Torus raises $500 million to build modular AI data center units

Solving the power bottleneck

The rapid expansion of artificial intelligence has created a massive demand for computing power across the United States. However, data center developers are increasingly struggling with electricity availability, high infrastructure costs, and lengthy permitting processes for large-scale projects. Conventional hyperscale data centers often require hundreds of megawatts, which can necessitate multi-billion dollar investments in new substations and transmission lines.

To address these hurdles, Utah-based startup Torus is developing a product called Foundry. This modular data center technology is designed to bring computing capacity to locations where electricity capacity is already available. By utilizing existing power, the company aims to bypass the delays and costs associated with building new grid infrastructure from scratch.

Torus recently secured $500 million in a Series C funding round to accelerate the manufacturing and deployment of these units. The round was led by Magnetar, with additional participation from Iconiq Capital, Origin Ventures, and Trinity Capital. This investment signals a growing interest in infrastructure models that combine power management with computing capacity.

How modular units work

Foundry consists of shipping-container-sized units that combine AI computing infrastructure with integrated battery energy storage. This design allows the systems to manage electricity demand more effectively while providing the necessary power for AI workloads. The technology is specifically intended for AI inference, which is the process of running trained models to generate outputs like predictions or responses.

Unlike traditional data centers that require massive, purpose-built buildings, these units are transportable and can be installed at commercial facilities or factories. Torus CEO and co-founder Nate Walkingshaw stated that the units can be installed at sites with excess electricity capacity. This allows businesses to accommodate additional computing loads without requiring major upgrades to their current electricity setup.

The Foundry units are designed to scale between 1MW and 20MW in capacity. This modularity means that businesses can expand their computing power incrementally as their needs grow. They can start with smaller units and add more as additional electricity becomes available or as their internal demand increases over time.

Manufacturing and deployment

Torus is leveraging its existing industrial foundation in Utah to produce the Foundry units. The company operates a 545,000-square-foot manufacturing facility called GigaOne in Salt Lake City. This facility is already used to produce energy storage systems that combine batteries with flywheel technology, providing a base for modular infrastructure production.

The company expects the deployment of these modular units to take approximately four months. This timeline is significantly faster than the years it can take to develop conventional data center campuses. By using factory-manufactured components, Torus intends to reduce the amount of civil works and infrastructure requirements typically needed for large-scale developments.

The $500 million in funding will support the production of these units and the expansion of manufacturing capabilities. While the company has not specified the exact allocation of the funds, the goal is to move from product launch to large-scale commercial deployment. Hydra Host has already been identified as one of the first customers for the Foundry technology.

Impact on the industry

The emergence of modular, power-integrated computing infrastructure could change how AI capacity is scaled in the United States. For industrial facilities and manufacturing plants, this technology offers a way to utilize underused electricity connections. It provides a plug-and-play style of expansion that avoids the local opposition often faced by massive, new industrial developments.

However, the success of the Foundry model will depend on Torus's ability to manufacture the units at scale and prove their economic reliability. The company will need to navigate various differences in electricity tariffs, cooling requirements, and local permitting frameworks across different regions. Despite these challenges, the move toward distributed data centers represents a shift toward more flexible infrastructure.

For construction and infrastructure stakeholders, this development introduces a new growth segment in the market. It moves the focus toward factory-manufactured units that integrate power storage directly into the computing hardware. This convergence of energy storage and data center construction is becoming a key trend as the AI market continues to expand.