Crusoe

Crusoe started in 2018 with a contrarian read on the AI boom: the bottleneck isn't chips, it's electrons. Chase Lochmiller and Cully Cavness built the company to own every layer between energy and compute, from finding stranded power to running the GPUs that sit on it.

That bet has scaled fast. Out of Denver, Crusoe now calls itself "the AI factory company," runs gigawatt-class data centers like the 1.2 GW Abilene campus behind OpenAI and Oracle's Stargate, and in late 2025 raised $1.375B at a valuation above $10B. It even sold off the bitcoin-mining business it was born from to go all-in on AI infrastructure.

Crusoe is vertically integrated in a way most cloud vendors aren't. It sources power, builds and operates hyperscale AI data centers, and rents the result as a GPU cloud, so a customer training a frontier model can lean on one company for the whole stack instead of stitching together a utility, a colo, and a cloud.

The pitch is energy-first: pair AI compute with wind, solar, hydro, geothermal, and gas so growth isn't capped by the grid. The through-line, in their own words, is going "from electrons to tokens."

AI's real ceiling right now is power and the time it takes to stand up capacity. Crusoe's answer is to attack both at once: secure large blocks of power (often stranded or new-build), then drop modular, fast-to-deploy data centers onto it.

For the teams renting the compute, the problem is different but related. Crusoe says its cloud deploys models up to 20x faster and cuts costs up to 81%, with managed Kubernetes, Slurm, and fault-tolerant AutoClusters meant to take the operational grind off AI teams. Those are the company's own figures, so read them as claims, but the underlying pain (scarce GPUs, flaky long-running jobs) is real.

This is infrastructure for people who train and serve large models at serious scale, not hobbyists spinning up a notebook. Crusoe's customers skew toward AI-native startups, enterprise AI teams, and research shops that need thousands of top-tier GPUs and can tolerate a sales-led onboarding.

That focus is also the tradeoff reviewers flag: there's no self-serve signup, so smaller teams that just want to swipe a card and go are not the target.

Crusoe's lineup stacks three layers on top of the power it controls. Crusoe Cloud is the GPU platform with NVIDIA and AMD compute, accelerated storage, and RDMA networking. Crusoe Managed Inference wraps its proprietary inference engine (MemoryAlloy) and the Intelligence Foundry for teams that want speed without running the plumbing themselves.

Beneath both sits the part few cloud vendors own: the AI data centers themselves, built modular and fast, plus managed Kubernetes, Slurm, and AutoClusters to keep large jobs alive.

Crusoe makes money renting AI infrastructure: cloud GPU capacity, managed inference, and increasingly large data-center capacity contracts with hyperscalers and labs. Because it sources its own energy and builds its own facilities, the margin story rests on controlling cost at the bottom of the stack rather than reselling someone else's cloud.

There's no public price list. Engagements run through sales with custom contracts, which fits the enterprise and lab customers it's chasing.

Crusoe sits in the "neocloud" pack (CoreWeave, Lambda, Nebius and the like) all fighting to place scarce NVIDIA GPUs, and above them the hyperscalers whose ecosystems it can't yet match. Its edge is structural: it owns the energy and the buildings, not just the servers, which is hard to copy quickly.

The flip side shows up in reviews: a thinner native integration story than AWS or GCP, and fewer GPU SKUs than some rivals. The bet is that owning power and data-center speed matters more as the AI buildout keeps hitting grid limits.

The numbers Crusoe points to are about scale of buildout, not just uptime. Its contracted AI infrastructure capacity is approaching 5 GW across data centers and cloud, its power pipeline tops 45 GW, and the first phase of the 1.2 GW Abilene campus went live roughly a year after groundbreaking.

On the cloud side it cites 99.98% uptime and reports Crusoe Cloud bookings grew about 5x across the first three quarters of 2025. It also ranked #1 on Artificial Analysis for fastest Kimi K2 inference.

The consistent praise is about the hardware and the reliability: access to scarce top-tier NVIDIA and AMD GPUs, and long training runs that stay up, with the 99.98% cluster uptime figure showing up in customer accounts.

The recurring gripes are the other side of the same coin. Crusoe isn't self-serve, so you need a sales cycle and engineers to operate it; the native integration ecosystem is thinner than the hyperscalers'; and it stocks fewer GPU types than some competitors. Glassdoor reviews split too, with good pay and benefits set against complaints about frequent pivots inside a company scaling this fast.

Crusoe's October 2025 Series E came in oversubscribed at $1.375B, pushing its valuation above $10B and co-led by Valor Equity Partners and Mubadala Capital. NVIDIA, Founders Fund, Fidelity, T. Rowe Price, Tiger Global, and Salesforce Ventures are among the backers, with Blue Owl slated for a later close.

That round sits on top of a fast climb: a $600M Series D in 2024, $350M Series C, and earlier rounds that add up to roughly $3.9B raised since 2018.

Crusoe is pouring capital into being the company that turns raw power into AI compute faster than anyone else, with contracted capacity nearing 5 GW and a pipeline past 45 GW. If AI demand keeps outrunning the grid, owning energy and data-center speed is a strong place to stand.

The risk is the same size as the ambition. This is a capital-heavy, execution-heavy bet on a market still finding its price, and the reviews already hint at the strain of scaling this fast. The next proof point is turning contracted gigawatts into live, booked, paying capacity.

Crusoe hires people who are comfortable on "a path not fully paved." Its values read like a climbing expedition: think like a mountaineer, move fast and make things, and bend energy toward sustainability. Leadership has been stacking veteran operators, adding a former MongoDB exec as COO/CFO and a Google Cloud AI leader to run product.

The work spans an unusually wide range for one company: cloud and systems engineers alongside electrical engineers, data-center construction, and manufacturing. Roles are hybrid and office-anchored across Denver, San Francisco, Bellevue, Tulsa, Dublin, and Tel Aviv, and the company is candid that teams and processes are still being built.

Crusoe discloses pay on many of its US roles, and the spread is wide because the company hires everything from staff software engineers to CNC operators and data-center construction managers. Engineering and sales roles reach the highest, with a large chunk of postings carrying disclosed yearly bands in USD.

Restricted Stock Units are included in every offer, and many roles list a bonus on top of base, so equity is a standard part of the package rather than a perk for a few.

Crusoe picked up ISO 27001 for information security and ISO 42001 for AI management in early 2026, and runs a dedicated security engineering function. The registered entity is Crusoe, Inc., and the company publishes a "Do Not Sell or Share My Personal Information" control alongside its privacy notice.

Sustainability governance is part of the same story: it ships an annual ESG/impact report and frames responsible AI as a governance commitment, not just marketing.

Crusoe's newsroom reads like an energy company's as much as a cloud's. Recent months brought a 900 MW Abilene campus to support Microsoft, a ~750 MW power deal with Bergen Engines, a 12 GWh iron-air battery agreement with Form Energy, and a deeper NVIDIA collaboration across the full AI factory stack.

The headlines that travel furthest, though, are the money and the milestones: the $1.375B Series E at a $10B+ valuation, the Abilene campus going live, and a Fast Company "Most Innovative" nod in 2026.