← News

Torus raises $500M to drop AI computers wherever the power already is

Torus, a Salt Lake City energy storage maker, said on Nov. 11, 2026, that it raised $500 million in a round led by Magnetar and launched Torus Foundry, prefabricated units that put 1 to 20 megawatts of AI inference computing at existing sites using grid power those sites already have.

The AI buildout's real bottleneck isn't chips, it's the multi-year line for a grid connection, and voters just made clear they don't want giant data centers next door. Torus's bet is the opposite of a gigawatt campus: lots of small boxes hiding behind factories, eating power that's already paid for. Sixteen weeks from contract to compute is a bold promise; 47,000 GPUs by 2027 is the number to hold them to.

Torus, a Salt Lake City company that has been selling modular battery systems, said it raised $500 million in a round led by Magnetar. ICONIQ, Origin Ventures, Ridgeline Partners, and Trinity Capital also took part. Bloomberg, which reported the raise first, called it a Series C. A Series C is a later private funding round, after earlier ones. No valuation was disclosed. The announcement is dated Nov. 11, 2026. The amount and the investors are Torus's, via Business Wire. The Series C label is Bloomberg's.

The money expands GigaOne, Torus's factory in Salt Lake City, and pays for Foundry sites across the United States. Those uses are the company's.

Nate Walkingshaw, founder and chief executive, is quoted: "Most providers are bringing power to the chips. Foundry brings the chips to power." The quotation is his.

Each Foundry site packs power, storage, cooling, networking, and computing into one prefabricated system. Prefabricated means the unit is built at the factory and delivered, rather than constructed on the site. Sites range from 1 to 20 megawatts. A megawatt is a million watts. The smallest unit is about the size of a small cargo van. Those lines are Torus's.

Torus says a site can go from a signed contract to running AI workloads in 12 to 16 weeks. Site preparation is about a week, and the install takes one to two days. Those timelines are the company's claims.

The units sit behind the meter, on spare capacity at transformers the site already has, so no new grid connection is needed. Behind the meter means the box plugs in on the customer's side of the electric meter, using power that connection already carries. Each unit is paired with Torus Reserve, a flywheel plus a battery, to cover demand peaks and to supply backup power. A flywheel stores energy by spinning a heavy wheel. Those lines are Torus's.

Foundry is built for inference, which means running a model that is already trained, rather than the long job of training one. Torus says the cooling is closed-loop and uses no water. Closed-loop means the coolant stays inside the machine. That no-water line is a company claim. Each system donates GPU time to medical research through Folding@home. A GPU is the chip that does the math. Folding@home is a project that uses spare computer time to simulate proteins. Those lines are Torus's.

Torus says thousands of GPUs already run in Foundry units, and it plans more than 47,000 by the end of 2027. The first customers are Hydra Host and a second AI cloud provider that is not named. A buyer can take an empty powered shell, a turnkey unit that already has the GPUs, or reserved capacity. Those options, the customer names, and the GPU counts are the company's. The 47,000 figure is a plan, not a count of machines already installed.

Before Foundry, Torus sold modular battery systems. It raised $200 million from Magnetar in September 2025. Nearly 400 people work at GigaOne, which opened its first production line in early 2026. The Foundry line can produce 500 megawatts of IT load, and the company plans to expand that to 1 gigawatt. IT load is the computing equipment the power is meant to run. A gigawatt is a thousand megawatts. Those figures are Torus's.

The Next Web, citing Bloomberg, notes the local politics around big server buildings. In last week's midterms, U.S. voters rejected data centers in 40 of 41 local ballot measures. That count is Bloomberg's, as The Next Web carried it. It is context for why a smaller box might be easier to site. It is not a Torus result.

The picture is grey Torus Foundry units outside GigaOne, with power lines behind them. The credit is Ken Hovley, Torus, as carried by The Next Web.

In plain terms, a Salt Lake City battery company raised $500 million, led by Magnetar, to park van-sized computers at factories and businesses that already have power, instead of waiting years for a new grid hookup. The 12-to-16-week promise and the water-free cooling are Torus's claims. Bloomberg called the round a Series C. No valuation was disclosed.

RELATED

ONLINE…

Comments

guidelines

Loading…

Loading…

Sources