Amazon winds down flagship AI models in strategic overhaul

Amazon is winding down many of its in-house AI models as it revamps its artificial intelligence strategy and sharpens focus on a new frontier-model effort. The shift marks a retreat from building out a broad flagship portfolio and a bet that fewer, more targeted systems will carry more weight inside the company’s AI stack.

Business Insider said the overhaul centers on Amazon’s internal model lineup and on Peter DeSantis, Amazon’s senior vice president of Foundational AI Models, Custom Silicon and Quantum Computing. That places a longtime infrastructure executive, rather than a pure model evangelist, at the middle of Amazon’s next AI phase.

The reorganization fits a pattern Amazon has described before. In December 2025, the company tapped DeSantis to lead a new group focused on AI, silicon and quantum computing after AI chief Rohit Prasad planned to step down at the end of the year. Amazon has also said it often starts new businesses where there is an initial customer need and later reassesses where they belong as they grow and gain momentum.

That matters because Amazon has been trying to spread AI across retail operations, customer service tools, product recommendations, devices and Amazon Web Services, where it competes with Microsoft and Google for enterprise AI workloads. A narrower model strategy could cut duplication and concentrate engineering and capital on systems that matter most to customers and developers, especially those building on Amazon’s cloud. It also suggests the company is less interested in winning a public arms race over model count than in controlling the layers that make AI usable at scale.

The change lands at a moment when large-model development has become more expensive and more unforgiving. Investors want proof that heavy AI spending can turn into durable products, not just headline demonstrations. Amazon’s decision to wind down much of its flagship model portfolio signals that the company is choosing consolidation over expansion, and that its internal AI race now looks increasingly like an infrastructure play.