Amazon is undergoing a significant reorganization of its artificial intelligence division, shifting away from a wide-ranging portfolio of in-house models toward a more concentrated strategy. While competitors like OpenAI, Anthropic, and Google have established household names for their flagship models, Amazon’s Nova series has struggled to achieve similar prominence. Following the perceived failure of several power-hungry large language models (LLMs) to meet expectations, the company is now gutting much of its internal AI lab infrastructure to narrow its focus on high-priority goals.
As part of this overhaul, Amazon is deprecating many of its flagship Nova models, including high-end text, image, and video-generation versions. These models are being moved to a "keep the lights on" (KTLO) status—an engineering designation meaning they will receive only the minimum resources required to remain functional for existing customers rather than receiving active development. This shift follows the closure of the company's AGI Lab, a research group established in 2024 that was intended to pursue long-term research into artificial general intelligence (AGI), the point at which AI achieves human-level reasoning.
The company is instead diverting its primary engineering talent and scarce computing power toward a new initiative called Frontier Model Research (FMR). Led by Pieter Abbeel—a robotics expert who joined Amazon following the acquisition of the robotics startup Covariant—FMR is tasked with developing a new flagship foundation model. This pivot suggests a move toward integrating AI more deeply with robotics and automation, aligning with Amazon’s broader corporate goal of automating logistics and warehouse operations. While the company maintains a support structure for the older Nova models to ensure customer stability, the new strategy prioritizes a singular, high-impact frontier effort over a diverse but less successful model lineup.
The article discusses Amazon's pivot toward robotics and automation in its new Frontier Model Research initiative.