2 weeks ago
11
Training a humanoid robot to navigate the real world turns out to require an absurd amount of computing power. Nscale, the UK-based AI cloud and data center company, just agreed to supply at least $3.5 billion worth of that power to Figure, the robotics startup building general-purpose humanoids.
The deal also includes an undisclosed strategic investment by Nscale into Figure itself, tying the two companies together in a way that goes beyond a simple vendor-customer relationship. The commitment could scale to more than $6 billion over time, covering the planned acquisition of up to 100,000 NVIDIA Vera Rubin GPUs.
What the deal actually looks like
The partnership centers on deploying massive GPU clusters at a facility in Barstow, Texas, with hardware installations expected to begin in the second half of 2027. Those Vera Rubin chips represent NVIDIA’s next-generation architecture, purpose-built for the kind of dense AI workloads that training embodied intelligence demands.
Figure’s Helix AI model handles perception, reasoning, and motor control simultaneously.
Two companies on a tear
Figure, founded in 2022 by Brett Adcock, has raised more than $2 billion in total funding. Its Series C round valued the company at $39 billion, with capital from investors including NVIDIA, Microsoft, and OpenAI.
Nscale closed its own $2 billion funding round in March 2026, landing at a $14.6 billion valuation. The company recently locked in a $45 billion, six-year agreement with Anthropic to provide AI computing power at its West Virginia campus, which boasts over 1.35 gigawatts of planned capacity.
Why robotics needs its own compute layer
The AI industry has spent the last few years building infrastructure primarily for text and image generation. Robotics represents a fundamentally different computational challenge.
Language models process tokens, essentially chunks of text. Robotics models process continuous streams of sensor data: cameras, lidar, force sensors, joint encoders. They need to make decisions in real time, with physical consequences for getting things wrong.
Figure’s ambition to deploy humanoids across various real-world environments, from warehouses to manufacturing floors, means the demand for this specialized compute will only grow as the company moves from prototype to production. Every new deployment scenario requires additional training data and simulation cycles.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·