Amazon is winding down most of its Nova AI models, shifting resources to a single next-generation system as it seeks to compete with OpenAI and Google. The move signals a major reset for AWS customers and investors watching AI spending
Amazon is making a dramatic shift in its artificial intelligence strategy, discontinuing most of its Nova family of AI models less than two years after their high-profile launch. The company is now consolidating its AI development around a single, large-scale model, aiming to better compete with industry leaders like OpenAI, Anthropic, and Google. This overhaul marks Amazon's most significant AI reset since it introduced the Nova lineup at re:Invent in December 2024, according to Reuters.
The Nova family originally included Nova Premier, Nova Omni, Nova Reel, and Nova Canvas, each targeting different AI applications from text to video and image generation. These models were pitched as affordable, capable alternatives for AWS customers, signaling Amazon's ambition to be seen as more than just a cloud infrastructure provider. But as of mid-2026, Amazon is quietly winding down development on most of these models. Nova Premier and Nova Omni, the company's high-end foundation models, along with Nova Reel and Nova Canvas, have been placed into "keep the lights on" (KTLO) status. This means they will remain available and supported for existing customers but will no longer receive significant updates or new features.
Only a handful of Nova offerings remain in active development. Nova 2 Lite and Nova 2 Sonic continue to be supported, while Nova Forge, which allows customers to build and customize their own models, and Nova Act, Amazon's AI agent platform, are still being developed. The company says it is reallocating resources to a new internal group called Frontier Model Research (FMR), led by Pieter Abbeel, a UC Berkeley professor who joined Amazon through its acquisition of Covariant in 2024. FMR is now the top priority within Amazon's AGI (artificial general intelligence) organization, which itself has been restructured and now reports to Peter DeSantis, a longtime Amazon executive overseeing cloud, custom silicon, and quantum computing initiatives.
Strategic Reset
The decision to scale back the Nova lineup reflects both market realities and internal challenges. Despite initial ambitions, Amazon's models failed to gain the same brand recognition or industry benchmarks as those from OpenAI or Google. While AWS customers used Nova models for enterprise workloads, they did not become a central story in the broader AI race. Maintaining multiple model families across text, image, audio, and video proved costly and diluted engineering focus, according to reporting by TheStreet.
Amazon's move comes as the company faces rising AI infrastructure costs and increased scrutiny from investors. The company's investment in Anthropic, its custom AI chips (Trainium and Inferentia), and its cloud services for AI workloads have performed well, but its own foundation models have struggled to stand out. The closure of Amazon's AGI Lab, which was built in 2024 after hiring much of the team from AI startup Adept, underscores the company's pivot. Remaining staff have been absorbed into FMR or reassigned elsewhere.
What's Next for AWS Customers and Investors
For AWS customers currently relying on Nova Premier, Nova Omni, Nova Reel, or Nova Canvas, the KTLO designation means their existing workloads will remain supported, but they should not expect further improvements. Customers seeking Amazon's most advanced AI capabilities will eventually need to migrate to whatever new model emerges from the FMR group, which is expected to debut its flagship system at re:Invent in late 2026. This new approach will focus on a single, frontier-scale model designed to compete directly with the latest offerings from OpenAI and Google, rather than maintaining a broad portfolio of specialized models.
For investors, the shift signals a more disciplined allocation of resources. Rather than spreading engineering talent and compute power across a dozen model families, Amazon is concentrating its efforts on building a single, highly competitive system. This could help the company better manage costs and improve its chances of delivering a model that can serve as a true benchmark in the AI space. With Amazon set to report earnings on July 30, CEO Andy Jassy is likely to face questions about the AI reorganization, the timeline for the new model, and how the company plans to compete at the frontier of AI development.
Amazon's AI overhaul comes at a time when tech giants are under pressure to justify heavy AI spending. In a related development, Amazon's recent Prime Day results showed record sales but also highlighted shifting consumer behavior and rising costs, adding to the scrutiny on the company's broader strategy.
According to Amazon's most recent quarterly filing, the company's AWS segment generated $25 billion in revenue for the first quarter of 2026, up 13% year-over-year. However, operating income for AWS was pressured by increased investment in AI infrastructure and research, reflecting the high costs associated with developing and maintaining advanced AI models. Industry analysts estimate that training a single frontier-scale AI model can cost hundreds of millions of dollars, making strategic focus and resource allocation critical for long-term competitiveness.
Amazon's decision to consolidate its AI efforts highlights the trade-offs companies face when balancing innovation with operational efficiency. Building and maintaining multiple specialized AI models can offer flexibility and address diverse customer needs, but it also increases complexity and costs. By shifting to a single flagship model, Amazon is betting that a focused approach will yield better results in a market where only a handful of models set the standard for performance and adoption. For AWS customers, this means monitoring Amazon's next moves closely, as future AI capabilities and support will increasingly depend on the success of the new frontier model.
