Particle.news

Nscale Pledges $3.5 Billion in GPU Compute to Humanoid Robot Maker Figure

The deal pairs a strategic equity stake with massive GPU capacity to fund the large-scale training runs robotics developers say are far more compute-intensive than language models.

Overview

  • On Thursday Nscale announced a multi-year commitment to supply at least $3.5 billion in AI cloud and GPU capacity to Figure, with the agreement able to scale past $6 billion and making Nscale a shareholder and preferred compute provider.
  • The companies plan to deploy up to 100,000 NVIDIA Vera Rubin GPUs, with initial hardware installations targeted to begin in the second half of 2027 at Nscale’s Barstow, Texas site.
  • Figure says the capacity will power its Helix embodied-AI models that train on continuous, multi-modal sensor streams for real-time motor control, a workload the companies and analysts describe as often more compute-heavy than training large language models.
  • The partnership includes reciprocal operational plans such as exploring the use of Figure’s humanoids in Nscale’s data center supply chain and a closed loop of training, simulation with NVIDIA Isaac Sim, and deployment on NVIDIA hardware.
  • The deal highlights a wider 2026 surge in funding and infrastructure for humanoid robotics but also raises questions about the economics and governance of vendor-investor ties and whether commercial robot deployments can soon cover these large upfront compute costs.