Torus Raises $500M to Bring the Chips to Power, Not the Other Way Around

Torus announced $500 million in growth capital on November 11, 2026, money it will use to scale a factory in Salt Lake City that builds AI infrastructure one cargo-van-sized unit at a time. Magnetar led the financing. ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital also participated.

The round landed alongside the launch of Foundry, Torus's prefabricated modular infrastructure product for AI inference. Foundry integrates power, storage, cooling, networking and compute in a single unit, and Torus says it can compress deployment schedules that stretch to years for a conventional hyperscale campus.

The stakes here are straightforward. Inference demand is growing, but grid interconnection queues and construction timelines are the bottleneck for most providers. Torus is betting that prefabrication plus existing electrical connections can sidestep both.

Why does the funding matter?

Half a billion dollars is a serious bet on a contrarian idea. Most AI infrastructure companies build massive campuses and then fight for the power to feed them. Torus flips that sequence.

Torus founder and CEO Nate Walkingshaw summarized the approach: "Most providers are bringing power to the chips. Foundry brings the chips to power."

That sentence is the whole business model. Instead of waiting years for new grid infrastructure, Torus looks for capacity that already exists and goes unused. The design starts with spare electrical capacity between substations and existing commercial or industrial buildings. Each system sits behind the meter, on the customer's side of the utility connection, and draws on available transformer headroom rather than requiring new grid buildout.

That constraint shapes every deployment. Usable capacity on the existing connection is the central design limit for each site. It's a different engineering problem than the one hyperscalers face, and Torus has raised $500 million on the premise that it's the more solvable one.

How fast can Foundry actually deploy?

Torus reports concrete numbers. Foundry can supply 1–20 megawatts of inference capacity per site. It moves from contract to running AI workloads in 12 to 16 weeks.

The physical footprint is small. The smallest system is roughly the size of a small cargo van. No building construction is required. Site preparation takes about one week. Installation and commissioning take one to two days.

Compare that to a conventional hyperscale campus, where deployment schedules can stretch to years, and the appeal is obvious: a customer with an underutilized industrial site and available transformer headroom can be running inference workloads in a single quarter.

Every Foundry ships paired with Torus Reserve, the company's energy-storage and management product. Battery and inertial flywheel storage, governed by predefined dispatch controls, shape demand, cover peaks, support ramping and provide backup power during grid outages. Torus says this configuration lets more compute operate on the same interconnection.

The company doesn't hand over hardware and walk away. Torus provides ongoing operations and maintenance, and Torus Protect adds integrated networking and security to the stack.

What are the deployment options?

Foundry launches with two customers: Hydra Host and another AI cloud provider that Torus did not name.

Customers choose from three deployment models, each defining how much of the stack they take on:

• Strategic — a powered enclosure without GPUs, for customers supplying their own silicon
• Partner — a turnkey bare-metal unit with GPUs included
• Direct — sovereign inference capacity through single-tenant hardware allocated exclusively to one enterprise's workloads

The Direct model is notable. Sovereign inference, where an enterprise runs workloads on hardware nobody else touches, is a growing requirement for regulated industries and governments, and Torus is building for it from day one.

How many GPUs are we talking about?

Torus reports thousands of GPUs already deployed in Foundry units. Its target is more than 47,000 by the end of 2027.

That number matters for two reasons. First, it's a substantial increase from the installed base described at launch. Second, it gives the $500 million financing a concrete measure of success. The manufacturing expansion in Salt Lake City isn't capacity for its own sake; it exists to hit that 2027 GPU target, and investors will judge the round against it.

What is Torus building in Utah?

Foundry is designed, engineered and manufactured at GigaOne, Torus's 545,000-square-foot facility in Salt Lake City. Nearly 400 employees work there, and key components are built in-house.

The Foundry production line is already online. Torus says the factory will be capable of producing 500 megawatts of IT load — the power capacity of computing equipment — with plans to expand to 1 gigawatt.

Resource use is part of the site proposition. Torus says Foundry's closed-loop cooling system is waterless and uses 95 percent less water than evaporative coolers. Water consumption has become a flashpoint for data center development, particularly in water-stressed regions, so a waterless design is a genuine selling point for siting.

There's a civic angle too. Each Foundry system donates GPU time to medical research through Folding@home, the distributed computing project for protein folding simulations.

What could slow this down?

The financing ties Torus's manufacturing and deployment plans together, but it also exposes where the plan could strain.

Two constraints stand out. The first is site availability: Torus's expansion depends on finding sites with usable transformer headroom, and that supply of underused electrical connections is finite and increasingly sought after. The second is factory throughput: GigaOne must convert its production capacity into operating systems at a scale that supports the 47,000-GPU target for 2027.

Execution at the factory and access to existing electrical connections are the central constraints on Foundry's buildout. The $500 million buys Torus the ability to scale both, but it doesn't guarantee the sites exist where customers want them.

For now, Torus has the capital, a production line running in Salt Lake City, at least two launch customers, and a deployment model that turns a years-long grid problem into a 12-to-16-week installation. Whether it hits 47,000 GPUs by the end of 2027 will tell us whether "bringing the chips to power" is an engineering insight or a slogan.