Snapshot

  • Status: Private — no 13F position; this is a private book holding
  • Sector: AI datacenters / energy-first compute infrastructure (“AI factories”)
  • Latest known valuation: **>1.375 B oversubscribed Series E, initial close announced Oct 24, 2025
  • Atreides involvement: Confirmed — named participant in the Series E (Crusoe press release). Not named in any earlier round’s disclosed investor list.
  • HQ: Denver, CO (founded 2018 by Chase Lochmiller, CEO, and Cully Cavness, President/COO)
  • One line: builds and operates gigawatt-scale AI datacenter campuses where stranded or fast-to-contract energy is, rather than where the grid is congested — anchor developer of the Stargate campus in Abilene, TX.

Business Overview

Crusoe began as a flared-gas Bitcoin miner — modular datacenters parked on oil wells, converting waste methane into compute. That origin business was sold to NYDIG in March 2025 (including the digital flare mitigation technology and 425+ modular datacenters), completing a deliberate pivot to a pure-play “AI factory company.” By 2024, AI cloud was already 45% of revenue (276 M total); Sacra estimated revenue would reach ~$998 M in 2025 on the back of the Abilene buildout (~estimates, not audited figures).

The flagship asset is the 1.2 GW Abilene, Texas campus — the first operational Stargate site, developed by Crusoe and financed through a $15 B joint venture with Blue Owl Capital and Primary Digital Infrastructure, with Oracle as tenant serving OpenAI. First GB200 racks were delivered mid-2025; the full eight-building, 1.2 GW second phase targets energization in mid-2026. The model has since replicated: a 900 MW adjacent Abilene campus for Microsoft (announced Mar 2026, with onsite power plant) and a 1.8 GW campus in southeast Wyoming with Tallgrass (announced late 2025, designed to scale toward 10 GW). Crusoe Cloud, the company’s own GPU cloud, reported bookings up 5x in the first three quarters of 2025 versus the prior year.

Crusoe’s edge is the energy layer: it co-develops generation (gas turbines, behind-the-meter power) alongside the datacenter shell, compressing time-to-power — the binding constraint in the AI buildout. Note one 2026 wrinkle: Oracle/OpenAI dropped a planned 600 MW Abilene expansion, and Crusoe re-leased that capacity (reportedly with NVIDIA paying a $150 M deposit and Meta in talks to take it) — evidence both of tenant-concentration risk and of how liquid gigawatt-class capacity currently is.

Atreides’ Involvement

  • Series E, Oct 2025 — confirmed. 10 B, co-led by Valor Equity Partners and Mubadala Capital; Atreides Management is explicitly listed among participants alongside Founders Fund, NVIDIA, Fidelity, Altimeter, T. Rowe Price and others. Size of Atreides’ check not disclosed.
  • Series D, Dec 2024 (2.8 B, led by Founders Fund) — no evidence of participation. The disclosed investor list (Fidelity, Long Journey, Mubadala, NVIDIA, Ribbit, Valor) does not include Atreides. Any claim that Atreides entered at the Series D is unsupported.
  • Vehicle — inferred, not disclosed. Plausible carriers are the Valor Atreides AI JV (Valor co-led the round; the ~55.5 M, Form D filed Nov 2025) is the closest timing match to the October close. Underlying assets of these vehicles are never disclosed; treat the mapping as inference.

Why Atreides Owns It

Crusoe is the purest private expression of the “watts” half of Baker’s “watts and wafers” thesis (Invest Like the Best EP.473, May 2026): the AI buildout is constrained by power and fab capacity, not by demand. On the public side Atreides plays this through Vistra, CoreWeave, and the optics/interconnect complex (see themes); Crusoe is the same bet one layer earlier — a developer that originates watts rather than renting them, capturing development economics on gigawatt campuses with hyperscaler-grade tenants (Oracle/OpenAI, Microsoft). It also fits the crossover-flywheel pattern: like CoreWeave and Astera Labs before it, Crusoe is a plausible IPO candidate where a private position converts into a day-one public one. The Valor relationship (JV partner, Series E co-lead) likely provided the access.

Risks

  • Tenant concentration / counterparty churn: the Oracle–OpenAI 600 MW expansion cancellation at Abilene shows offtake can evaporate mid-buildout; backfilling depends on continued hyperscaler capex appetite.
  • Leverage and project-finance stack: $15 B+ JV structures (Blue Owl, Primary Digital) layer debt-like obligations ahead of equity; an AI capex pause would hit development-stage equity hardest.
  • Power execution: gas-turbine procurement, interconnect queues, and water/permitting in Texas and Wyoming are all on the critical path; delays compress returns.
  • Valuation re-rating: 10 B in ten months prices flawless execution; late-2025 entry leaves Atreides with little cushion versus earlier investors.
  • Commodification: neocloud/datacenter capacity is scaling industry-wide; if time-to-power normalizes, Crusoe’s edge narrows toward a cost-of-capital business.
  • Opacity for LPs: exposure sits in undisclosed vehicles (JV/SPVs); no public mark until an IPO or secondary event.

Sources