Amazon Scales Back AI Division Amid Model Failures

Amazon is substantially curtailing its artificial intelligence operations after its large language models failed to deliver results. The company is placing most of its Nova model variants—including text generation, video, and image synthesis—into minimal maintenance status, freeing resources for a single robotics-focused initiative. This includes closing an 80-person artificial general intelligence research lab in San Francisco.

Leadership of the new priority project falls to Pieter Abbeel, director of Berkeley’s Robot Learning Lab, whose robotics company Covariant was acquired by Amazon in 2024. The shift concentrates resources on robotics applications aligned with Amazon’s push to automate its operations—a contrast to the company’s previous broad investment in general AI model development.