Amazon is pulling the plug on most of its flagship Nova AI fashions — and the actual query is whether or not the corporate is retreating or repositioning for one thing greater. The transfer to wind down Nova Premier, Omni, Reel, and Canvas alerts a pointy strategic reset, one which trades breadth for a single, concentrated wager on a frontier basis mannequin.
Key takeaways
- Amazon is deprecating most flagship Nova AI fashions together with Premier, Omni, Reel, and Canvas, placing them in “hold the lights on” mode for current clients solely.
- Sources are shifting to Frontier Mannequin Analysis (FMR), a brand new inner effort led by Pieter Abbeel, aimed toward constructing one aggressive basis mannequin.
- A brand new flagship AI mannequin is predicted to debut at Amazon’s re:Invent convention this autumn, and will nonetheless carry the Nova identify.
- Amazon closed its AGI Lab and lowered employees; Peter DeSantis has led the consolidated AI group since December, narrowing focus to fewer frontier bets.
- AWS holds compute commitments price $138bn from OpenAI and greater than $100bn from Anthropic, underlining the place Amazon’s actual AI leverage sits.
Amazon Winds Down Most Nova AI Fashions
The scope of the pullback is wider than it first appeared. In line with reporting by Enterprise Insider, Amazon has begun deprecating the high-end Premier and Omni fashions, the Reel video generator, and the Canvas picture generator. Internally, workers have described these merchandise as working in “KTLO” mode — engineering shorthand for “hold the lights on” — that means they continue to be supported for current clients however now not obtain lively improvement funding.
That framing issues. These aren’t merchandise being deleted in a single day. However they’re successfully frozen, whereas the engineers and compute assets behind them transfer elsewhere.
What Will get Minimize — and What Stays
Not all the pieces within the Nova line goes away. Amazon confirmed it’s retaining Nova 2 Lite and Nova 2 Sonic as basis fashions, alongside Nova Forge, a service for constructing and customizing fashions, and Nova Act, its AI agent software. The brand new frontier mannequin that FMR is growing might finally debut underneath the Nova model as nicely.
Only a 12 months in the past, AWS used its re:Invent stage to unveil Nova Omni 2 as its flagship multimodal reasoning mannequin. Lower than three years after CEO Andy Jassy personally championed AGI because the staff that might construct Amazon’s most bold basis fashions — prompting the creation of six new analysis teams — the group is now retiring vital chunks of that very same lineup.
New Concentrate on Frontier Mannequin Analysis Led by Pieter Abbeel
The centerpiece of Amazon’s new route is Frontier Mannequin Analysis, or FMR — an inner initiative that has grow to be the corporate’s high AI precedence this 12 months. Sources that when fed a number of mannequin households at the moment are flowing right here.
One Mannequin to Win With
FMR is led by Pieter Abbeel, who joined Amazon via its 2024 acquisition of AI robotics startup Covariant. Beneath his route, the staff is constructing a single flagship basis mannequin anticipated to debut at Amazon’s re:Invent convention this autumn, in accordance with Reuters and Enterprise Insider.
The logic is easy, if uncomfortable to confess publicly. Beneath former AI chief Rohit Prasad, Amazon unfold its engineering assets throughout textual content, picture, and video mannequin households concurrently. That strategy produced a large portfolio however nothing that really competed on the frontier. Peter DeSantis, who took over the consolidated AI group in December after Prasad’s departure on the finish of 2025, has since concentrated expertise and scarce compute on fewer, higher-stakes efforts.
Management Transition and the Price of Making an attempt to Do All the things
Prasad’s exit was adopted in February by the departure of David Luan, the Adept cofounder who led Amazon’s AGI Lab. The lab itself was subsequently shut down. AGI Lab had been established in 2024 after Amazon employed many of the staff behind AI startup Adept and licensed its know-how — a big funding that’s now being written off organizationally.
The sample right here displays a broader business pressure. Constructing aggressive frontier fashions requires not simply expertise and capital, however sustained, targeted compute allocation. Spreading that throughout six or seven parallel mannequin households dilutes all three. DeSantis seems to have made the unsentimental name that Amazon wanted to choose a lane.
Organizational Modifications and What They Sign
The AGI group Amazon created in 2023 had additionally developed its personal inner tradition — separate leveling and compensation methods designed to compete aggressively for AI expertise. Staff informed Enterprise Insider that the current layoffs shocked many, partly as a result of frontier mannequin researchers had lengthy been thought of amongst Amazon’s most valued technical employees. The restructuring collapsed AGI underneath DeSantis’s broader group, which additionally covers silicon improvement and quantum computing.
Amazon’s San Francisco AGI web site, an 80-person analysis group, has additionally closed, in accordance with GeekWire.
Amazon’s Infrastructure Benefit — and the Sincere Reckoning
An Amazon spokesperson pushed again on the narrative of retreat. “AI fashions stay probably the most necessary issues we’re engaged on, and that hasn’t modified,” the spokesperson informed Enterprise Insider, including that the corporate “frequently evolve[s]” its mannequin lineup based mostly on what clients want and offers clear migration paths as fashions advance.
That official place is defensible, however the strategic context tells a extra nuanced story. Amazon by no means established Nova as a well known AI model the way in which OpenAI, Anthropic, or Google did with their fashions. Its methods additionally proved costly to run relative to the worth they returned, at a second when cheaper rivals had been flooding the market. The aggressive stress was actual.
What Amazon does have, and what no quantity of mannequin deprecation can diminish, is its infrastructure place. AWS carries compute commitments price $138 billion from OpenAI and greater than $100 billion from Anthropic. Jeff Bezos has publicly referred to as the corporate’s customized Trainium chips a fourth firm pillar — a multibillion-dollar enterprise constructed on the premise that Amazon wins by being the engine room for everybody else’s AI ambitions.
That could be a highly effective place. But it surely additionally creates a strategic pressure that FMR must resolve: Amazon is the owner for 2 of its most formidable AI rivals. Whether or not it will probably concurrently host Claude and GPT-4 on AWS whereas producing a frontier mannequin that builders actively favor over both is the wager that re:Invent will put to the take a look at.
FAQ
Which Amazon Nova AI fashions are being discontinued?
Amazon is winding down most flagship Nova AI fashions together with Premier, Omni, Reel, and Canvas. These are being positioned in “hold the lights on” mode, that means they continue to be supported for current clients however are now not an lively improvement precedence.
What’s Amazon’s new focus in AI mannequin improvement?
Amazon is concentrating assets on Frontier Mannequin Analysis (FMR), led by Pieter Abbeel, with the purpose of growing a single aggressive basis AI mannequin reasonably than sustaining a number of parallel mannequin households.
When will Amazon’s new basis mannequin debut?
The brand new flagship AI mannequin is predicted to debut at Amazon’s re:Invent convention this autumn. It might nonetheless carry the Nova model identify.
Has Amazon stopped investing in AI fashions altogether?
No. An Amazon spokesperson acknowledged that AI fashions stay a high precedence, and the corporate says it frequently evolves its lineup based mostly on buyer wants whereas offering clear migration paths as fashions advance.
Article produced with the help of synthetic intelligence and reviewed by the editorial staff.