On 3 September Figure and Nscale announced a partnership for NVIDIA Vera Rubin with up to 100,000 GPUs, with initial deployment targeted from the second half of 2027 in Barstow, Texas. The initial commitment is $3.5 billion of compute, with intent to exceed $6 billion. Nscale also comes in as an investor in Figure.
- Up to 100,000 GPUs on Vera Rubin, initial deployment from the second half of 2027.
- $3.5 billion initial commitment, with intent to go above $6 billion; Nscale also makes a strategic investment.
- Figure writes that it is now largely bound by data and compute for training Helix.
Every second Index, Figure's humanoid training data effort, generates thirty-five minutes of data. That is the company's own number.
And it is still not enough. In the announcement of 3 September the company writes that it is now largely bound by data and compute, and that data alone will not solve the problem.
Notice the loop. Nscale sells Figure compute, invests in Figure and will explore using its robots in its own supply chain. The money goes out one way and comes back the other two. That is not necessarily bad. But it is good to see it when you read the figure of 6 billion.
Almost a year remains until the second half of 2027. Until then Helix will learn on what it has, and the test is simple: whether the robots in factories do more work, not whether the contracts grow.