Understand this story
Figure has released a one-hour video of an internal all-hands meeting. Founder and CEO Brett Adcock disclosed in the video that the fourth-generation robot, Figure 04, has completed design lock, with components entering the delivery and validation cycle; he called it the largest improvement of any generation to date. The Index global data collection project supporting model iteration has accumulated 37.5 million human manipulation videos, at a collection rate of about 81 minutes of video per second, with weekly data spending reaching $8.1 million. Figure has also finalized a compute partnership with Nscale at an initial scale of $3.5 billion, planning to deploy up to 100,000 GPUs based on NVIDIA's Vera Rubin platform, going live in the second half of 2027 at a site in Bastrop, Texas. Adcock said the team's new model, trained with the largest-scale pretraining data and GPU resources to date, has completed testing in an internal lab codenamed Black Box, but he did not provide evaluation details.
AI-generated from public reports
Developments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 31
East Money
Figure All-Hands Meeting Reveals Roadmap: Figure 04 Enters Component Manufacturing, Betting on Large-Scale Pretraining and $3.5 Billion in Compute
Figure has released a one-hour internal all-hands meeting video. Founder and CEO Brett Adcock disclosed that the fourth-generation robot Figure 04 has completed design lock and that its components have entered the delivery and validation cycle, calling it the largest improvement of any generation. The Index global data collection project supporting model iteration has accumulated 37.5 million human manipulation videos, collecting at a rate of about 81 minutes of video per second, with weekly data spending reaching $8.1 million; Figure has also finalized a compute partnership with Nscale at an initial scale of $3.5 billion, planning to deploy up to 100,000 GPUs based on Nvidia's Vera Rubin platform, going live in the second half of 2027 at the Bastrop, Texas site. Adcock said the team's new model, trained with the largest-scale pretraining data and GPU resources to date, has completed testing in the lab internally codenamed Black Box, but he did not provide evaluation details.