Amazon reportedly shelves most Nova AI models to back single frontier-model push

Amazon is sharply scaling back its in-house AI development. According to Bloomberg reporting relayed by Futurism, the company is reducing the scope of most of its Nova AI models and reorganizing its AI teams around a short list of "highest priority" goals, after its own large language models failed to gain traction.

Most Nova models moved to bare-minimum support

Amazon insiders told Bloomberg that the company has put the bulk of its models on hold, including the text-based Nova line as well as its video- and image-generation systems. Those models are now described as being in a maintenance-only state: they remain technically supported, but receive just enough staff and computing resources to stay operational rather than to improve.

Futurism attributes the retreat to underperformance. Amazon's models were reportedly expensive to run in compute terms, and unlike OpenAI's ChatGPT, Anthropic's Claude or Google's Gemini, the Nova line never came close to becoming a recognizable name among the general public.

One frontier bet, led by a roboticist

Rather than maintaining a broad portfolio, Amazon is funnelling staff and computing power into a single frontier-model effort led by Pieter Abbeel, director of the Berkeley Robot Learning Lab. Abbeel joined Amazon through its 2024 acquisition of Covariant, a startup whose founders were building AI models for robots — a background Futurism notes aligns with Amazon's push to automate as much of its operations as possible.

An AGI office closed days earlier

The strategy shift follows another sign of retrenchment. A week before the news broke, Amazon closed one of its key AI offices in San Francisco — an 80-person site dedicated to research on artificial general intelligence, the hypothesized stage of AI development at which systems achieve human-level intelligence and reasoning.

Why it matters

One of the world's largest technology companies concluding that its first-party foundation models are not worth the compute being spent on them is a significant data point in the ongoing debate over whether large language models can ever generate a return on their enormous costs — a question Futurism explicitly raises in its coverage.

The reorganization also represents a clear strategic choice: instead of chasing consumer chatbot mindshare against OpenAI, Anthropic and Google, Amazon is consolidating around a single frontier project with a distinctly robotics-flavored leadership team, an area where its automation-heavy business gives it an obvious commercial fit.

The closure of the San Francisco AGI office, coming so soon before this pullback, further suggests lowered near-term ambition on the general-intelligence front. Futurism reads the combined moves as possibly Amazon stepping out of the AI hype cycle early, while it still can. If other major AI spenders reach similar conclusions about returns, Amazon's retrenchment could look less like an isolated failure and more like an early signal of where the wider industry is heading — though whether the consolidated, automation-oriented bet succeeds remains to be seen.