Skip to main content
Amazon Web Services (AWS) logo, symbolizing Nova AI model scaling back for new Frontier team.

Editorial illustration for Amazon Scales Back Nova AI Models, Bets on New Frontier Team

Amazon Scales Back Nova AI Models, Shifts Strategy

4 min read

Amazon is pulling back on most of the Nova AI models it introduced with fanfare in December 2024, according to Business Insider, which cites people familiar with the matter. The flagship Nova Premier and Omni models, the Reel video generator, and the Canvas image tool are all losing active development. They'll stay running for existing customers in what insiders describe as "keep the lights on" mode, but Amazon isn't putting new engineering effort behind them.

The shift comes after layoffs hit Amazon's AGI division and the company shut down the AGI Lab it built in 2024 following its acquisition of Adept. That's a notable retreat for a unit that was supposed to anchor Amazon's homegrown AI ambitions.

The money and attention are moving elsewhere. Pieter Abbeel, who came to Amazon through its acquisition of robotics startup Covariant, now leads the Frontier Model Research group, which is set to unveil a new foundation model at the re:Invent conference this fall. Amazon says AI models remain central to its plans, even as it keeps pouring money into outside bets like Anthropic and OpenAI.

Amazon isn't exiting the AI race, though. Resources now flow to the Frontier Model Research group under Pieter Abbeel, who joined through the acquisition of robotics startup Covariant. A new foundation model is due at the re:Invent conference this fall.

Why this matters

Amazon spent a year telling developers Nova was its answer to GPT-4 and Gemini, then quietly moved the flagship models to maintenance mode less than a year after launch. That's a fast reversal for a product line unveiled in December 2024 with real marketing behind it. If you built on Nova Premier or Omni, or planned to use Canvas or Reel in production, you're now on a model family Amazon isn't actively improving. That's a real signal for anyone weighing platform risk when picking a foundation model vendor: Amazon can cut a division as fast as it stood one up.

The bigger story is where the money and talent went. Pieter Abbeel's Frontier Model Research group, built on the Covariant acquisition, is now the priority, while the AGI Lab from the Adept deal got shut down entirely. That's two acquisitions absorbed and one already discarded in under two years. For researchers watching AWS as an employer, or founders wondering if Amazon will still be a serious model provider next year, the churn itself is the answer worth tracking.

Common Questions Answered

Which Amazon Nova AI models are being scaled back and moved to maintenance mode?

Amazon is pulling back on the flagship Nova Premier and Omni models, along with the Reel video generator and Canvas image tool. These models will remain running for existing customers in "keep the lights on" mode, but Amazon is no longer putting active engineering effort into their development.

What is Amazon's new strategic focus after scaling back Nova models?

Amazon is redirecting resources to the Frontier Model Research group led by Pieter Abbeel, who joined through the acquisition of robotics startup Covariant. The company plans to introduce a new foundation model at the re:Invent conference this fall, signaling a shift in its AI development priorities.

How quickly did Amazon reverse course on Nova after its December 2024 launch?

Amazon moved the Nova flagship models to maintenance mode less than a year after their December 2024 launch, despite initially marketing them as the company's answer to GPT-4 and Gemini. This represents a notably fast reversal for a product line that received significant marketing support at its introduction.

What are the implications for developers who built applications on Nova Premier or Omni?

Developers who built on Nova Premier, Omni, Canvas, or Reel are now on a model family that Amazon is no longer actively improving, creating platform risk concerns. This signals that developers should carefully evaluate the long-term stability and support of AI platforms when making production decisions.

LIVE20:02GM Engineers Now Spend Just 15% of Time Writing Code After AI Overhaul