Amazon Scales Back Nova AI Models — Tech Giant Restructures AI Strategy

Introduction: A Cloud Giant Changes Course

Amazon, one of the world's largest technology companies, is making a surprising strategic move: scaling back development of its Nova Premier, Omni, Reel, and Canvas AI models. These models, unveiled less than two years ago in December 2024, will now enter "keep the lights on" mode — remaining available for existing customers but no longer actively developed.

According to a Business Insider report, the cuts follow layoffs in the AGI division and the closure of the AGI Lab Amazon founded in 2024 after acquiring Adept. The pullback is significant for a company often considered one of the few capable of competing in the frontier AI race.

But Amazon isn't exiting the AI race — it's redirecting its investments. Resources now flow to the Frontier Model Research group under Pieter Abbeel, the renowned robotics scientist who joined through the acquisition of Covariant. A new foundation model is expected at the re:Invent conference this fall.

The Rise and Fall of Nova

When Amazon unveiled the Nova models in December 2024, it seemed ready to compete head-to-head with OpenAI, Anthropic, and Google. The plan was ambitious: multimodal models (Nova Omni), video models (Reel), image generators (Canvas), and high-performance frontier models (Nova Premier).

Less than two years later, development has stopped. According to Business Insider sources, Amazon faced technical difficulties reaching frontier-level performance with its own Nova models. This highlights a crucial reality: frontier AI models require extremely rare expertise — expertise that is currently concentrated in a small number of companies.

Pieter Abbeel's Frontier Research Group

Amazon's answer to this challenge is the Frontier Model Research group under Pieter Abbeel. Abbeel, a UC Berkeley professor who previously co-founded Covariant (a robotics AI startup), brings deep expertise in reinforcement learning and robotics to Amazon's AI efforts.

The group is expected to deliver a new foundation model at Amazon's re:Invent conference this fall. Rather than competing on the number of models, Amazon appears to be focusing on quality and specific use cases.

The $100 Billion Paradox

One of the most interesting aspects of this story is the paradox of Amazon's AI investments. While scaling back its own Nova models, Amazon remains one of the biggest backers of Anthropic and OpenAI — where total investments could exceed $100 billion.

This "hedging" strategy makes strategic sense: Amazon needs the best AI models for its AWS cloud platform, regardless of who builds them. If Anthropic or OpenAI produce better models than Amazon's in-house efforts, AWS customers still get access to them.

AGI Lab Closure

The closure of Amazon's AGI Lab, founded after acquiring Adept in 2024, signals a broader shift. Adept, a startup building AI agents for enterprise use, was acquired primarily for its talent. But the AGI Lab never produced a significant model.

Industry Impact

Amazon's Nova retreat sends several signals to the AI industry. First, that frontier model development is brutally expensive — even for a company with Amazon's resources. Second, that the AI talent market remains extremely tight. Third, that the "model as infrastructure" approach (investing in external models while building your own) is becoming standard.

For Georgian startups and developers, this is a reminder that AI model development requires massive resources and specialized talent. Building practical applications on top of existing models — rather than training your own — remains the most viable strategy for smaller players.

Conclusion: Infrastructure over Models

Amazon's Nova decision suggests a strategic shift: from being a model maker to being the platform where the best models run. For Amazon, this means AWS becomes even more critical as the distribution channel for AI.

📖 Source