Amazon AI strategy makes a big U-turn! Gradually phasing out most Nova flagship models, putting all resources behind “frontier models”

With intense competition in the generative AI sector, tech giant Amazon is now undergoing a major strategic shift. As reported by Business Insider, Amazon is currently boldly restructuring its AI strategy, gradually phasing out most of its in-house flagship Nova models, and fully concentrating its core engineering resources and compute power on developing brand-new “frontier models.” This shift comes right after recent layoffs and executive reshuffling in its AGI division, suggesting Amazon is trying to meet OpenAI and Google and other formidable rivals with a more focused “forward firepower” approach.
(Background: Amazon warns: memory crisis will kill game consoles and PCs; cloud gaming is the only way out)
(Additional context: “White Hair” stock god Serenity: OpenAI and Amazon’s new network architecture “won’t weaken the demand for optical modules”)

Table of contents

Toggle

• Most Nova models enter a “keep the lights on” state
• Resources heavily reassembled, pushing for a new “frontier models”
• Executive shake-up, shifting from “betting on multiple fronts” to “highly focused”

As the AI arms race moves deeper into the “deep water” phase, Amazon has decided to readjust its pace. On July 28 in Taipei time, citing informed sources, Business Insider reports that Amazon is comprehensively and significantly reorganizing its AI development strategy.

Most Nova models enter a “keep the lights on” state

The most notable move in this shift is that Amazon has started gradually phasing out most of its in-house flagship Nova models. The affected product lines include the high-end Premier and Omni models, the Reel models focused on video generation, and the Canvas models that emphasize image generation.

According to some internal employees, these models, once expected to deliver great things, have now been downgraded to a “KTLO (Keep The Lights On)” state. This means that while the company will still provide basic support to existing customers, it no longer views these models as a primary focus for development and updates. However, this does not mean Amazon will completely abandon the Nova brand—basic models including Nova 2 Sonic and Nova 2 Lite, as well as Nova Forge and the AI agent technology Nova Act, will still continue to operate.

Resources heavily reassembled, pushing for a new “frontier models”

The massive engineering talent and scarce compute resources that are freed up are now being largely redirected to a new initiative called “Frontier Model Research” (FMR). The program is led by top researcher Pieter Abbeel, who joined Amazon after the acquisition of AI robotics startup Covariant. It is said that FMR has risen to become Amazon’s top strategic priority for this year, and the team is fully developing a brand-new flagship foundation model, which is expected to be officially unveiled at this year’s re:Invent conference in the fall.

Executive shake-up, shifting from “betting on multiple fronts” to “highly focused”

This major strategic pivot is closely tied to an internal personnel earthquake at Amazon. Looking back to 2023, CEO Andy Jassy even personally backed the AGI organization’s most ambitious effort to build a foundation model, with the team led by Rohit Prasad, an executive who had long been responsible for Alexa. However, Prasad left at the end of 2025, and the AGI Lab, focused on long-term research and led by David Luan, co-founder of Adept, also officially closed last week (Luan himself left in February of this year).

Currently, the entire AGI organization has been merged under senior vice president Peter DeSantis. Under his strong leadership, Amazon’s AI strategy has shifted from the previous “betting on multiple fronts” approach—developing multiple text, image, and video models at the same time—to an “elite and highly focused” strategy centered on a small number of frontier models.

In response, an Amazon official spokesperson said: “AI models are still one of our most important tasks, and nothing has changed. Like any AI product portfolio, we will continue to adjust the model lineup based on customer needs, and provide clear guidance and migration paths when models are upgraded.” Whether this timely strategic reorganization can help Amazon claw back against formidable rivals such as OpenAI, Google, and Anthropic remains to be seen.