Amazon is overhauling its artificial intelligence strategy, phasing out most of its in-house flagship Nova models and reorganizing its Artificial General Intelligence division after a round of job cuts. The restructuring, first reported by Business Insider on July 28, 2026, signals a significant pivot for the tech giant as it attempts to compete at the frontier of AI development. For comprehensive coverage of the AI industry, see our latest AI business news.

Nova models put on life support

Multiple sources told Business Insider that Amazon has started phasing out the majority of its in-house Nova model lineup, including the high-end Nova Premier, Nova Omni, the Nova Reel video-generation model, and the Nova Canvas image-generation model. Some employees described these models as being in "KTLO" mode — an engineering shorthand for "keep the lights on" — meaning the software continues to support existing customers but is no longer a primary development priority.

Amazon disputed the characterization in a statement provided to PYMNTS. The company said it continues to invest in and support the Nova models, including Nova Forge, a platform that allows businesses to build their own custom models. "As with any AI portfolio, we continually evolve our model lineup based on what customers need, and we always provide customers clear guidance and migration paths as models advance," Amazon said. "We're also investing in the next generation of frontier model research, and Amazon's commitment to building advanced AI models is as strong as it's ever been."

AGI Lab closure and layoffs

The restructuring follows job cuts the previous week at Amazon's AGI division. According to reporting from the Wall Street Journal, Hindustan Times, and other outlets, Amazon laid off employees within its Artificial General Intelligence unit as it sharpened its focus on key AI initiatives. The cuts also included the closure of AGI Lab, a research group Amazon launched in 2024 after acquiring most of the team from AI startup Adept.

The layoffs come months after Amazon's broader workforce reduction that eliminated approximately 16,000 positions, making the AGI cuts a targeted continuation of its cost discipline even as the company dramatically increases its overall AI spending.

A new frontier-model effort under Pieter Abbeel

According to Business Insider's sources, Amazon is shifting resources away from the Nova models and toward a new frontier-model effort overseen by researcher Pieter Abbeel. Abbeel joined Amazon through its acquisition of Covariant, an AI robotics startup he co-founded. A prominent figure in reinforcement learning and robotics, Abbeel is now tasked with leading Amazon's push to build models capable of competing directly with those from OpenAI, Anthropic, and Google.

The reorganization reflects a broader strategic calculation. Rather than maintaining a wide portfolio of text, image, and video models — an approach Amazon has pursued since unveiling the Nova family — the company is concentrating talent and computing resources on its largest priorities. Sources told Business Insider that engineers are being redirected toward the new frontier-model strategy.

Amazon's competitive position in AI

The pivot underscores the competitive pressure Amazon faces in the generative AI market. While Amazon Web Services remains the dominant cloud provider, the company's first-party AI models have struggled to match the capabilities of frontier models from OpenAI's GPT family, Anthropic's Claude lineup, and Google's Gemini series. Amazon has relied heavily on its Bedrock platform, which offers customers access to third-party models from Anthropic, Meta, and others, alongside its own Nova models.

By consolidating its in-house AI research under a single frontier-model effort, Amazon appears to be betting that a focused approach will produce better results than spreading investment across multiple specialized models. The company's financial commitment to AI remains substantial — Amazon's capital expenditure continues to climb as it builds data center infrastructure to support both its own models and the cloud workloads of AI companies.

The decision to wind down most Nova models while continuing to support Nova Forge suggests Amazon sees more value in enabling customers to build custom models on its infrastructure than in maintaining a broad first-party model lineup. Whether the new frontier-model effort under Abbeel can close the gap with industry leaders remains to be seen.

A pattern of strategy shifts

The Nova model phase-out is not Amazon's first AI strategy reversal. The company's AGI Lab was launched in 2024 after Amazon absorbed most of the team from Adept, an AI startup focused on building AI agents that could automate software tasks. That acquisition was part of a broader talent grab in which Amazon hired key Adept engineers rather than acquiring the company outright — a structure designed to avoid antitrust scrutiny. The subsequent closure of AGI Lab less than two years later underscores the difficulty even well-resourced companies face in building competitive frontier models from scratch.

Amazon's AI journey has been marked by similar pivots. The company was an early leader in machine learning through its AWS offerings and recommendation systems, but it was caught flat-footed by the generative AI wave launched by OpenAI's ChatGPT in late 2022. Since then, Amazon has invested billions in Anthropic, integrated AI across its retail and cloud operations, and rushed to build its own model lineup — only to now scale much of it back.

The broader industry context adds pressure. Microsoft, Amazon's primary cloud rival, reported strong quarterly earnings driven by Azure AI revenue, though it also took a writedown on its OpenAI stake. Meta's profit fell as AI infrastructure spending surged. And Google continues to invest heavily in Gemini models. For Amazon, the question is whether concentrating its AI bets under a single frontier-model team will produce results that justify the resources being poured into them.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →