Skip to content
AI Usage3 min read

Nscale and Figure Ink $3.5B Compute Deal for AI Robotics

Figure AI signs a $3.5 billion compute deal with Nscale for up to 100,000 Nvidia Vera Rubin GPUs in Texas to power physical AI models.

AB

Author

AUG Bot

Published

Digital representation of high-density AI compute servers and humanoid robotics infrastructure

Nscale and Figure Ink $3.5B Compute Deal for AI Robotics

Multibillion-dollar commitment secures up to 100,000 GPUs in Texas

Humanoid robotics developer Figure AI signed a major multi-year compute partnership with cloud infrastructure provider Nscale. The initial $3.5 billion agreement secures high-density GPU capacity to train physical intelligence models, with plans to expand total compute commitments beyond $6 billion.

Key details

Under the deal announced on September 3, 2026, Nscale will serve as Figure's preferred cloud compute provider and acquire an equity stake in the robotics company. The agreement guarantees access to compute clusters running up to 100,000 GPUs built on Nvidia's next-generation Vera Rubin rack-scale architecture. Deployments are scheduled to begin in the second half of 2027 at a dedicated data center site in Barstow, Texas.

The massive compute cluster will support Helix, Figure's foundational AI model for controlling autonomous humanoid robots. Figure's Index data-collection effort generates 35 minutes of real-world training video every second, creating huge data processing demands. Nvidia's Vera Rubin platform combines the Vera CPU and Rubin GPU, promising up to a 10x reduction in inference token costs and a 4x reduction in GPU requirements for training mixture-of-experts architectures compared to Blackwell systems.

Why this matters

This multibillion-dollar contract underscores how physical AI and robotics are becoming major drivers of hyperscale compute demand alongside large language models. Training general-purpose physical models requires massive clusters operating at gigawatt scales, shifting capital expenditure directly into specialized cloud infrastructure and hardware supply chains.

Context

The deal highlights the skyrocketing capital costs required to maintain competitive AI training pipelines. Figure has raised nearly $2 billion in equity funding to date, yet its compute commitment with Nscale exceeds $3.5 billion, demonstrating how AI startups rely on debt financing, cloud partner equity investments, and multi-year capacity reservations to secure required compute.

What happens next

Nscale and Figure will prepare the Barstow, Texas facility for initial hardware installations starting in 2027 while exploring opportunities to integrate Figure's humanoid robots into Nscale's infrastructure supply chain. As Nvidia rolls out Vera Rubin platform hardware in late 2026, other cloud providers are expected to establish similar multi-gigawatt compute agreements.


Source: Unite.AI Published on AI Usage Global, author: AUG Bot

Older post
Related

Read more

More posts that expand on the topics, companies, and AI trends covered in this story.

Solar panels alongside electrical power grid infrastructure supplying an AI data center campus
AI Usage

The AI Boom Is Accelerating US Solar Buildout While Raising Data Center Emissions

IEA projections show AI data centers will drive nearly 50% of US electricity demand growth through 2030, funding solar expansion while increasing reliance on natural gas power.

Digital representation of global data center capital expenditure and AI hardware infrastructure
AI Usage

PwC Forecasts $31.6 Trillion Global AI Data Center Capex

PwC forecasts cumulative global data center capex to reach $31.6 trillion by 2050, driven by recurring GPU hardware refresh cycles and power grid availability.

Digital representation of power grid infrastructure and data center land acquisition in Texas
AI Usage

Bitdeer buys 200 acres in Milam County, Texas, to expand AI and HPC data center

Bitdeer acquires 200 acres in Milam County, Texas, near its Rockdale site to expand AI and HPC data center infrastructure backed by 563 MW of interconnected power.