Skip to content
Tech News
← Back to articles

This startup believes smaller, modular data centers can solve AI's infrastructure bottleneck

read original more articles
Why This Matters

Crusoe's $3.9 billion raise and shift toward factory-built, modular data centers highlights a growing split in AI infrastructure between massive training campuses and smaller, faster-to-deploy inference facilities. This matters because power, labor, and construction bottlenecks have slowed AI capacity growth, and modular designs could let companies scale compute more quickly without years-long buildouts. The move also signals investor confidence in diversified AI infrastructure business models beyond giant campuses like Crusoe's Abilene site.

Key Takeaways

In context: Crusoe, the AI infrastructure company behind a major data-center campus in Abilene, Texas, has raised $3.9 billion and is expanding into smaller, factory-built facilities aimed at handling AI inference workloads. It is counting on those modular sites to add computing capacity more quickly than traditional projects, which can be held up by shortages of power, labor and construction resources. The move broadens Crusoe's business beyond massive data-center campuses and gives it another way to serve AI customers.

The funding round values Crusoe at about $30.9 billion after the investment, Chief Executive Chase Lochmiller said. Atreides Management, Valor Equity Partners, and Mubadala Investment co-led the round. Other investors include Founders Fund, TPG, and several sovereign wealth funds, including the Qatar Investment Authority.

Crusoe is known for building large-scale AI computing infrastructure, including the Abilene campus developed for Oracle, which rents computing capacity to OpenAI. The company now sees an opportunity for smaller installations that it can produce in its own facilities, ship by truck, and set up in areas with available electricity.

The units, called Spark, are intended mainly for inference. Training leading models can require massive clusters with hundreds of thousands of chips. Inference does not always require that scale, Lochmiller said.

"You don't actually need an Abilene to do that," he told The Wall Street Journal. Running inference at that scale, he added, "can be a bit of overkill."

The shift reflects a growing divide in AI infrastructure. Large campuses remain important for model training, where developers need vast numbers of Nvidia GPUs, high-bandwidth networking, and large power supplies. But inference can be spread across smaller facilities, particularly as companies look for ways to add computing capacity without waiting years for design, permitting, and construction.

Crusoe's approach is to build Spark facilities in advance, then move them to places where power is available. That could reduce the need for large construction crews and shorten the time required to bring new AI capacity online. Access to electricity and labor has become a central issue for data-center developers nationwide.

Crusoe has a manufacturing facility outside Denver that Lochmiller said could eventually produce as much as one gigawatt of Spark capacity each year. The company already operates Spark units in Reno, Nevada, using power from solar panels and former electric-vehicle batteries.

"We're going to put them everywhere," Lochmiller said.

The company is also pursuing larger developments. Crusoe initially developed eight buildings in Abilene for Oracle. It had expected to expand the campus with Oracle and OpenAI, but those companies later signed deals with other developers. Crusoe instead leased additional Abilene buildings to Microsoft.

... continue reading