Torus raises $500m to deploy AI computing units using existing grid power
The Salt Lake City company aims to bypass data center power shortages by placing prefabricated AI units at existing business and factory sites.
The short version
- Torus raised $500m in a Series C round led by Magnetar to expand its Salt Lake City factory and launch new US sites.
- Foundry units are prefabricated systems that pack power, cooling, and networking into units as small as a cargo van.
- The systems use spare capacity at existing transformers, allowing for deployment in 12 to 16 weeks without new grid connections.
- Torus plans to have more than 47,000 GPUs running in Foundry units by the end of 2027.
- The cooling systems for these units run in a closed loop and do not use water.
Why it matters
As data centers face increasing regulatory and physical hurdles to secure new power, this model allows companies to scale AI without heavy infrastructure construction. It provides a faster path to deployment for businesses needing inference capacity.
Bypassing the power bottleneck
Data center developers are facing a significant hurdle in the United States: finding enough electricity to power new facilities. Recent local ballot measures have shown a trend of voters rejecting new data centers, and new state leadership in regions like Texas and Ohio has pushed for tighter limits. These obstacles make it difficult for companies to build massive new hubs from scratch.
Torus, a company based in Salt Lake City, is attempting to solve this by changing where the computing happens. Instead of trying to bring massive amounts of power to new chips in isolated locations, they are bringing the chips to the power. Their new Foundry units are designed to sit at existing business and factory sites that already have electricity available.
By utilizing spare capacity at existing transformers, Torus allows companies to add AI computing without needing to request new grid connections. This approach avoids the noise, pollution, and infrastructure strain that often lead to local opposition. As founder and CEO Nate Walkingshaw told Bloomberg, the model allows companies to add computing without raising power bills or adding significant environmental impact.
How the Foundry units work
The Foundry system is a prefabricated unit that packs several complex components into a single package. These units include power, storage, cooling, networking, and the actual computing hardware. The smallest of these units is roughly the size of a small cargo van, while larger sites can reach up to 20 megawatts of capacity.
A key part of the technology is the Torus Reserve, which is a flywheel and battery storage product. This system covers demand peaks, which allows the site to fit more computing onto the same electrical connection. Additionally, the Reserve provides backup power during outages, ensuring that the AI workloads remain stable and uninterrupted.
The setup is designed for speed and efficiency. Torus claims a site can move from a signed contract to running AI workloads in just 12 to 16 weeks. The actual site preparation takes about one week, and the physical installation of the units takes only one to two days. The cooling for these systems also runs in a closed loop, meaning it uses no water.
Focusing on AI inference
Torus is specifically targeting the inference side of the AI market. Inference is the process of running a trained AI model to provide answers or perform tasks for users. This is distinct from training, which is the initial process of teaching an AI model how to behave. Torus notes that demand for inference is expected to grow faster than training demand over the next few years.
The company cites figures from McKinsey suggesting that data center demand for inference will grow 35% annually through 2030. In comparison, the growth rate for training is projected to be 22% during that same period. This shift in demand helps define the target market for the Foundry units, which are built to handle these active workloads.
To support this growth, Torus plans to have more than 47,000 GPUs running in Foundry units by the end of 2027. They already have thousands of GPUs running in these units today. The company is also contributing to scientific progress by giving some of this GPU time to medical research through the Folding@home project.
Customers and expansion
The company has already secured its first customers, which include Hydra Host and a second unnamed AI cloud provider. These customers have different options for how they want to interact with the hardware. They can purchase an empty powered shell, a turnkey unit that comes with GPUs already installed, or they can simply reserve capacity for their own exclusive use.
The funding from the $500m round will be used to expand GigaOne, which is Torus's factory in Salt Lake City. This facility opened its first production line in early 2026 and is where the Foundry units are manufactured. The production line is currently capable of producing 500 megawatts of IT load, but Torus intends to expand that capacity to 1 gigawatt.
This expansion comes on the heels of a previous $200m raise in September 2025. That earlier funding was specifically for Torus's modular battery systems, which help businesses manage energy use and lower power bills. The company is now combining its expertise in energy storage with high-performance computing to create a more flexible infrastructure for the AI era.
The broader energy landscape
The move by Torus reflects a broader trend of tech giants trying to manage the strain AI puts on the power grid. For example, Google is currently funding a 1GW virtual power plant to help balance energy needs. Similarly, OpenAI has launched a demand response program, which is a strategy to adjust power usage based on grid conditions.
By using existing grid power at diverse sites, Torus provides a decentralized alternative to the traditional data center model. This helps mitigate the need for massive new infrastructure projects that often face political and environmental hurdles. It allows for a more modular way to scale AI capacity as the market continues to expand.
As the industry moves forward, the success of this model will depend on the availability of spare capacity at existing sites. If companies can successfully deploy these units without triggering new regulations or high costs, it could become a standard way to scale AI. Torus is positioning itself as a primary provider of this
Sources and links
- ReportingThe Next Web: the article we sourced
First reported by The Next Web. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.