Descriptions:
Bloomberg Technology sits down with the CEO of Luma AI to discuss the company’s launch of a Physical AI Lab—an open-science initiative targeting one of robotics’ most stubborn problems: generalization. Unlike large language models, which learn to handle diverse and novel tasks from a single training run, today’s robots are trained one task at a time, creating what the CEO calls a “valley of specific tasks.” Luma’s proposed solution draws on four years of work building general systems from internet-scale multimodal data, using that foundation to extract control signals that can simulate physical reality and drive robotic behavior.
The open-science and open-source commitment is central to the lab’s identity, and the CEO frames it as both an ethical imperative and an economic argument: physical AI will eventually permeate homes, hospitals, factories, and public infrastructure, making single-company control of that stack geopolitically and economically untenable. Nations, the CEO argues, will not accept their means of production being controlled by one or two foreign corporations—making an open ecosystem model not just philosophically preferable but commercially viable. The lab is designed around partnerships with chip providers and model infrastructure companies.
Luma AI, which recently closed a $900 million Series C, is positioning its multimodal generative AI foundation—best known publicly through its video generation work—as a unique asset in the physical AI race, distinguishing itself from robotics-native competitors like World Labs. The segment is a concise but substantive introduction to the company’s new research direction and its broader vision for distributed, open physical intelligence.
📺 Source: Bloomberg Technology · Published June 01, 2026
🏷️ Format: Interview
