Torus, a U.S. manufacturer of distributed power and compute infrastructure, has launched Foundry, a modular AI infrastructure product designed to meet fast-growing demand for inference compute.

Alongside the launch, the company announced USD 500 million in growth capital led by Magnetar.

Modular Inference Capacity in Weeks, Not Years

Foundry packages power, storage, cooling, networking, and compute into a single prefabricated system that delivers new capacity to existing sites in weeks, rather than the years it takes to build a traditional hyperscale campus.

Each unit delivers 1 to 20 megawatts of inference capacity per site, going from contract to running AI workloads in 12 to 16 weeks.

A Shift From Training to Inference

The launch comes as AI demand shifts from training models to running inference for enterprises. According to McKinsey, data center demand for AI inference is projected to grow 35 percent a year through 2030, compared with 22 percent for training.

Inference compute runs continuously, serves users in real time, and benefits from being distributed closer to where demand sits, the company said. That difference is pulling compute toward on-premises and edge sites, and inference providers need capacity they can deploy in more places sooner.

Enterprise customers, Torus noted, want security, IP protection, and control over where their workloads run, requirements Foundry's distributed design is intended to meet in ways centralized capacity can be harder to match.

Targeting the Power Bottleneck

Torus argues that securing power is the biggest bottleneck in AI infrastructure today and that the constraint is available capacity rather than a lack of generation.

The company's technology unlocks unused grid capacity that sits between substations and existing commercial and industrial buildings by using existing interconnections.

"Inference doesn't have to wait. We've spent years learning how to make the grid more affordable, flexible, resilient, and intelligent, and Foundry applies that to compute capacity in weeks with power that's already available," said Nate Walkingshaw, Torus's founder and CEO. "Most providers are bringing power to the chips. Foundry brings the chips to power."

Walkingshaw said the industry has been trying to solve AI's power problem by building net-new infrastructure, which can raise costs and takes years to bring compute online.

The company frames its underlying premise as a capacity problem rather than a generation problem: the grid has unused capacity at many sites today, and Torus systems make that capacity usable.

Behind-the-Meter, No New Grid Infrastructure

Foundry is installed behind the meter, using available headroom and available electricity, at individual transformers. The company says the units add steady, high-load-factor demand without requiring infrastructure upgrades that often lead to rate increases and without affecting grid reliability.

The prefabricated, modular systems require no building construction, and the smallest unit is roughly the size of a small cargo van. Site preparation takes about a week, and installation and commissioning take one to two days.

Units are prefabricated at GigaOne, Torus's Salt Lake City manufacturing facility, a step the company says eliminates construction time at deployment sites and accelerates time to deploy.

Energy Storage Paired With Every Deployment

Every Foundry is paired with Torus Reserve, the company's energy storage and management product.

Reserve provides grid-enhancing technologies including inertial-based firm power, grid flexibility with load curtailment capabilities, and ramp-rate support, all delivered through what Torus calls load-shaping recipes: predefined controls that govern how power is dispatched. These capabilities shape loads to manage demand, support grid stability, and ramp power in milliseconds.

Integrated inertial flywheel and battery storage and management cover demand peaks so that more compute can fit on an existing interconnection and provide backup power during grid outages.

Vertically Integrated Stack

Torus describes Foundry as a full-stack offering from a single provider: fiber to site, chiller to coolant distribution unit, and grid to rack.

The stack includes Torus Reserve for load shaping and redundancy, Torus Foundry for inference compute, and Torus Protect for integrated networking and security. Every deployment is turnkey, with ongoing operations and maintenance provided by Torus.

Foundry is designed, engineered, and manufactured in Utah by Torus's team of nearly 400 employees at GigaOne, with key components built in-house.

Deployment Models and First Customers

Torus is launching Foundry with its first customers, Hydra Host and another AI cloud provider. The product is available in three models.

The Strategic model is a powered shell that includes power and enclosure without GPUs; the Partner model is a turnkey bare metal unit that includes power, enclosure, and GPUs; and the Direct model provides sovereign inference capacity directly to enterprise customers, with single-tenant hardware allocated exclusively to a company's workloads.

Stay Ahead of the United States Data Center Construction Boom

Hyperscale campuses, colocation expansions, and edge facilities are reshaping digital infrastructure across the United States, but the contractors, suppliers, and investors who win in this market are the ones who see opportunities before they become headlines.

The Global Project Tracking (GPT) platform by Blackridge Research gives you real-time visibility into data center developments nationwide, from early-stage announcements through final commissioning, so your teams can act on intelligence rather than react to news.

Whether you're tracking utility-scale builds in Northern Virginia, greenfield sites in the Midwest, or retrofit projects in major metros, the platform puts every decision-critical milestone at your fingertips.

• Upcoming Projects

• Tender Notices

• Contract Awards

• Projects Under Construction

• Completed Projects

Book a Free Demo today and discover how the Global Project Tracking (GPT) platform by Blackridge Research can sharpen your competitive edge in the United States data center market.