Torus puts AI chips inside battery containers and raises $500 million

Torus, a Salt Lake City maker of power and compute infrastructure, has raised $500 million in growth capital, led by Magnetar, and launched Foundry, a prefabricated system that bundles power, storage, cooling, networking and GPUs for AI inference. The company says it can go live in 12 to 16 weeks.

Bloomberg reported the round on November 11 and described the product as shipping-container sized. The details below come from Torus’s own release on Business Wire.

Foundry units of 1 to 20 MW, installed behind the meter

According to Torus, each Foundry site runs from 1 to 20 megawatts and needs no building construction. The smallest unit is about the size of a small cargo van. The company puts the timeline at 12 to 16 weeks from contract to live inference: roughly a week of site preparation, then one to two days of installation and commissioning.

The units connect behind the meter and draw on spare grid capacity, including headroom at individual transformers. Torus says this avoids new grid infrastructure and does not affect grid reliability. Flywheel and battery storage is built in to smooth demand peaks, so more compute fits on a single interconnection, and it provides backup during outages. Every Foundry is paired with Torus Reserve, the company’s storage and energy management product, and with Torus Protect for networking and security.

Cooling is closed-loop and waterless, which Torus says uses 95% less water than evaporative coolers. The company sells Foundry in three ways: a powered shell without GPUs, a turnkey bare-metal unit with GPUs, and single-tenant “sovereign” inference capacity sold directly to enterprises.

Who is behind the money and who is buying

Magnetar led the round. ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital also took part. Torus says the money will expand its GigaOne factory and fund deployment of Reserve and Foundry across the United States.

GigaOne is a 545,000-square-foot plant in Salt Lake City with nearly 400 employees. Torus says it opened its first production line in early 2026 and has added a new line about every 21 days since. The Foundry line can build 500 MW of IT load, with a plan to reach 1 GW.

Launch customers are Hydra Host and a second AI cloud provider that Torus does not name. The company says thousands of GPUs are already deployed in Foundry and targets more than 47,000 by the end of 2027. The release does not say which GPU vendor or model the units use, and it gives no pricing or valuation.

“Most providers are bringing power to the chips. Foundry brings the chips to power.”

That line is from founder and CEO Nate Walkingshaw. Dave Snyderman of Magnetar said in the release that access to power is among the biggest challenges in scaling AI infrastructure.

A bet that power, not chips, is the bottleneck

The pitch lands as power has become the constraint that AI builders talk about most. Torus argues that waiting for new transmission and generation delays compute, and that existing sites with spare capacity can be used sooner. The release cites McKinsey projections that inference demand in data centers will grow about 35% a year through 2030, against 22% for training.

The round is also a step up for Torus. In September 2025 the company announced a $200 million raise from Magnetar, when it was pitching modular power plants to utilities and data centers. The new $500 million is 2.5 times that amount, and it moves the company from selling storage and power to also selling the compute that runs on it.

Torus is not alone in the modular idea. The Wall Street Journal reported, via Techmeme, that Mavria, which wants to build modular data centers that can be moved by truck, raised $130 million at a $1 billion valuation in pre-seed and seed rounds led by Paradigm. Others chase the same scarce power from the generation side: Venture Global said this week it is eyeing 5 GW of power for AI. Torus differs in starting from a factory that already builds storage hardware, which it can fit into each unit.

Meanwhile, the cost of building AI capacity is pushing others toward creative financing, as seen in how CoreWeave is turning to chip-backed debt and customer pre-payments.

What the release does not establish

Every figure above is Torus’s own claim, published in a company press release with no independent verification. The 12 to 16 week schedule, the grid-reliability claim and the water saving are untested by outside parties. The release names no GPU supplier, gives no price per megawatt or per GPU hour, and does not say how many Foundry units are in service. Torus also describes the offering as launching now with its two launch customers, so general availability for other buyers is not spelled out.

The company also says it donates GPU time to Folding@home for medical research.