Summary
- Nscale and Figure have agreed an initial $3.5bn multi-year compute commitment.
- The relationship could expand beyond $6bn and potentially reach 100,000 Nvidia Vera Rubin GPUs.
- Initial deployment is targeted for H2 2027 at a location Nscale identifies as Barstow, Texas.
Nscale has signed a multi-year compute agreement with humanoid-robotics company Figure covering an initial $3.5 billion commitment and potentially scaling to more than $6 billion.
The London-based AI infrastructure company says the partnership could ultimately support deployment of up to 100,000 Nvidia Vera Rubin GPUs. Initial systems are targeted for the second half of 2027 at a US location that Nscale’s announcement identifies as Barstow, Texas.
The agreement goes beyond a conventional cloud-capacity contract. Nscale is becoming a shareholder in Figure, making a strategic investment in the robotics company, and has been named its preferred compute provider. Nscale is expected to provide power, the compute platform, and orchestration around the Nvidia infrastructure.
A commitment measured in tens of thousands of next-generation GPUs becomes an infrastructure programme long before those processors are installed. The machines require electrical capacity, network fabric, cooling plant, rack systems, controls, and a commissioning sequence able to bring the cluster online without leaving expensive hardware waiting for the surrounding facility.
The announcement does not provide a detailed electrical or mechanical specification for the US deployment, so the 100,000-GPU figure should not be read as an already commissioned data centre. It is a potential compute scale attached to a multi-year commercial agreement whose first deployment is still targeted for the second half of next year.
That gap between contracted compute and physical delivery is becoming increasingly important as AI infrastructure commitments grow. The processor order is only one part of the capital stack. High-density deployments can require redesigned power distribution, liquid cooling, larger network fabrics, and significant upstream grid or generation capacity. The availability dates of those systems have to align with the GPU schedule.
Nscale’s position in the deal reflects the changing role of specialist AI infrastructure providers. Rather than selling rack space alone, the company is combining energy, physical data centre capacity, hardware, and orchestration into a service intended to deliver usable compute to customers whose core business sits elsewhere.
For Figure, that transfers part of the infrastructure build problem to a specialist provider. Training and operating increasingly capable robotics models can create large and variable compute requirements, but owning a data centre portfolio would add an entirely different set of planning, power, construction, and operational responsibilities.
For Nscale, the agreement represents substantial demand visibility if the deployment reaches the scale described. It also creates execution exposure. Securing up to 100,000 Vera Rubin GPUs is only useful if the facilities, power systems, cooling equipment, and network architecture needed to operate them arrive on schedule.
The deal is a global comparator rather than a new UK data centre build, but the supplier is headquartered in London and is assembling an international infrastructure portfolio. Nscale lists owned or partner capacity across locations including the UK, Norway, Portugal, Iceland, and the United States.
The $3.5 billion initial commitment gives the project commercial weight, yet the most informative milestones will come later: the facility and power specification, the size of the first deployment tranche, commissioning dates, and whether the proposed scale above $6 billion moves from an option into contracted capacity.
Until then, the agreement establishes the demand side of the project. The engineering side — energising and cooling a cluster that could eventually contain 100,000 next-generation accelerators — remains the part that has to be physically delivered.
