Figure AI agrees to multi billion dollar compute expansion for Helix models

Figure AI, a developer of physical automation, signed a deal valued at 3.5 billion dollars with computing provider Nscale, according to initial partnership disclosures. Under the agreement, the company will gain access to up to 100,000 advanced Nvidia Vera Rubin graphics cards. This massive processing infrastructure is designed to train Helix, the artificial intelligence system responsible for controlling the company's humanoid robots.

For non technical observers, this deal highlights the immense computational power required to teach machines how to operate in physical spaces. Teaching a robot to navigate cluttered rooms, manipulate tools, or react to human movement requires training on massive datasets. If this computational push succeeds, practical humanoid helpers could transition from experimental prototypes into actual factory workers and home assistants much faster.

Even so, hardware scale does not automatically solve physical engineering challenges. Teaching software in virtual environments is very different from managing physical hardware around real people, and it is unclear if raw computing strength can overcome real world mechanical constraints. Furthermore, the exact timeline for deploying these specialized chips at scale has not been publicly specified.