Amazon is quietly winding down most of its flagship Nova AI models to bet everything on one new frontier model

[Taz18]

Amazon is overhauling its AI strategy, deprecating most of its flagship Nova models, including the high end Premier and Omni models, the Reel video generation model and the Canvas image generation model, according to Business Insider reporting. Some existing models are reportedly being shifted into what employees describe internally as keep the lights on mode, meaning they'll continue receiving support for current customers but are no longer active development priorities. In their place, Amazon is redirecting resources toward a single new frontier model effort led by researcher Pieter Abbeel, with frontier research becoming a top company priority this year and the new flagship model expected to debut at Amazon's re:Invent conference later in 2026

An Amazon spokesperson told Reuters the company continually evolves its model lineup based on customer needs, and confirmed it will keep investing in and supporting the Nova models customers currently rely on while also investing in next generation frontier research. The new model may still emerge under the existing Nova brand rather than as an entirely new product line. This restructuring follows job cuts last week within Amazon's artificial general intelligence group and the shutdown of its AGI Lab research unit, created back in 2024 after Amazon hired away several prominent AI researchers, with a spokesperson describing the changes as sharpening focus on the initiatives that matter most for customers

Amazon has struggled to generate the same public attention around Nova that rivals OpenAI, Anthropic and Google have achieved with their own flagship models, instead leaning more heavily on providing core AI infrastructure for enterprise customers through AWS. The company consolidated its AGI operations under longtime cloud executive Peter DeSantis back in December, following the departure of several senior AGI executives over the past year. The reorganization suggests Amazon is betting that concentrating compute, talent and budget on fewer, more ambitious projects gives it a better shot at genuinely competing at the frontier, even if it means some short term uncertainty for customers currently building on the models now being deprioritized

Amazon Is Rethinking Its Biggest AI Betuk.finance.yahoo.com

[SammyZayn]

Keep the lights on mode is such a telling internal description, that's corporate speak for we've basically given up actively improving this but don't want to strand existing customers entirely

[Terry84]

Consolidating around one frontier model led by a single prominent researcher after years of a scattered, multi-model Nova lineup suggests real institutional acknowledgment that the spread-thin approach simply wasn't working

[Quiet Depot]

Struggling to generate buzz compared to OpenAI, Anthropic and Google despite Amazon's massive AWS distribution advantage is a surprising gap, having that much infrastructure reach and still not building models people talk about says something about execution rather than resources

[Calm Charlotte]

The AGI Lab shutdown after being created specifically to attract star researchers back in 2024 is a real admission that effort didn't pan out the way Amazon hoped when it made those high profile hires

[LurkingLegend]

Betting everything on one frontier model debuting at re:Invent is a real moment of truth for Amazon's whole AI strategy, that single launch now carries enormous weight for how seriously the company gets taken in this race going forward

[AJStyles92]

This being the second major AGI related restructuring at Amazon within a week shows how much internal pressure there must be to actually show competitive progress rather than just steady incremental Nova updates

[DigitalNomad62]

Existing customers on the deprecated models are the ones who should be watching this most closely, migration guidance and support quality will matter enormously for anyone who built real production systems on Nova Premier or Omni specifically

[NeonPilot]

This is a good example of a major tech company publicly admitting a scattered strategy wasn't working and making a real, concentrated bet rather than continuing to hedge across too many products at once