Amazon Phases Out Most Nova AI Models, Shifts to New Frontier Push
The company is concentrating resources on a single next-generation model under researcher Pieter Abbeel as it reorganizes its AGI division.

Amazon is executing a significant pivot in its artificial intelligence strategy, moving away from its diverse Nova model portfolio to concentrate engineering resources on a single next-generation foundation model, according to people familiar with the company's plans.
The shift follows recent layoffs in Amazon's Artificial General Intelligence organization and the closure of AGI Lab, a research group established in 2024 after Amazon hired most of the team from AI startup Adept. Business Insider first reported the details.
The Nova wind-down
Amazon has begun deprecating most of its flagship Nova models, including the high-end Premier and Omni models, the Reel video-generation tool, and the Canvas image-generation system. Internal employees describe these products as operating in "KTLO" mode—"keep the lights on"—meaning they remain available to existing customers but are no longer active development priorities.
The remaining Nova portfolio includes the Nova 2 Sonic and Nova 2 Lite foundation models, Nova Forge for model customization, and Nova Act, Amazon's AI agent technology. The company has not ruled out branding its upcoming model under the Nova name.
Frontier Model Research takes center stage
Resources have shifted to an initiative called Frontier Model Research, or FMR, led by researcher Pieter Abbeel. Abbeel joined Amazon through its acquisition of AI robotics startup Covariant and now heads the development of a new flagship foundation model expected to debut at Amazon's re:Invent conference this fall.
The reorganization accelerated after senior vice president Peter DeSantis took control of the AGI organization in December 2025, following the departure of longtime leader Rohit Prasad. People familiar with the transition say DeSantis has pursued a more concentrated strategy than his predecessor, who oversaw multiple model families spanning text, image, and video generation.
Why it matters
Amazon's consolidation reflects the brutal economics of frontier AI development. Training and operating multiple large models requires enormous computing resources and specialized talent—both in short supply. By narrowing its focus, Amazon is making a calculated bet that concentrating firepower on a single advanced model will yield better competitive positioning against OpenAI, Google, and Anthropic than spreading resources across a broader portfolio. The move also signals that even well-funded tech giants must make hard choices about which AI bets to pursue as the costs of staying competitive continue to escalate.
Organizational upheaval
The strategic shift has created uncertainty within Amazon's AI teams. Employees report receiving limited guidance about the long-term future of Nova models. The recent layoffs surprised many, particularly because frontier model researchers had been considered among Amazon's most valued technical talent.
AGI Lab, which focused on long-term AI research under Adept cofounder David Luan, shut down last week after Luan's departure in February. The lab had operated with separate compensation and leveling systems designed to compete aggressively for AI talent.
An Amazon spokesperson told Business Insider the company remains committed to investing in frontier models and continues supporting AI models in production because customers depend on them. "As with any AI portfolio, we continually evolve our model lineup based on what customers need," the spokesperson said.
The changes mark a sharp reversal from Amazon's positioning just two years ago, when CEO Andy Jassy championed AGI as the team building the company's most ambitious foundation models and AWS unveiled Nova Omni 2 as its flagship multimodal reasoning system at re:Invent.
Business Insider reported these developments based on information from people familiar with Amazon's AI operations.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
