Figure Inks Multi-Billion-Dollar Compute Deal With Nscale to Deploy 100,000 Next-Gen NVIDIA GPUs

Key Takeaways

Show all

• Figure AI has formed a multi-year strategic partnership with Nscale to deploy up to 100,000 GPUs on NVIDIA's next-generation Vera Rubin platform.
• The deal represents an initial compute commitment of $3.5 billion, with explicit plans to scale beyond $6 billion.
• Deployments are scheduled to begin in the second half of 2027 at Nscale's facility in Barstow, Texas.
• The compute capacity is earmarked to train Helix, Figure's proprietary foundation model, complementing its recently unveiled Index data-crowdsourcing initiative.
• As part of the pact, Nscale is taking a strategic equity stake in Figure, while both companies explore integrating humanoid units directly into Nscale's data center supply chain.

Humanoid robotics has reached an inflection point where progress is no longer governed solely by mechanical engineering, but by raw computational scale.

In one of the largest infrastructure commitments to date for embodied artificial intelligence, California-based startup Figure AI announced a major strategic partnership with AI cloud provider Nscale. Under the multi-year pact, Figure plans to deploy up to 100,000 GPUs built on NVIDIA’s upcoming Vera Rubin architecture.

The agreement carries an initial financial commitment of $3.5 billion in compute capacity, with provisions to scale beyond $6 billion over time. The hardware rollout is targeted to begin in the second half of 2027 at Nscale’s facility in Barstow, Texas. Alongside the cloud allocation, Nscale has made an undisclosed strategic investment in Figure, with both companies exploring operational deployments of Figure’s humanoids within Nscale's own infrastructure supply chains.

Moving Beyond Hardware Bottlenecks

The capital commitment reflects an aggressive strategic pivot. While Figure spent early development cycles refining actuators, modular assembly lines, and whole-body balance routines—scaling production past its 1,000th Figure 03 build—leadership has maintained that hardware manufacturing is rapidly commoditizing.

Instead, the limiting factor in autonomous physical labor has shifted toward onboard intelligence and general-purpose reasoning.

"Figure is entering a phase where we are largely bound by data and compute needed to train Helix," the company stated in its announcement, referring to the end-to-end AI architecture powering its machines across three hardware generations. "To ship a robot into every home, we need a massive amount of compute."

The move arrives on the heels of Figure's launch of Index, an expansive consumer data engine designed to crowdsource real-world video from tens of thousands of global contributors. While Index is generating roughly 35 minutes of human interaction footage every second, converting that firehose of unstructured video into robust motor policies requires a commensurate leap in processing power.

"Helix becomes more capable the same way every learned system does: with more data and compute," Figure founder and CEO Brett Adcock said. "Looking back, this will be a key inflection point to putting a robot into every home."

Activating the "Physical AI Flywheel"

The deal also deepens Figure’s reliance on NVIDIA’s hardware and software stack, particularly as frontier labs and hardware makers increasingly collide. Following Figure's public split with OpenAI—which CEO Brett Adcock attributed to his belief that Figure’s internal AI team had outpaced the frontier lab—securing sovereign compute outside traditional AI lab alliances has become essential. NVIDIA founder and CEO Jensen Huang characterized the agreement as a full activation of the "physical AI flywheel." The loop links training Figure’s models on Vera Rubin clusters via Nscale’s infrastructure, validating physical loco-manipulation in NVIDIA Isaac Sim, and deploying the optimized policies onto onboard NVIDIA GPUs nestled inside Figure's robots.

Nscale CEO Josh Payne framed the partnership as the logical progression of cloud infrastructure demands. "We've seen incredible growth with inference and agentic AI, and Figure is pushing the boundaries of AI even further," Payne said.

A Distant Runway for Next-Gen Scale

While the headline figures are massive, the timeline underscores the long developmental horizon for general-purpose physical agents.

Targeted for late 2027, the deployment of NVIDIA's Rubin-generation silicon means that Figure's immediate training runs will still depend on existing Hopper and Blackwell-class clusters. Procuring 100,000 Rubin GPUs in Barstow, Texas will also require immense power generation, cooling capacity, and capital allocation over the next three years—demands that have occasionally constrained data center buildouts across the broader tech industry.

Furthermore, whether scaling compute and passive video ingestion alone can overcome the deep nuances of contact-rich physical dynamics remains an open debate across robotics research. As developers continue to grapple with actuator placement and fine-grained hand dexterity, raw parameter count cannot always replace physical compliance and sensor fidelity.

Even so, by pledging several billion dollars toward frontier silicon, Figure is making its thesis unmistakable: embodied robotics will not be solved through incremental algorithmic tuning, but through hyperscale computing muscle.