Torus Raises $500 Million to Put AI Inference Where Spare Grid Power Already Sits, in Units It Says Go Live in 12 to 16 Weeks
Torus, a Salt Lake City maker of battery and flywheel storage, announced $500 million in growth capital led by Magnetar on November 11, which Bloomberg described as a Series C, and launched Foundry, prefabricated units that deliver 1 to 20 megawatts of AI inference computing behind the meter at existing commercial and industrial sites. Torus says a site can go from contract to running workloads in 12 to 16 weeks and plans more than 47,000 GPUs in Foundry units by the end of 2027. No valuation was disclosed.
Most AI data center plans start with finding new power. Torus is betting on the opposite: small amounts of unused capacity already sitting at factories and commercial sites. Its new Foundry units bring the chips to that power, and investors led by Magnetar have put up $500 million behind the idea.
What Torus announced#
Torus, which has sold battery and flywheel storage systems to businesses, launched Foundry and announced the funding on November 11. Magnetar led, with ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital participating. Bloomberg, which first reported the raise, described it as a Series C, according to The Next Web; neither gave a valuation.
Per Torus:
- Size and speed: each site delivers 1 to 20 MW, and a site can go from contract to running AI workloads in 12 to 16 weeks. The smallest unit is about the size of a small cargo van.
- Power: units sit behind the meter and use spare capacity at existing transformers, so they need no new grid infrastructure. Each is paired with Torus Reserve storage to cover demand peaks and provide backup.
- Scale: thousands of GPUs are already deployed in Foundry units, with plans for more than 47,000 by the end of 2027. First customers are Hydra Host and an unnamed AI cloud provider.
- Manufacturing: units are built at Torus's GigaOne factory in Salt Lake City, whose Foundry line can produce 500 MW of IT load, with plans for 1 GW.
The pitch is aimed at inference, which runs continuously and closer to users. The Next Web notes US voters rejected data centres in 40 of 41 local ballot measures in last week's midterms, per Bloomberg. Large AI buyers are working the same constraint from the other side, as with OpenAI's pledge to cut power use when the grid is strained.
Limits#
- Deployment times, GPU counts and grid claims are Torus's; we found no independent assessment of how much spare capacity existing sites can offer.
- The valuation and the round's mix of equity and other financing were not disclosed.