Amazon drops main Nova AI models to build one massive flagship
Standard tech hyperscaler behavior: throw billions at a dozen half-baked AI tools, realize nobody uses them, and panic-pivot. Amazon is officially trimming the fat on its corporate ego.
Internal restructuring inside Amazon has effectively killed active development for the heavyweight Nova Premier, Nova Omni, video generator Nova Reel, and image maker Nova Canvas. Existing enterprise customers will maintain access through AWS and Amazon Bedrock, but the corporate cash faucet for big feature updates on these models has been permanently shut off.
The company also quietely disbanded its dedicated AGI Lab—a unit assembled after buying out key talent from AI startup Adept—resulting in layoffs across the department. Corporate leaders apparently decided that spreading compute across five different mediocre models was less productive than funding a single competent system.
All remaining compute resources are now funneling into the Frontier Model Research group under AI expert Pieter Abbeel, who joined after Amazon acquired robotics outfit Covariant. This newly consolidated squad is rushing to build a next-generation foundational model, aiming for a grand reveal at the annual AWS re:Invent conference in late 2025.
Smaller lightweight variants like Nova 2 Lite and Nova 2 Sonic will remain on life support, alongside developer utilities like Nova Act. The upcoming frontier project may even inherit the Nova branding once it surfaces, completing a full corporate circle back to square one.
This strategic retreat comes less than eighteen months after Amazon launched the initial Nova lineup as a direct assault on OpenAI, Anthropic, and Google. While rivals pushed out reasoning architectures and autonomous agents, Amazon struggled to gain mindshare in a market already exhausted by endless hype, leaving GPU clusters idling on underperforming projects.
Watching hyper-scalers quietly bury their bloated multi-model zoo to save on GPU electricity bills is the ultimate proof that money can buy hardware, but it certainly cannot buy instant algorithmic relevance.
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.