While global tech giants frantically escalate generative AI investments, Amazon has chosen a starkly different path. Indeed, the company is contracting its battle lines, consolidating resources, and eliminating non-core teams. According to recent reports, Amazon overhauls its AI strategy by phasing out most Nova models and restructuring its divisions. Furthermore, the tech giant is executing widespread layoffs within its Artificial General Intelligence (AGI) organization. Consequently, Amazon is officially closing the AGI lab it established just last year in 2024.
Peter DeSantis, the senior vice president who assumed control of the AGI division last December, is spearheading this massive restructuring. This Amazon veteran also manages custom silicon and quantum computing operations. Currently, he is shifting the company’s AI resources from a scattered approach to a highly targeted strike. Therefore, he aims to consolidate engineering talent and scarce computational resources. Ultimately, this strategy focuses on developing a select few truly competitive frontier models.
The KTLO Phase: Nova Flagship Models Face Retirement
Multiple insiders revealed that Amazon has initiated the gradual phase-out of most Nova flagship models. Specifically, this retirement impacts Nova Premier, the high-end flagship model. Additionally, it affects Nova Omni, the multimodal reasoning model. Furthermore, Nova Reel and Nova Canvas, the video and image generation models, share this exact fate.
Internal employees describe these models as entering a “Keep The Lights On” (KTLO) state. Essentially, the company continues to support existing customers, but these models are no longer primary development priorities. In other words, Amazon will not immediately terminate these services. However, the company has officially stopped investing significant resources into iterative upgrades.
Last year, Nova Omni 2 debuted with immense fanfare at the re:Invent conference. Now, merely eight months later, it faces this exact same fate. Consequently, this demonstrates a fundamental shift in the strategic judgment of Amazon’s executive leadership.
New Strategic Core: Frontier Model Research
While systematically withdrawing resources from existing Nova models, Amazon is pivoting toward a brand-new initiative. This new focus is the Frontier Model Research (FMR) program. Pieter Abbeel, a renowned AI professor from UC Berkeley, leads this ambitious project. He joined Amazon after the tech giant acquired Covariant, the AI robotics startup he co-founded.
According to insiders, the Frontier Model Research initiative has become the absolute highest priority for Amazon’s AI division this year. Currently, this specialized team is developing a revolutionary flagship foundation model. Expectedly, this new model will officially debut at this year’s re:Invent annual conference. Interestingly, this upcoming model might still launch under the “Nova” brand name. Therefore, this suggests Amazon is not entirely abandoning the Nova identity, but rather redefining and relaunching it.
Organizational Restructuring: Lab Closures and Leadership Changes
Looking back at the evolution of Amazon’s AI organization, this adjustment represents a long-brewing strategic consolidation. In 2023, Amazon established the AGI organization under the leadership of Rohit Prasad. Initially, their goal was to build foundation models capable of powering future AI products. Subsequently, in 2024, the company established the AGI lab within the broader organization.
However, in December 2025, Rohit Prasad departed the company. Consequently, Peter DeSantis assumed command of the AGI organization. He successfully integrated it with the custom silicon and quantum computing divisions. Then, in February 2026, Adept co-founder David Luan officially left Amazon. Finally, in July 2026, Amazon formally closed the AGI lab and executed layoffs across the AGI organization.
An insider noted that the AGI organization previously pursued a multi-model family strategy under Prasad’s leadership. This broad approach covered text, image, and video generation simultaneously. Conversely, since DeSantis took charge, he has chosen to concentrate top talent and expensive computing resources. Thus, he strictly targets frontier models that possess genuine market competitiveness.
Executive Mindset Shift: Doing the Right Things
In 2023, Amazon CEO Andy Jassy personally positioned the AGI organization as the core team responsible for building foundation models. Furthermore, he established six new research groups specifically for this overarching purpose. Yet today, less than three years later, this exact organization faces severe contraction.
This dramatic pivot likely reflects a universal phenomenon across the entire AI industry. The sector is transitioning from an exploratory phase into a highly pragmatic era. As generative AI business models become clearer, enterprises are reaching a crucial realization. Rather than casting a wide net to develop ten mediocre models, it is better to concentrate firepower on one truly competitive product.
For Amazon, AWS remains the world’s largest cloud service provider. Therefore, the strategic significance of AI models extends far beyond end-user applications. Crucially, the goal is to attract enterprise customers to train, deploy, and run AI workloads on the AWS platform. Unquestionably, a single, exceptionally powerful flagship model holds far more commercial value than ten average alternatives.
Market Observations: AI Arms Race Enters Consolidation
Amazon’s strategic realignment serves as a microcosm of the generative AI industry’s broader shift. The market is moving from frenzied investment toward rational consolidation. Over the past two years, major tech giants have fiercely invested tens of billions of dollars into developing large language models. However, the market is now scrutinizing the return on AI investments much more strictly.
Consequently, the wide-net strategy is gradually yielding to highly targeted, precision-strike approaches. Amazon has actively chosen to consolidate its battle lines right now. The company is concentrating its vast resources on the Frontier Model Research program led by Pieter Abbeel. Simultaneously, it is systematically phasing out most of the existing Nova series models. Whether this bold decision allows Amazon to catch up in the fierce AI race remains uncertain. Ultimately, the industry awaits the unveiling of the new model at this year’s re:Invent conference.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
