Crusoe raises $3.9 billion to build its own AI factories

Crusoe has banked $3.9 billion in a Series F round that values the AI infrastructure company at $30.9 billion after the money, and it has been open about what the cash is for. The round closed on 17 September and was co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, with existing and new backers including Founders Fund, GIC, Nvidia, the Qatar Investment Authority, Radical Ventures and TPG.

The company builds what it calls AI factories rather than plain data centres, and the money is going into both ends of that phrase. Most of the capital scales existing campuses; the rest pays for units it calls Crusoe Spark, modular power-and-compute blocks that can be trucked to a site instead of waiting for a transmission line.

Electrons to tokens

“Getting there means controlling the infrastructure from electrons to tokens,” chief executive Chase Lochmiller said in the announcement, framing vertical integration as the whole strategy. In practice, that means Crusoe buys or develops its own power generation, builds the shell, racks the GPUs and then sells the compute through Crusoe Cloud. Atreides managing partner Gavin Baker put the investment case in a single line: as AI grows, the economics flow to the lowest-cost producer of intelligence.

The numbers attached to the round are the interesting part. Crusoe reports more than $140 billion in total contracted value across the platform and over 6GW of gross contracted capacity, of which 1GW is delivered and operational today. Bookings on Crusoe Cloud are up more than 20 times year on year so far in 2026, and the managed inference service it launched late last year has already contracted more than $100 million in annual recurring revenue.

A power company wearing a cloud company's coat

The pitch is not subtle, and it is not new: the bottleneck in AI is no longer chips but electricity and the buildings that turn it into tokens. Conventional developers pick a site and then negotiate for power. Crusoe's argument is that starting with the power is the only way to keep the maths working when every hyperscaler on the planet is hunting the same substation.

Our opinion

Investors throwing $3.9 billion at a company that owns its own generators tells you where the industry thinks the margin lives. It is not in the model weights and, increasingly, it is not in the GPU either. It is in whoever can put a substation next to a field of accelerators and sell the result at a price that undercuts a hyperscaler. Crusoe has the contracts and the story to match. The risk sits in the phrase “anticipated”: this is an initial close on a round that is not finished, and 6GW of contracted capacity is a promise about the future rather than a description of the present. The company still has to build it, and the last three years have shown that power projects are where AI timetables go to die.