LE PERSPECTIVE / THE STORY

Torus has launched Foundry, a modular AI infrastructure system designed to bring inference computing online faster, alongside $500 million in growth capital led by Magnetar. The Salt Lake City company is targeting one of AI’s biggest constraints: access to usable electricity.

Foundry combines GPUs, power, energy storage, cooling and networking in prefabricated units installed at existing sites rather than waiting for new data center campuses. Torus says customers can go from contract to running AI workloads in 12–16 weeks, with installation and commissioning taking one to two days.

Power Is the Real Bottleneck
Foundry’s central bet is bringing chips to available electricity, rather than building new power infrastructure around chips. That could shorten AI deployment cycles, but Torus still has to prove the model at scale.

Each Foundry site supports 1–20 megawatts. Its paired Torus Reserve system uses batteries and flywheels to smooth sudden GPU power spikes, manage peak demand and provide backup during outages. Torus says this lets operators make better use of existing grid connections.

Hydra Host and another unnamed AI cloud provider are Foundry’s initial customers. Torus says thousands of GPUs are already deployed and targets more than 47,000 by the end of 2027. That latter figure is a target, not installed capacity.

The funding also includes ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital. Torus plans to expand manufacturing and deployments across the United States. Foundry units are built at its 545,000-square-foot GigaOne factory in Salt Lake City.

Customers can choose a powered shell, a GPU-equipped bare-metal system, or dedicated single-tenant inference capacity. Torus says its closed-loop cooling uses 95% less water than evaporative cooling systems.

Rather than building massive AI campuses around future power connections, Torus is betting smaller compute sites can be placed wherever capacity already exists. The test will be how quickly it can turn scattered sites into repeatable, reliable GPU supply.

Get stories like this in your inbox every week. Subscribe to The Week in Perspective.