Introducing Torus Foundry: AI Compute Where the Power Already Is

Years. That’s how long most AI compute waits for power.

We get it online in weeks.

Everyone assumes the grid is out of power. That every new AI data center means new plants, new lines, and higher bills for everyone else.

The grid isn’t out of power. It’s congested. There’s stranded capacity sitting unused at sites across the country. Why? Because electricity only flows one way and can’t get to where it’s needed.

The answer isn’t waiting on new power. It’s using what’s already there. 

Today we’re launching Torus Foundry to put AI compute where the power already is, and we’ve raised $500M, led by Magnetar, to take Foundry nationwide. 

About Foundry

Foundry is modular AI inference compute that runs on power a site already has. What’s inference? Training is how an AI model learns. Inference is the model actually working, every time someone uses AI. It’s where demand is growing fastest, and it doesn’t need one giant campus. It needs smaller blocks of compute in a lot of places at once.

With Foundry, power, storage, cooling, networking, and compute come in one prefabricated unit we build at GigaOne in Salt Lake City. It goes where power is available, so nobody waits on the grid. Site prep takes about a week, install takes a day or two, and you go from contract to token in 12 to 16 weeks.

Bringing power to chips

Every Foundry unit pairs with Reserve, our energy storage and management system. AI inference load looks like an EKG to the grid, all spikes. Reserve smooths it out, so the grid sees steady, flexible load instead of a problem. That means no new strain on the community, no new generation or infrastructure to pay for, and no waiting in a years-long interconnection queue.

Torus builds energy storage and management systems that make the grid more reliable, scalable, secure, and affordable. We install them at commercial and industrial sites and work with utilities so those sites run as virtual power plants or distributed energy resources. We operate across the Western Interconnection and are expanding nationwide.

Foundry's initial customers

That work taught us how the grid behaves and where the power actually sits. So instead of waiting years to bring power to the chips, we bring the chips to the power.

Hydra Host and another AI Cloud provider are our first customers, with thousands of GPUs already running. Thanks to Magnetar, ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital for betting on us. Next stop: 47,000 GPUs by the end of 2027.

To the Foundry team and everyone at GigaOne, who took this from an idea to a unit on the floor in months: thank you.

If you’re waiting on power, you don’t have to. Come see one.