Nate Walkingshaw wasn’t trying to build an AI company. He was trying to store hydropower for his family’s tree farm in Utah.
That search led him to flywheels, then to a startup, and now to a box full of AI chips rolling out of a Salt Lake City factory.
On November 11, 2026, Torus announced $500 million in new funding and launched Foundry, a prefabricated, shipping-container-sized unit that puts power, energy storage, cooling, networking and AI computing in one package. Magnetar led the round. ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital joined in. No valuation was disclosed.
The company’s announcement calls it growth capital. Bloomberg, which interviewed Walkingshaw, reported it as a Series C, the third major funding round and usually the one a company raises to scale something that already works.
A tree farm and a spinning wheel
Walkingshaw founded Torus in 2021 with Gilbert Lee. Their first bet was the flywheel, a heavy wheel spun up to hold energy as motion and release it almost instantly.
The early systems went into homes. Then the company aimed bigger.
Torus paired its flywheels with batteries and started selling to businesses and utilities. TIME put its Torus Spin on the list of the best inventions of 2024. Rocky Mountain Power signed on for 70 megawatts of demand response, a setup where stored energy takes pressure off the grid during peak hours.
On September 9, 2025, Magnetar invested $200 million, and the partnership with PacifiCorp, Rocky Mountain Power’s parent, grew to cover up to 500 more megawatts. Part of that money went into GigaOne, a factory of roughly 545,000 square feet in Salt Lake City. Its first production line opened in early 2026. Torus says it has added a new line every 21 days since.
Nearly 400 people work there now.
The decision to flip the problem
Across the US, data-center developers are hunting for electricity. A new hyperscale campus can take years, much of it spent in the interconnection queue, the waiting line every big project joins before the utility will hook it up. Torus looked at the same grid and saw something else: plenty of power already sitting at factories and business sites, just not always where anyone needed it. So it made the call that defines this round. Rather than wait for power to reach the chips, Torus would ship the chips to where the power already is. It already knew how to build hardware, manage energy and work with utilities. Foundry wraps all of that around a rack of AI chips.
“Most providers are bringing power to the chips. Foundry brings the chips to power.”
Nate Walkingshaw, CEO and Founder of Torus
Foundry is built for inference, the everyday work of running a trained AI model when someone asks it a question. Each site gets 1 to 20 megawatts. From signed contract to running compute takes about 12 to 16 weeks.
Walkingshaw said the target sites are businesses and factories with spare grid capacity, avoiding costly upgrades that tend to land on utility bills. It’s a version of the shift toward AI computing at the edge, closer to the people using it.
Every unit comes paired with Reserve, the company’s storage system. According to Torus, AI inference hits the grid in sharp spikes, and Reserve smooths them into a steady load.
Customers can buy an empty powered shell, a turnkey unit with GPUs, or capacity reserved for one company. Hydra Host, an AI infrastructure company, and a second AI cloud provider Torus hasn’t named are the first buyers. Thousands of GPUs are already running.
Wait, the flywheel company sells computers now?
Sort of. And it’s less of a leap than it looks.
Every chapter of this company has been about one thing: getting power where it’s needed without building a new power plant. A tree farm. A house. A utility at peak hour. Now a rack of GPUs.
What changed is the customer. For years, Torus sold the energy. With Foundry, it also sells what the energy runs. That puts a company born on a tree farm inside the business history of AI, on the hardware side most people never see.
Magnetar has now backed Torus twice in 14 months, the second time with two and a half times as much money. That’s conviction.
The honest caveat: one named customer is still one named customer.
47,000 GPUs by the end of 2027
Torus says the new money will expand GigaOne and fund Reserve and Foundry deployments across the United States. The Foundry line can already turn out 500 megawatts of IT load, and the company plans to take it to 1 gigawatt. Its stated goal is more than 47,000 GPUs running in Foundry units by the end of 2027, with Foundry going nationwide.
