Figure bets on 100,000 GPUs via Nscale deal

Figure's announcement pairs its need for compute with its effort to gather training data through Index. The company identifies both as constraints on Helix, and the Nscale deal is meant to address the compute side. The plan also includes exploring using humanoids in Nscale's supply chain, though no order was announced.

The financial commitment is substantial, with an initial $3.5 billion and potential to go beyond $6 billion. But the timeline stretches far into the future, with initial deployment slated for the second half of 2027. In the meantime, Figure's own data shows growth in its Index collection, yet the company has not released detailed metrics on how this improves robot performance.

Research cited in the article suggests skill improvements from scaling may be bounded, especially within fixed settings. This makes the composition of Index's collection important. Whether the promised scale of compute and data can deliver reliable performance across varied environments remains an open empirical question, one that hinges on controlled comparisons and practical testing of the robots themselves.