Figure and Nscale Strike $3.5 Billion NVIDIA Vera Rubin GPU Deal for Humanoid Robotics

Picture a robot in every home. That’s the stated ambition of Figure, the humanoid robotics company that has just announced one of the largest compute commitments in the short history of AI-driven robotics. To get there, it’s going to need an enormous amount of raw computing power — and it’s just found its supplier.

On 3 September 2026, AI infrastructure company Nscale announced it had signed a multi-year strategic partnership with Figure, covering the deployment of up to 100,000 NVIDIA GPUs on NVIDIA’s next-generation Vera Rubin platform. The initial compute commitment stands at around £2.7 billion (announced as $3.5 billion), with plans to scale that figure to beyond around £4.7 billion ($6 billion). The first deployment is targeted to begin in the second half of 2027 at a site in Barstow, Texas.

It’s a deal that goes well beyond a straightforward hardware order.

More Than a Purchase Order

Nscale isn’t simply selling compute to Figure. As part of the agreement, Nscale will become a Figure shareholder and take on the role of Figure’s preferred compute provider. That combination — infrastructure supply, equity stake, and preferred-provider status — gives the partnership a structural depth that sets it apart from a typical vendor relationship.

The arrangement reflects how seriously both companies are treating the long-term compute demands of humanoid robotics. Training robots to move, reason, and respond in unpredictable real-world environments requires a scale of AI computation that dwarfs most other applications. And that demand is only going to grow.

What Is the Vera Rubin Platform?

NVIDIA’s Vera Rubin platform is the company’s next generation of AI computing infrastructure, positioned for what NVIDIA describes as agentic AI and large-scale reasoning workloads. Put simply: it’s built for the kind of complex, continuous AI tasks that teaching a robot to handle a kitchen or a factory floor would demand.

NVIDIA has named Nscale among its cloud partners expected to deploy Vera Rubin-based instances, with partner availability from the second half of 2026. The Figure deal, targeting deployment from the second half of 2027, sits within that broader rollout window — though the scale of 100,000 GPUs is at the ambitious end of what’s been publicly announced for any single partnership.

Nscale’s chief executive Alishah Chaudhry said: “This partnership reflects our belief that the next wave of AI will be physical, and that Figure is building the foundation for it. We’re proud to be their compute partner and a shareholder in that future.”

The Scale of the Bet

One hundred thousand GPUs is a number worth sitting with for a moment. For context, many significant AI deployments run on clusters numbering in the thousands. Getting to six figures requires not just hardware procurement but hefty site preparation, power infrastructure, and cooling capacity — all of which take time and capital to deliver.

The Barstow, Texas site will need to be built out to accommodate that kind of load. Power availability, in particular, is a genuine constraint for hyperscale AI infrastructure projects across the United States right now. The timeline — with initial deployment starting in the second half of 2027 — reflects that reality.

So the deal is best understood as a commitment and a direction of travel, rather than a facility that’s ready to switch on tomorrow. That said, the financial scale of the commitment, with around £2.7 billion already pledged, gives it considerably more weight than a letter of intent.

Figure’s Bigger Picture

Figure has been open about its goal: putting a robot into every home. That’s an enormous ambition, and one that requires the company to train increasingly capable humanoid models at scale. The Nscale partnership is framed by Figure as directly supporting that mission — giving it the compute headroom to develop the next generation of its robotics models without being constrained by infrastructure.

The deal also signals where the broader robotics industry is heading. The competition to build capable humanoid robots — from Figure, Boston Dynamics, Tesla’s Optimus programme, and others — is increasingly a competition over who can train the best models, fastest. And that means the race for compute is, in many ways, the race itself.

NVIDIA’s Vera Rubin platform sits at the centre of that contest. By locking in a preferred-provider relationship with Nscale and its access to Vera Rubin infrastructure, Figure is making a clear statement about how it intends to compete.

Whether the deployment delivers on its full potential will depend on execution — site build-out, power contracts, and the pace at which Figure’s robotics models actually require that level of scale. But the ambition, and the money behind it, is not in question.

What This Means for Kent Residents

There’s no direct operational impact on Kent from this announcement — no local site, no Kent-based contracts have been named. But for UK residents watching the AI and robotics sector, deals of this scale shape the trajectory of technology that will eventually reach British homes and workplaces. As humanoid robotics moves closer to commercial deployment, the policy and regulatory conversations happening in Westminster about automation and its effects on UK jobs will only become more pressing.