Crusoe, the Denver data center developer best known for building the Abilene campus in Texas that OpenAI uses, announced on 17 September 2026 the initial closing of a $3.9 billion Series F at a $30.9 billion post-money valuation. The round was co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, and the company called it oversubscribed.
The money is aimed at two very different kinds of building. One is the giant, vertically integrated campus of the Abilene type. The other is Spark, a small modular data center that Crusoe manufactures in its own facilities and trucks to wherever power is available. Chief executive Chase Lochmiller summed up the strategy as “controlling the infrastructure from electrons to tokens”.
This article works through what was actually announced, what the numbers imply, and why the modular half of the plan may matter more than the headline figure, especially for anyone buying compute or planning around local opposition to data centers.
Table of contents
- What Crusoe Announced
- The Numbers, Worked Through
- From Flare Gas to a $30.9 Billion Valuation
- Why Crusoe Is Building Small as Well as Huge
- Board Changes Signal an IPO Path
- How the Company Makes Money
- Questions the Release Leaves Open
- Claims to Read Carefully
- What Crusoe’s Raise Means for Buyers of Compute
- Series F FAQ
- References
What Crusoe Announced
The company’s own release and TechCrunch’s report agree on the essentials. A few details in the release are easy to miss, and they change how the headline should be read.
| Item | Detail |
|---|---|
| Round | Series F, “initial closing of its anticipated $3.9 billion” round |
| Valuation | $30.9 billion post-money |
| Co-leads | Atreides Management, Mubadala Capital, Valor Equity Partners |
| Other named backers | Founders Fund, GIC, Nvidia, Qatar Investment Authority, Radical Ventures, TPG |
| Further investors | 30 more by our count, including ARK Invest, Baillie Gifford, Fidelity, Salesforce Ventures, T. Rowe Price and Tiger Global |
| Use of funds | Scale existing programmes; build campuses and modular Crusoe Spark units; grow Crusoe Cloud |
| New directors | Thomas Seifert, Bill Stein, JB Straubel |
“Initial closing” is doing work
The release describes the “initial closing of its anticipated $3.9 billion Series F”. That wording means not all of the money has necessarily landed yet. It is common in very large rounds, and nothing suggests the rest is in doubt, but “raised $3.9 billion” is the headline version of a slightly more careful statement.
A very long cap table
Beyond the nine lead and named investors, the release lists 30 further backers, from sovereign funds to crypto investors such as Polychain Capital, a retail vehicle in Robinhood Ventures Fund I, and the research firm SemiAnalysis’s investment arm. TNW had reported in July that the company was in talks to raise about $3 billion. The round closed higher.
The founding story
Crusoe was founded in 2018 as a crypto mining operation powered by natural gas that would otherwise have been flared at oil wells. It pivoted to AI infrastructure as demand for compute exploded. That origin explains the energy-first pitch: the company learned to put computers where stranded power already was.
The Numbers, Worked Through
Headline funding figures are easy to repeat and hard to interpret. A little arithmetic on the company’s own numbers makes the round clearer.
Pre-money and dilution
A $3.9 billion investment at a $30.9 billion post-money valuation implies a pre-money valuation of $27.0 billion. The new money buys roughly 12.6% of the company, which is modest dilution for a sum that large, and a sign of how much investors value what already exists.
Three times in under a year
TechCrunch notes that the company raised $1.38 billion at a $10 billion valuation “last October”, about ten months earlier. Going from $10 billion to $30.9 billion is a 3.09-fold increase. The takeaway is that investors are pricing the company on contracts signed since, not on assets built.
Contracted value versus built capacity
The company reports more than $140 billion in total contracted value and more than 6 gigawatts of gross contracted capacity, “including 1 GW of gross capacity delivered and operational today”. So about one sixth of contracted capacity, roughly 17%, is running. The other five-sixths is future construction, which is exactly what the new money is for.
Growth claims
The release also claims Crusoe Cloud bookings are up more than 20 times year on year, and that its Managed Inference product, launched late last year, has already “contracted over $100M in ARR”. TNW adds, from the Wall Street Journal, that the inference business went from almost nothing at the start of 2026 to that run rate by the summer.
One very large customer contract
Separately, Bloomberg reported a $13 billion, five-year cloud contract with the quantitative trading firm Jane Street. Averaged evenly, that is $2.6 billion a year, from a customer outside the usual AI-lab list. Its customers also include Meta, Microsoft and Oracle, according to TechCrunch.
From Flare Gas to a $30.9 Billion Valuation
The round is easier to judge against the company’s recent history. Most of the events below happened in the last twelve months, which is the real story: an infrastructure business that is being re-rated almost month by month.
| When | Event | Source |
|---|---|---|
| 2018 | Founded to mine crypto using natural gas that would otherwise be flared | TechCrunch |
| 2021 | JB Straubel invests personally | TechCrunch |
| October 2025 | $1.38 billion raised at a $10 billion valuation | TechCrunch |
| July 2026 | Reported to be in talks to raise about $3 billion | TNW |
| August 2026 | Meets Goldman Sachs and Morgan Stanley about a potential IPO | Axios, via TechCrunch |
| September 2026 | $13 billion, five-year cloud contract with Jane Street | Bloomberg, via TechCrunch |
| 15 September 2026 | Multi-year partnership with Perplexity announced | Company newsroom |
| 17 September 2026 | Series F initial close and three new directors | Company release |
The pivot explains the pitch
A business that began by moving computers to stranded gas wells learned early that power, not land or chips, is the scarce input. The same instinct runs through Spark. Instead of waiting years for a grid connection at a chosen site, the company builds the computing and takes it to the power.
Speed has been rewarded
The gap between the July report of a $3 billion raise and the September close at $3.9 billion is a sign of demand for the round. So is the sheer number of investors. In a market where many AI infrastructure deals are debt-financed, a large equity round with sovereign, crossover and retail money is a vote of confidence in the model.
The next milestone is public
With bankers engaged, a public-company audit chair recruited and a board reportedly asking for focus, the next step looks like a listing. That would bring quarterly disclosure, which would answer many of the questions this release leaves open.
Why Crusoe Is Building Small as Well as Huge
The more interesting part of the announcement is not the size of the round but the split in what it pays for. The same company that built one of the largest AI campuses in the world is now betting heavily on boxes that fit on a truck.
What Spark is
Crusoe describes Spark as modular data centers that it “manufactures in the US”, designed “for the next generation of silicon and networking systems”. Its claim is that Spark deployments “shorten the timeline of data center construction in the field from years to weeks”, and let customers “incrementally grow capacity over time as workloads and demand require”.
Training versus inference
Lochmiller explained the logic to the Wall Street Journal, as reported by TNW. Serving a model needs far fewer chips than training one. “You don’t actually need an Abilene to do that,” he said, adding that running inference from such a site “can be a bit of overkill”. As AI spending shifts from training frontier models to serving them to users, smaller sites near demand and power start to make more sense.
| Factor | Large campus (Abilene type) | Crusoe Spark modular unit |
|---|---|---|
| Typical job | Training frontier models | Inference and incremental cloud capacity |
| Build time in the field | Years | Weeks, per Crusoe |
| Where it is built | On site, with a large construction workforce | In Crusoe’s own factories, then trucked |
| Power | Grid and dedicated generation at scale | “Large power sources almost anywhere”; one site runs on solar and reused EV batteries |
| Scaling | Large steps | Add units as demand grows |
| Local visibility | High, a frequent target of opposition | Lower, though not invisible |
A factory for data centers
According to TNW’s account of the Journal interview, a plant outside Denver will eventually produce up to a gigawatt of Spark capacity a year, and units already run in Reno, Nevada, on repurposed electric-vehicle batteries and solar panels. A gigawatt a year from one factory would equal everything the company says it has delivered to date.
The community angle
TechCrunch points out that smaller centers “could also help the company sidestep, at least in part” the backlash from communities protesting “massive complexes near their neighborhoods”. That is a real advantage. It is also only partial: a cluster of containers still needs power, cooling and planning consent, and critics of data center water and energy use will not ignore them because they arrived on a truck.
Board Changes Signal an IPO Path
The same day, the company named three independent directors. Read together with recent reporting, the appointments look like preparation for public markets.
| New director | Current or past role | What they bring |
|---|---|---|
| Thomas Seifert | Chief financial officer, Cloudflare | Public-company finance; will chair the audit committee |
| Bill Stein | Former Digital Realty chief executive; partner and CIO at Primary Digital Infrastructure | Data center real estate at listed-company scale |
| JB Straubel | Founder and CEO, Redwood Materials; Tesla co-founder and CTO until 2019 | Batteries and on-site energy systems |
An audit committee chair from a listed company
TNW reports that Seifert will chair the audit committee. Recruiting a sitting public-company CFO for that role is a standard step before an initial public offering. Axios reported last month that the company had met investment bankers, including Goldman Sachs and Morgan Stanley, to discuss a potential listing.
Straubel ties power to the pitch
Straubel is not a new relationship. He personally invested in the company in 2021, and it became the first customer of Redwood’s energy storage business. Redwood supplies the microgrid of solar and repurposed EV batteries that powers Spark units. “Grid demand is growing faster than infrastructure can keep up, and that bottleneck is real for AI,” he said.
Board tension over focus
TNW, citing the Journal, adds that the strategy “has at times unsettled the board, which has suggested narrowing the focus”. Selling “data centres, GPUs and tokens”, in Lochmiller’s phrase, is three businesses with different margins and risks. A board preparing for public investors will want each to justify itself.
How the Company Makes Money
TechCrunch describes a “three-pronged business model”, and it is the key to understanding why investors are paying up.
Leasing space
The first line leases data center space to customers who bring their own GPUs. This is closest to the traditional colocation business and depends on securing power and land faster than rivals.
Renting GPUs
The second rents out the company’s own GPUs through Crusoe Cloud, where the release names Cognition, Figure and Perplexity as customers. Nvidia’s participation in the round underlines how closely GPU supply and cloud capacity are now linked.
Selling tokens
The third sells inference, the compute used to run AI models, priced by output rather than by hardware. This is the newest and fastest-growing line, and the one Spark is designed to serve. It also puts the company in competition with some of its own customers.
Why vertical integration appeals to investors
Atreides’s Gavin Baker put the thesis plainly: “As AI grows, the economics flow to the lowest-cost producer of intelligence.” Owning power, buildings, chips and software lets one company capture margin at every layer, if it can execute on all of them at once.
Questions the Release Leaves Open
Funding announcements are designed to impress, not to answer every question. Several points matter to customers, communities and future shareholders, and none is addressed in the material published so far.
How big is a Spark unit?
Neither the release nor the coverage gives the power rating, footprint or GPU count of a single Spark unit, or a price. Without those, it is impossible to compare Spark with conventional builds on cost per megawatt, or to judge how many units a gigawatt a year of factory output means.
How much of the round has closed?
The release speaks of an “initial closing”. It does not say how much has been received or when the rest is expected. For a company whose growth depends on financing construction ahead of revenue, the timing matters.
How concentrated are the contracts?
The $140 billion in contracted value is a headline total. The release does not break it down by customer or by term. If a small number of AI labs and one trading firm account for most of it, the business is more exposed to their plans than the total suggests.
How will local opposition respond?
Smaller sites avoid some objections to giant campuses, but they also multiply the number of places where a planning decision, a noise complaint or a water question can arise. Whether modular units are welcomed, tolerated or resisted will only become clear as they spread beyond the first few sites.
Claims to Read Carefully
A funding release is a sales document. Most of what the company says is specific and checkable over time, but three claims deserve a second look.
“Its first artificial general intelligence”
The release says OpenAI “trained Astra, its first artificial general intelligence, at the Abilene campus Crusoe designed and built”. That is a supplier describing a customer’s model in the customer’s most ambitious terms. Whether any system counts as artificial general intelligence is disputed, and a data center builder is not the party to settle it.
Benchmarks against vLLM
Crusoe claims its inference engine delivers “up to 9.9x faster time-to-first-token and 5x higher throughput vs. vLLM”, powered by its MemoryAlloy technology. “Up to” figures describe best cases, and the comparison is against an open-source serving stack rather than rival clouds. Buyers should test on their own workloads.
“From years to weeks”
The Spark claim is about field construction time, not total delivery time. Manufacturing, permitting, grid connection and chip supply all still take time. The comparison is fair on its own terms, but it is not a promise that capacity appears weeks after an order.
What Crusoe's Raise Means for Buyers of Compute
For most organisations, this company is not a direct supplier. Its round still says something useful about where AI infrastructure is heading, and about the choices facing any business buying cloud capacity.
Inference capacity is becoming local
If modular units can be placed near power and demand in weeks, inference capacity will spread beyond the few regions that host giant campuses. That could help organisations with data residency or latency requirements, and it is worth raising with cloud suppliers when planning cloud infrastructure.
Energy is now a buying criterion
Crusoe’s pitch starts with power, and so do its critics. Energy source, water use and end-of-life hardware are becoming questions procurement teams are asked about. We have covered both the grid problem, in Google, Nvidia and Anthropic want Emerald AI to find space on the grid, and the hardware waste problem, in the AI data center e-waste problem.
Concentration risk cuts both ways
A company with $140 billion in contracts and a few very large customers is exposed to any of them changing plans. Customers of neoclouds should apply the same diligence they would to any critical supplier, including exit plans and cybersecurity assurances, as part of normal data center operations planning.
The bigger picture
A developer valued at $10 billion last autumn is now worth $30.9 billion, is recruiting a public-company audit chair, and is building a factory to mass-produce data centers. Whatever happens to the AI boom, the infrastructure layer is professionalising quickly. For more on the companies and models driving that demand, see our AI models, tools and releases hub.
Series F FAQ
How much did Crusoe raise?
The company announced the initial closing of an anticipated $3.9 billion Series F on 17 September 2026, at a $30.9 billion post-money valuation.
Who led the round?
Atreides Management, Mubadala Capital and Valor Equity Partners co-led it, with Founders Fund, GIC, Nvidia, the Qatar Investment Authority, Radical Ventures and TPG also named.
What is Crusoe Spark?
Spark is the company’s modular data center, manufactured in its own US facilities and transported by truck to sites with available power. The company says it cuts field construction from years to weeks.
Is Crusoe going public?
It has not announced an IPO. Axios reported that it met Goldman Sachs and Morgan Stanley about a potential listing, and its new audit committee chair is a sitting public-company CFO.
Who are Crusoe’s customers?
OpenAI uses the Abilene campus. TechCrunch lists Meta, Microsoft and Oracle; the release names Cognition, Figure and Perplexity on Crusoe Cloud; and Bloomberg reported a $13 billion contract with Jane Street.
References
Modular data center (Wikipedia)
Crusoe raises $3.9B to build massive data centers and small modular AI factories (TechCrunch)
Crusoe raises $3.9bn at a $30.9bn valuation (TNW)
Crusoe and the communities where it operates
Google, Nvidia, and Anthropic Want Emerald AI to Find Space on the Grid for More Data Centers