Figure’s Multibillion-Dollar Compute Deal Meets Nscale’s Data-Center Buildout Challenge

Figure’s multibillion-dollar compute agreement with Nscale is part of a much larger infrastructure expansion whose financing and construction remain unfinished, according to new reporting from The Information.

The report adds context to the humanoid maker’s ambitious training plans: contracting for future GPU capacity is one step; getting the data centers built, powered and equipped is another. It does not establish that Figure’s allocation has been delayed.

Key Takeaways

Show all

• The Information reports that nearly all of Nscale’s $103 billion in contracted commitments depend on data centers that have yet to be built, with further financing needed for its expansion.
• Figure’s agreement starts with a $3.5 billion compute commitment, with an intention to exceed $6 billion, and targets initial deployments in the second half of 2027.
• The report identifies risks across Nscale’s business but does not establish a financing shortfall or missed deadline specific to Figure’s capacity in Barstow, Texas.

The infrastructure behind Figure’s training plans

As we covered in our original report on the partnership, Figure plans to access up to 100,000 GPUs on NVIDIA’s Vera Rubin platform through Nscale. The capacity is intended to support training of Helix, the AI system behind its humanoids.

Figure’s September 3 announcement explicitly put initial deployment in Barstow in the second half of 2027. Nscale is also making a strategic investment in Figure, and the companies plan to explore using humanoids in Nscale’s supply chain.

Figure described data and compute as constraints on Helix’s development. Brett Adcock has since predicted that humanoids will ultimately need more data and compute than large language models, arguing that scaling robot intelligence and manufacturing will eventually require hundreds of billions of dollars. That is his broader investment thesis, rather than a disclosed Figure budget, but it helps explain the company’s appetite for a compute agreement of this scale.

Delivery of the planned infrastructure is therefore relevant to Figure’s longer-term ambitions, although its announcement did not establish how dependent its overall training program would be on this one agreement.

A large backlog, and substantial work ahead

The Information’s analysis of Nscale’s IPO plans describes a company whose contracted business is far larger than its current operations. According to the report, Nscale operates five data centers delivering roughly half a gigawatt of capacity, with another 12 contracted to take that total to 10 gigawatts.

Financing is a central issue. Citing Nscale’s IPO filing, the publication reports $24 billion in technology-equipment purchase commitments payable this year and next. It also reports substantial borrowing since June, including a recently announced $3.36 billion convertible-note raise. Those figures describe Nscale’s broader expansion, rather than the funding status of Figure’s particular project.

Nscale has attracted major backers. Its own March funding announcement disclosed a $2 billion Series C at a $14.6 billion valuation, with NVIDIA among the participants. Further investment and borrowing nevertheless remain important to turning its pipeline into operating capacity.

What this means for Figure

The robotics angle is the infrastructure dependency behind the promise of much larger training runs. A compute agreement can reserve future access, but delivery still depends on construction, electricity, equipment and capital arriving on schedule.

The Information does not report a missed Figure delivery milestone. Nor does it establish Figure’s cancellation rights, alternative capacity arrangements or whether the Barstow allocation is fully financed. Contract protections described for another Nscale customer should not be assumed to apply to Figure.

For readers following Figure’s progress, the useful milestones remain concrete: financing and construction progress at Barstow, power availability, and delivery against the initial second-half-of-2027 target. The new report makes those dependencies more visible; it does not show that Figure’s compute deal has failed.