CNBC reports that Amazon SVP Peter DeSantis acknowledged the company has "not been at the very frontier" of the largest AI workloads and said he hopes Amazon will be "in the conversation about leading models in the coming year." The interview recapped Amazon's dual approach to AI: a model marketplace, Bedrock, and in-house model Nova 2, which launched December 2025 and has about 50,000 customers per DeSantis. Amazon is also investing in custom chips (Trainium, Inferentia) as part of its AI infrastructure effort. The comments frame Amazon as aiming to close a gap with industry leaders OpenAI and Anthropic, while continuing to build model, data, and hardware capabilities, according to the CNBC report.
What happened
CNBC published an interview with Amazon Senior Vice President Peter DeSantis in which he said "I think it's a fair narrative that our models haven't been at the very frontier for the very largest, most demanding workloads," and added he hopes Amazon will be "in the conversation about leading models in the coming year." DeSantis stated Nova 2 has about 50,000 customers since its December 2025 launch on Bedrock. Amazon also operates Bedrock as a model marketplace and is developing custom silicon - Trainium and Inferentia chips - as core AI infrastructure.
Technical context
Separately, at VivaTech 2026 in Paris on the same day, DeSantis described the chips-models relationship as requiring lockstep development: "If the chips are not telling the model designers what capabilities are coming and where they can optimize, then we're not doing the science necessary to take advantage of those capabilities until the chips are available, and then you're waiting months and months," per About Amazon. He described the outcome when done right as a "flywheel of better models, better chips, lower cost, and better efficiency."
Editorial analysis
Companies attempting to close a frontier-model gap typically need three elements at scale: high-quality training data and architecture investment; large-scale compute and custom hardware; and production-grade serving infrastructure. Amazon's public comments name all three: architecture work, custom silicon, and the Bedrock plus in-house model stack. The gap DeSantis acknowledges is real - Nova 2's December 2025 launch came well after GPT-4-class and Gemini-class models had established consumer and enterprise dominance.
What to watch
External signals of progress include public benchmarks or technical papers from Amazon on Nova 2, Trainium availability and throughput announcements, changes in Bedrock partner listings, and enterprise case studies showing Nova 2 on latency-sensitive workloads. DeSantis did not provide a detailed roadmap beyond the "coming year" timeframe.
Scoring Rationale #
Amazon's AI SVP making a frank public admission that the company is not at the frontier - while naming a one-year catch-up target - is notable strategy-level news from a major cloud provider. The story is supported by verified product context (Nova 2, Bedrock, Trainium) but remains primarily executive commentary without new technical releases or benchmark evidence, which caps the score in the notable range.
Practice with real Retail & eCommerce data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Retail & eCommerce problems