier. That is either a sign of unusual corporate honesty or an effective move to lower expectations before Nova3 ships.
The structural context makes the gap more pointed. Amazon Web Services generates over $100 billion in annualised revenue and provides the compute infrastructure on which Anthropic’s Claude and OpenAI’s GPT-5 family run. Amazon has placed $25 billion in Anthropic directly, while OpenAI models became available on Amazon Bedrock in April 2026 – meaning AWS now distributes both of the frontier labs it cannot directly match. DeSantis described Amazon’s approach as “deliberate”: getting the data, architecture, and infrastructure foundations right before pushing for frontier capability. Amazon’s Nova2 model, released in December 2025, has attracted roughly 50,000 customers – a respectable early-adoption number, but not a number that implies frontier competitive position.
Freddy Camacho, who studies the political economy of computation and capital as dominance assets, draws the infrastructure power argument: “Amazon distributing OpenAI and Anthropic models on Bedrock is not a concession – it is a dominant-infrastructure play. The company that owns the pipe earns revenue from every model that runs through it. DeSantis is essentially saying: we don’t need to be at the frontier if we own the infrastructure layer that every frontier model depends on. That is a qualitatively different competitive position than being model-best.” YourNewsClub pins DeSantis’s framing as the most commercially honest description of Amazon’s AI strategy that any major tech executive has delivered on record this year.
Alex Reinhardt, who tracks financial systems and settlement infrastructure through digital protocols, draws the capital allocation implication: “Amazon’s $25 billion in Anthropic is a hedge and a dependency simultaneously. If Anthropic’s Claude displaces Nova on enterprise workloads, Amazon’s infrastructure revenues still grow. If Nova eventually matches Claude, Amazon captures both the model margin and the infrastructure margin. The investment architecture is designed to win regardless of which model wins.” YourNewsClub spots the Bedrock-as-distributor-of-competitors arrangement as historically unusual: few cloud providers have simultaneously invested in, competed with, and distributed the models of their most direct AI competition in a single platform.
DeSantis’s “coming year” timeline maps onto a period when OpenAI expects to list publicly and Anthropic has filed confidentially. If Amazon does not close the frontier gap before both companies complete public offerings, the window to reposition Nova as a genuine frontier competitor – rather than a capable but second-tier enterprise option – narrows considerably. The frontier model market may lock its competitive ordering before Amazon’s next major model release.
Stack this up against Amazon’s actual infrastructure position. AWS hosts Anthropic’s training runs under a $25 billion commitment, and added OpenAI models to Bedrock in April under a formal expanded partnership. Amazon earns revenue from both frontier labs whether or not Nova closes the capability gap. DeSantis acknowledged that: “We’ve been taking a very deliberate approach to get our foundations right.” The foundations he cites – custom Trainium chips, data architecture, inference infrastructure – are the same foundations AWS sells to every AI company that competes with Nova. Your News Club places the Q1 2027 earnings call as the first quarterly moment where the “coming year” claim becomes verifiable or doesn’t.