Amazon is winding down four of its flagship Nova artificial intelligence models by September 2026, redirecting resources toward a new frontier-model research effort instead of competing mainly on price and efficiency. The company launched Nova in 2025 as a budget-friendly alternative to OpenAI's GPT series, Google's Gemini, and Anthropic's Claude, but the rapid pivot signals a major strategic shift in how Amazon approaches AI development.

Which Amazon Nova Models Are Being Retired?

Amazon is deprecating four core Nova models that represented the lineup's flagship tier. According to AWS Bedrock's official model lifecycle documentation, these models are transitioning to "Legacy" status before reaching end-of-life dates in September 2026.

• Nova Premier: Amazon's largest and most capable text and reasoning model, scheduled for end-of-life on September 14, 2026.
• Nova Canvas: The image generation model, set to be retired on September 30, 2026.
• Nova Reel: The video generation model, also ending on September 30, 2026.
• Nova Omni: A multimodal model designed to handle text, images, and other formats simultaneously, named in reporting as being wound down but without a published end-of-life date on AWS's public documentation.

The retirement timeline is unusually aggressive. Under AWS policy, models must remain on the platform for at least 12 months before reaching end-of-life status, and they spend a minimum of six months in Legacy status first. New customers cannot start using a Legacy model at all, and existing customers risk losing access after 15 days of inactivity.

Why Is Amazon Abandoning Nova So Quickly?

Amazon's official statement frames the move as resource reallocation rather than retreat. "We're continuing to invest in and support the Nova models our customers rely on today. We're also investing in the next generation of frontier model research," an Amazon spokesperson told Reuters. The surviving models, including Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic, and Nova Forge, will continue receiving updates.

The underlying reason appears to be competitive pressure. Throughout 2025, Amazon positioned Nova as a cost-effective alternative to frontier models, emphasizing latency, customization, and agentic capability rather than raw performance. That strategy worked when Nova competed mainly on price within AWS Bedrock. By mid-2026, however, rivals like Google and Anthropic shipped models that were both cheaper and more capable, making Amazon's price-focused positioning harder to defend.

Concentrating resources on a single frontier-model effort is a bet that Amazon can close the capability gap instead of continuing to compete mainly on cost. This mirrors the strategy already pursued by OpenAI, Google, and Anthropic, which each maintain one primary model family rather than half a dozen parallel product lines.

What Does This Mean for Businesses Using Nova?

For companies running production workloads on Nova through AWS Bedrock, the shift requires immediate migration planning. Customers relying on Nova Premier, Canvas, Reel, or Omni will need to transition to alternative models before the September end-of-life dates arrive.

The speed of this pivot is notable. It is unusual for a company the size of Amazon to publicly wind down flagship products barely 18 months after launch, especially in a category where every major cloud provider is racing to prove its in-house models can stand next to OpenAI, Google, and Anthropic. The move signals that Amazon believes the only viable path forward is building models that match frontier performance, not just offering cheaper alternatives.

How to Plan Your Migration Away from Nova Models

• Audit Current Workloads: Identify which Nova models your team is using in production and document their specific use cases, including text generation, image creation, video production, or multimodal tasks.
• Evaluate Replacement Options: Test surviving Nova 2 models (Lite and Pro for text, Sonic for speech) or explore third-party alternatives on Bedrock from providers like Anthropic and Cohere to find the best fit for your performance and cost requirements.
• Plan Migration Timeline: Schedule your transition before the September 14 and September 30 end-of-life dates to avoid service disruptions; note that Legacy models restrict new customer access immediately.
• Monitor AWS Announcements: Watch for updates on Amazon's new frontier-model research effort, which may eventually offer a more capable in-house alternative to current Nova offerings.

The broader context matters too. Anthropic, Cohere, and AI21 Labs all have their own models cycling through Legacy and end-of-life status on Bedrock right now, so version churn on the platform is routine. What makes the Amazon Nova case different is that these are Amazon's own proprietary models, not a third party's, and four flagship products are exiting at once rather than one older version being swapped for a newer point release.

Amazon's pivot reflects a hard truth in the current AI landscape: competing on efficiency and cost alone is no longer enough. The company is betting that it can build frontier-class models that match the capability of GPT-4, Claude, and Gemini, and that doing so requires concentrating engineering and compute resources rather than maintaining a broad catalog of specialized models. Whether that bet pays off will shape how Amazon positions itself in enterprise AI over the next two years.