Crusoe has raised $3.9 billion in a Series F at a $30.9 billion valuation, ten months after raising $1.38 billion at $10 billion. The eight-year-old company will put the money into projects already running, including the large Abilene, Texas site that OpenAI uses, and into small modular "AI factories" that travel by truck and connect to heavy power almost anywhere. It also added three directors: Cloudflare finance chief Thomas Seifert, Primary Digital Infrastructure partner and chief investment officer Bill Stein, and Redwood Materials founder and CEO JB Straubel, who also sits on Tesla's board.
The valuation tripled in ten months. That is the number to sit with, because the round is not the largest figure Crusoe has announced recently. That would be the five-year cloud contract it signed with the quantitative trading firm Jane Street, worth $13 billion — more than a third of what the whole company is now said to be worth. Bloomberg reported that Crusoe will supply Jane Street with GPUs and AI infrastructure.
A quant shop as the anchor of a publicly known AI cloud contract is worth pausing on. Crusoe hosts an OpenAI site and counts Meta, Microsoft and Oracle as customers, but the single contract that has a dollar figure attached comes from finance, not from a frontier lab. Either Jane Street is buying compute at a scale that puts it in the same conversation as the model builders, or the labs are simply not signing deals they let anyone disclose. Both readings are interesting; neither is in the announcement.
The business underneath is three revenue lines: renting customers space in its data centers for their own GPUs, renting out GPUs it owns, and selling compute for running models — inference. That mix is what carried Crusoe into the most valuable tier of AI infrastructure companies.
The modular product is where Crusoe is trying to be something other than a landlord. It builds these units, called Spark, in its own factories, which lets it stand up capacity faster and without a large construction crew on site. Compact sites also blunt a problem the industry has been losing on locally: residents organizing against enormous complexes next to their neighborhoods.
Crusoe's co-founder and CEO said AI should produce an era of abundance, and that getting there means the company controls the infrastructure "from electrons to tokens," an approach he said its investors share.
The board additions read as preparation rather than governance. Last month Axios reported that Crusoe had met with investment banks including Goldman Sachs and Morgan Stanley about a possible IPO in the near future. A sitting public-company CFO and an infrastructure investor are the people you seat before an S-1, not after. Straubel is the exception and the oldest relationship: he invested personally in 2021, and Crusoe later became the first customer of Redwood Materials' energy division.
Here is what I think this round actually buys. Crusoe started in 2018 mining cryptocurrency on flared natural gas — a business whose entire premise was arbitraging stranded energy — and switched to AI infrastructure when demand for compute went vertical. The "electrons to tokens" line is a claim to have kept that origin as a structural advantage. But all three revenue lines are rental. Crusoe is paid for capacity, not for the electrons, and a tripled valuation on a rental book is a bet that the scarcity holds for the length of the leases. The Spark units are the most defensible part of the story precisely because they attack something money cannot fix quickly: siting, permitting and neighbors. Speed of deployment is a real moat when the binding constraint is not chips but the year it takes to get a building approved.
The exposure runs the other way through the customer list. Meta, Microsoft and Oracle are not just Crusoe's customers; they are the three most aggressive builders of their own data centers on earth. Crusoe is selling overflow capacity to companies whose explicit plan is to stop needing overflow capacity. That is a fine business in a shortage and a difficult one after it, and five-year contracts against GPUs that depreciate faster than buildings are where the two timelines collide.
An IPO would force the disclosure that a $30.9 billion private mark does not: how much of the revenue is contracted, for how long, and to whom. Crusoe has spent ten months tripling in value on the strength of a market that prices capacity by scarcity. Going public means being priced by cash flow instead.