Torus raises $500M for Foundry container-sized AI data centers

The pitch is a shipping container that holds 576 HGX B300 GPUs in 577 square feet, runs on 1.2 megawatts of power, and goes from contract to operational in 12 to 16 weeks. Salt Lake City energy-storage startup Torus has raised $500 million in a Series C to build more of them, Bloomberg reported. The round was led by Magnetar with ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital participating.

The product is called Foundry. It comes in 1-to-20-megawatt sites that stack those 1.2 MW units. CEO Nate Walkingshaw framed the design as what the company's grid work was always pointing toward: "our version of a distributed data-center approach to move the chips to the power that's already there," he told Bloomberg. His line in the launch materials was blunter: "Foundry brings the chips to power."

His case against the greenfield build: "The industry has been trying to solve AI's power problem by building net-new infrastructure, which causes costs to potentially rise and takes years to bring compute online. Inference doesn't have to wait."

Torus is manufacturing Foundry out of GigaOne, its 545,000-square-foot Salt Lake City plant, with a current line rated for 500 MW of IT load and plans to extend to 1 GW. The company says it is targeting more than 47,000 GPUs deployed in Foundry units by the end of 2027. Hydra Host is named as the first customer; a second AI cloud provider is cited but not identified.

Dave Snyderman, managing partner at Magnetar, said Foundry is "bringing compute to available power, helping unlock capacity within the existing grid." Torus itself puts annual AI inference data-center demand growth at 35% through 2030. That figure, like the deployment timeline, is the company's own. It lands in a busy week on our AI-infrastructure tracker.