Torus Raises $500M Growth Round to Bring AI Inference Sites to Spare Grid Power
"Most providers are bringing power to the chips. Foundry brings the chips to power." That line from Torus founder and CEO Nate Walkingshaw is the whole pitch. On Wednesday the Utah company announced $500M in growth capital led by Magnetar, with ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital participating, and launched a product called Foundry. The release doesn't name the round; Bloomberg calls it a Series C. Valuation was not disclosed. Release: https://www.torus.co/news/torus-launches-modular-ai-infrastructure-foundry
Foundry is a prefabricated box that packs power, storage, cooling, networking and compute together, at 1 to 20 MW per site. The smallest one is roughly the size of a small cargo van. It goes behind the meter at existing commercial and industrial sites, using grid capacity that is already connected but sits unused between the substation and the building. Torus says it takes 12 to 16 weeks "from contract to token." Each unit is paired with Torus Reserve, its flywheel-plus-battery storage system, which shapes the load so the grid doesn't notice a sudden data center.
The money goes into expanding GigaOne, its factory in Salt Lake City, and deploying Reserve and Foundry across the US. Magnetar already put $200M in last year, so this is a second, bigger check from the same backer.
The insight worth stealing: the scarce thing in AI inference is not chips or land, it is the time it takes to get power. A hyperscale campus takes years. A box at a factory that already has a fat grid connection takes a season. If agent traffic keeps growing the way the labs say, inference will end up spread across thousands of small sites like this, not just a few giant ones.
← Back to all articles
Foundry is a prefabricated box that packs power, storage, cooling, networking and compute together, at 1 to 20 MW per site. The smallest one is roughly the size of a small cargo van. It goes behind the meter at existing commercial and industrial sites, using grid capacity that is already connected but sits unused between the substation and the building. Torus says it takes 12 to 16 weeks "from contract to token." Each unit is paired with Torus Reserve, its flywheel-plus-battery storage system, which shapes the load so the grid doesn't notice a sudden data center.
The money goes into expanding GigaOne, its factory in Salt Lake City, and deploying Reserve and Foundry across the US. Magnetar already put $200M in last year, so this is a second, bigger check from the same backer.
The insight worth stealing: the scarce thing in AI inference is not chips or land, it is the time it takes to get power. A hyperscale campus takes years. A box at a factory that already has a fat grid connection takes a season. If agent traffic keeps growing the way the labs say, inference will end up spread across thousands of small sites like this, not just a few giant ones.
Comments