Torus Secures $500M to Scale Foundry AI Infrastructure

A factory's electrical connection can become part of the pitch for its next AI system. Torus is building a business around that possibility, combining its experience in energy storage with modular infrastructure for running AI models. The Salt Lake City company wants buyers to consider the usable power already at a site when deciding where computing should go.

On November 11, 2026, Torus announced $500M in growth capital, led by Magnetar with participation from ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital. The financing accompanies the launch of Foundry, its prefabricated AI inference infrastructure, and will support manufacturing expansion and US deployments of Foundry and Torus Reserve energy systems. The official announcement describes growth capital without specifying the financing mix or a formal series designation.

For enterprise buyers and compute providers, the proposition brings several interdependent purchases into one delivery conversation. Electricity, storage, cooling, networking and computing have to work together before a model can serve a customer; Torus is extending its responsibility across that chain.

How Foundry turns site power into a compute decision

Foundry is designed for deployment on existing grid interconnections, with energy storage and management helping make available capacity useful for computing. Torus describes systems that grow in 1.2 MW increments across sites supporting 1-20 MW. Those are company specifications, and their commercial significance starts with the buyer's ability to add capacity in smaller steps.

The product page gives Torus responsibility for system design, permitting, installation, commissioning and lifetime operations and maintenance. That breadth matters to a customer coordinating several technical teams. A rack cannot deliver an inference service while the electrical and cooling work remains somebody else's unfinished project, so a common supplier has value beyond the enclosure arriving on a truck.

Torus says the overall interval from contract to running workloads is 12-16 weeks. The much shorter 1-2 day figure refers to installation and commissioning after preparation, an important distinction for anyone turning the specification into a project schedule. Factory assembly moves work away from the customer site, while design, approvals and integration still have their place in the delivery plan.

Inference means running an already-trained model to produce an answer or perform a task. Torus's founder explanation of Foundry places that work near available power and enterprise demand, expanding the locations buyers can consider. Choosing among those locations then becomes an operating decision shared by the people buying compute and the people responsible for the building.

The storage business underneath the AI expansion

Torus Reserve combines mechanical flywheel storage with batteries. The company describes the flywheel as the rapid-response component and batteries as the longer-duration resource, with software coordinating load shaping and energy use. That division of labor helps explain why Torus treats the power system as an active part of the compute installation.

A customer's connection has limits, and demand changes over time. Storage can respond to a rapid increase in load or shift when electricity is drawn, allowing the system to manage more of the interaction between the equipment and the grid. The business proposition rests on applying that control to the particular site being developed, rather than assuming that the same amount of usable headroom exists everywhere.

Torus was founded in 2021 in Springville, Utah. Origin Ventures identifies Nate Walkingshaw and Gilbert Lee as its co-founders and former Pluralsight executives, giving the company a software leadership history alongside its manufacturing work. Nate Walkingshaw remains CEO in the current announcement.

That history makes the expansion into compute more legible. Torus already designs energy systems around a facility's requirements; Foundry gives it another demanding load to serve and another layer of customer responsibility. Its delivery team now has to connect the customer's power arrangement with the infrastructure on which the AI workload depends.

What the capital supports, and how the history fits

The new capital is intended to expand GigaOne, Torus's 545,000-square-foot manufacturing base in Salt Lake City, and support Reserve and Foundry deployments across the United States. That allocation ties the financing to production and installation work, a connection also visible in solar and storage project finance. For this business, expanding supply means coordinating the equipment manufactured in Utah with the conditions at each eventual customer site.

Magnetar already had a financial relationship with Torus. A separate September 9, 2025 announcement disclosed a $200M investment for modular energy infrastructure, establishing a history before the Foundry launch. The latest commitment extends that relationship as Torus adds computing to its energy platform.

There is also a distinct equipment-financing record. On April 30, 2026, Trinity Capital announced a commitment of up to $35M to help expand manufacturing at GigaOne. A financing commitment describes available support under an arrangement; it should not be added to an equity tally as if every dollar had been drawn.

The current Torus newsroom reports $767M raised. That company-reported figure supplies context, while the public disclosures leave valuation and the composition of the latest $500M unspecified. The useful connection across the announcements is the continuing investment in the manufacturing capacity needed to deliver the company's broader infrastructure offering.

Three configurations change the buyer's job

Foundry's launch offers three commercial configurations. A buyer can take a powered shell without GPUs, a turnkey bare-metal system that includes them, or dedicated single-tenant inference capacity. Each arrangement gives the customer a different combination of equipment ownership, integration work and reliance on Torus's ongoing service.

For an enterprise, that creates a procurement discussion extending beyond the price of a machine, into questions also shaping AI compute finance. Buying a shell leaves a different hardware decision on the customer's desk than buying access to capacity. A compute provider may bring its own requirements and infrastructure expertise, while another organization may want more of the operating work carried by its supplier.

Hydra Host is the named initial Foundry customer. Torus also identifies an unnamed AI cloud provider and reports thousands of GPUs already deployed, with a plan for more than 47,000 by the end of 2027. Those remain company-reported deployment figures and a forward target, respectively, rather than independently established site economics.

As Foundry reaches additional locations, its commercial configurations will keep meeting the particulars of the facility: available power, the customer's workload and the responsibility each party accepts. The buyer choosing an AI system now has a reason to bring the person who understands the existing transformer into the conversation, while there is still time for that knowledge to shape the order.