Amazon Winds Down Flagship Nova Models in AI Strategy Shift

AI News

July 30, 2026

Amazon Winds Down Flagship Nova Models in AI Strategy Shift

Amazon is reportedly reducing development of several flagship artificial intelligence models as part of a broader effort to concentrate its engineering talent and computing resources on a smaller number of advanced projects.

The company has begun deprecating Nova Premier, Nova Omni, Nova Reel and Nova Canvas, according to people familiar with the changes. These models cover high-end reasoning, multimodal processing, video generation and image creation.

Rather than maintaining a wide portfolio of specialized systems, Amazon is shifting resources toward a new frontier foundation model expected to compete more directly with leading models from OpenAI, Google and Anthropic.

Existing models will remain supported

The change does not mean Amazon is immediately shutting down every affected product.

Some employees reportedly describe the older models as entering a “keep the lights on” phase. They will continue to receive enough support for existing customers, but they are no longer expected to remain major development priorities.

An Amazon spokesperson said the company regularly updates its model portfolio according to customer requirements and will provide guidance and migration paths as its technology advances.

Amazon also said it remains committed to the Nova models customers currently depend on while continuing to invest in next-generation frontier research.

New frontier model becomes the priority

Amazon is moving more resources toward an internal initiative called Frontier Model Research.

The group is led by AI researcher Pieter Abbeel, who joined Amazon after the company hired leaders from robotics startup Covariant. Abbeel was placed in charge of frontier-model research during an earlier restructuring of Amazon’s AI organization.

The group is developing a new flagship foundation model that could be introduced during Amazon’s annual re:Invent conference later in 2026. The model may still be released under the Nova brand, although Amazon has not officially confirmed its name or launch date.

This approach would allow Amazon to focus scarce chips, training capacity and specialized researchers on one strategically important model rather than distributing them across several separate projects.

Amazon narrows its AI portfolio

The reported changes follow a wider reorganization inside Amazon’s artificial general intelligence division.

Amazon recently reduced jobs within the group and closed AGI Lab, a long-term research unit created after the company hired much of the team behind AI startup Adept. The restructuring suggests Amazon is prioritizing projects that could have the greatest near-term impact on customers and its AWS business.

The company had previously pursued a broad model strategy spanning text, speech, images and video. At AWS re:Invent 2025, Amazon presented Nova 2 as a major expansion of its internal AI portfolio, emphasizing reasoning, multimodal capabilities and conversational AI.

Less than a year later, the company appears to be adopting a more concentrated strategy.

Some Nova products will continue

Amazon is not abandoning its entire in-house model program.

The remaining portfolio includes Nova 2 Lite, a cost-efficient reasoning model, and Nova 2 Sonic, which is designed for conversational speech applications. Amazon is also continuing Nova Act, its technology for agents that interact with software interfaces.

Nova Forge also remains an important part of the strategy. The AWS service allows enterprise customers to customize Nova models using proprietary data and access checkpoints from different stages of model training.

These products support Amazon’s wider goal of selling AI infrastructure and development tools even when customers choose models from other providers.

AWS remains central to the strategy

Amazon has struggled to generate the same level of attention for Nova as OpenAI has achieved with GPT, Google with Gemini and Anthropic with Claude.

However, Amazon remains a major AI infrastructure provider through AWS, Bedrock and its Trainium chips. Bedrock offers customers access to models from multiple companies rather than requiring them to use Amazon’s own systems exclusively.

Amazon has also expanded its partnerships with external AI developers. In February 2026, the company announced a major partnership with OpenAI covering AWS infrastructure, Trainium capacity, enterprise agents and customized models for Amazon products.

This combination suggests Amazon is pursuing a two-part strategy: build a smaller number of competitive internal models while continuing to profit from the infrastructure used to train and operate models from other companies.

A more focused attempt to compete

The decision to reduce investment in several Nova models reflects the growing cost and difficulty of competing at the frontier of AI development.

Training advanced models requires enormous computing clusters, specialized talent and continuous investment. Maintaining separate text, image, video and multimodal systems can divide those resources and make it harder to match competitors that concentrate on a smaller number of flagship products.

Amazon’s new strategy is intended to reduce that fragmentation.

The company still needs to prove that its next frontier model can attract developers and enterprise customers. But narrowing its portfolio may give Amazon a better chance of creating a model capable of competing at the highest level while preserving its stronger position as a provider of AI infrastructure.

Amazon Productivity