Torus launches prefabricated AI sites using spare power at existing buildings

How it works

Plug-and-play computer units that run AI using spare electricity already available at factories and warehouses.

1. 🏭Workers prep the site in about a week
2. 🔌Units plug into the building's existing transformer
3. ⚡Storage covers peak power demands
4. ✅Running AI in 12 to 16 weeks total

Founded and led by Nate Walkingshaw, Torus said the funding round was led by Magnetar, with ICONIQ, Origin Ventures, Ridgeline Partners, and Trinity Capital also investing ¹. Bloomberg reported the raise as a Series C round, though neither source disclosed the company's valuation .

Guess

How long to go from signed contract to running?

12–16 weeks2–4 years3–6 months

What Foundry does

Each Foundry site packs power, battery storage, cooling, networking, and computers into one prefabricated system ¹. Sites range from 1 to 20 megawatts. The smallest unit is roughly the size of a small cargo van .

Torus says a site can go from signed contract to running AI workloads in 12 to 16 weeks ¹. Preparing the site takes about a week, and installation takes just one or two days .

The key trick is that each unit sits behind the meter — meaning it plugs into the existing electrical transformer at a factory or warehouse, using spare capacity rather than requiring a brand-new grid connection ¹. Each unit is paired with Reserve, Torus's flywheel and battery storage product, which covers peak power demands so more computing fits on the same connection . It also provides backup power during outages .

An exampleInvented to explain it

Dana runs a warehouse needing faster AI answers

DanaCan we run AI here without waiting years for power?

Torus repYes, we use your spare electricity right now.

DanaHow long until it's running?

Torus repAbout 12 to 16 weeks from contract to live.

DanaWhat kind of AI does it handle?

Torus repInference — answering questions from real users.

Dana's warehouse starts running AI workloads within months instead of years.

Built for inference

Foundry is designed for inference, the work of running trained AI models when real users ask questions or make requests — the kind of task you do every time you chat with an AI assistant ¹.

Torus cites McKinsey estimates that demand for inference will grow 35% per year through 2030, compared to 22% for training, the process of teaching AI models new skills ¹. The inference-focused startup Fal recently raised $1.3 billion in a round that doubled its value .

Torus claims thousands of GPUs are already running inside Foundry units, with plans to deploy more than 47,000 by the end of 2027 ¹. Two firms have become Foundry's initial clients: Hydra Host, which builds AI infrastructure, and another AI cloud provider that was not identified by name . Companies can choose between buying an empty powered shell, a turnkey unit with computers already installed, or reserved capacity meant only for them .

From batteries to computing

Until now, Torus sold modular battery systems helping businesses manage their energy use and cut power bills ¹. The company raised $200 million from Magnetar for that earlier business in September 2025 .

The pivot comes as US data centre developers face growing difficulty finding power. In last week's midterms, voters rejected data centres in 40 of 41 local ballot measures, and Democrats who campaigned for tighter limits on data centres won governor races in Texas and Ohio ¹.

In an interview with Bloomberg, Walkingshaw pointed out that voter frustration opens the door for companies offering more computing power without driving up electricity bills or generating noise and pollution ¹. He pointed to Google funding a 1 gigawatt virtual power plant and OpenAI launching a demand response programme to ease strain on power grids .

According to Torus, Foundry's cooling system operates in a sealed loop and uses zero water ¹. Each system also allocates some GPU time to Folding@home, a project helping medical researchers study diseases .

Torus employs nearly 400 people at GigaOne, its factory that opened its first production line in early 2026 ¹. The Foundry production line can currently build enough for 500 megawatts of computing load, and the company plans to expand it to 1 gigawatt .

This story reports on a trend for information only. It is not advice or encouragement to do any of it.

Key facts

• Torus raised $500 million in a Magnetar-led Series C round
• Foundry units range from 1 to 20 megawatts and take 12-16 weeks from contract to operation
• Units plug into existing transformers, needing no new grid connection
• Torus plans over 47,000 GPUs in Foundry by end of 2027
• Foundry cooling uses no water and runs in a closed loop

Sources

1. Torus Foundry launches as Torus raises $500m in a Magnetar-led round — thenextweb.com