Crusoe closed more than $3 billion at roughly a $30 billion valuation, nearly tripling the mark set in its Series E eleven months ago. Almost every account of the round leads with that multiple. The more useful fact is who signed the contract that made the round possible: not OpenAI, not Microsoft, not Meta, but a proprietary trading firm that sells no AI product to anyone.
Bloomberg reported days earlier that Crusoe had landed a five-year cloud agreement worth about $13 billion with Jane Street, the largest customer the company has signed for its cloud business. Jane Street already carries a roughly $6 billion commitment with CoreWeave. Its combined contracted GPU capacity across the two providers now runs to about $19 billion, which is the part nobody has stopped to examine. The neocloud bull case is built on demand from frontier labs and hyperscalers. The marquee deal underwriting the sector’s newest $30 billion valuation came from a customer class that appears in nobody’s addressable-market slide.
From Flared Gas to $30 Billion in Eight Years
Crusoe started in 2018 burning stranded natural gas at wellheads to mine bitcoin. The pitch was energy arbitrage: find power that has no route to market, put computation next to it, sell the output. When bitcoin economics deteriorated and AI training did not, the company pointed the same thesis at GPUs. That is a genuinely good business and a better founding story than most of what the sector produces, and it explains why Atreides Management and Valor Equity Partners co-led the new round with Mubadala Capital participating again.
It also explains the valuation trajectory. Crusoe raised $1.375 billion at north of $10 billion in October 2025. Eleven months later the mark is roughly $30 billion. Nothing about the underlying asset changed in that window. What changed is the contracted revenue attached to it, and the largest single contract came from Jane Street.
The Customer Nobody Modeled
Jane Street is one of the most profitable trading firms in the world and it is not building a chatbot. Quantitative trading has always been compute-hungry, but $19 billion of five-year GPU commitments is not a research cluster. It is infrastructure spending at the scale of a frontier lab, booked by a firm whose output is positions rather than models.
There are two readings, and they carry very different implications for everyone holding neocloud exposure. The generous one is that machine learning has become load-bearing enough in systematic trading that the largest players now need lab-scale capacity, in which case the addressable market for AI infrastructure is materially larger than the lab-plus-hyperscaler frame assumes, and finance is an entire demand tier nobody has been counting. The skeptical one is that a firm sitting on extraordinary profits is buying scarce capacity partly because it is scarce, and multi-year GPU contracts have become a way to lock in access ahead of a shortage rather than a signal of committed workload.
Both readings can be partly true. Neither is being tested in public, because Jane Street does not have shareholders to explain itself to, and that opacity is the actual problem. When Meta or Microsoft commits to this kind of capacity, the commitment shows up in capex guidance, gets interrogated on an earnings call, and can be checked against revenue. A private trading firm’s contract is visible only through its counterparties’ fundraising announcements.
Concentration Priced as Diversification
Our read is that the $30 billion mark is being justified with a demand story that does not match the contract book, and that the gap is not being priced.
Crusoe’s valuation nearly tripled on the strength of a customer concentration that would draw immediate scrutiny at a public company. A single five-year agreement representing a substantial share of contracted cloud revenue, from a counterparty whose financials nobody outside it can see, is a real risk, and it is the same shape of risk we flagged when CoreWeave’s $104 billion backlog turned out to double as the collateral for its debt. Backlog is being treated across this sector as though it were revenue, and contracted capacity as though it were diversified demand. It is neither. It is a promise from a small number of counterparties, and in Crusoe’s case one of them is not in the AI business at all.
The counter-argument is fair and worth stating: a five-year contract from Jane Street is better credit than a five-year contract from most AI startups, several of which are funded by the very compute providers they buy from. Jane Street pays in cash it actually earned. On counterparty quality Crusoe has done well.
But quality is not the same as breadth, and the sector keeps confusing the two. Baidu’s GPU cloud grew 283% while total revenue still fell, which is what it looks like when spectacular infrastructure growth sits on top of a business whose economics have not been settled. The neocloud category is now several years into a buildout financed against contracts, priced against a demand curve nobody has audited, and increasingly underwritten by whoever happens to have the most cash on hand in a given quarter.
What would settle it is disclosure that nobody is required to provide. Until a neocloud publishes customer concentration the way a public company must, investors are marking these valuations off press releases about deals whose durability they cannot evaluate. The next Crusoe round will arrive with a bigger number attached. Whether it arrives with a broader customer list is the question that actually matters, and it is the one the announcements keep declining to answer.