Crusoe Energy
energy-first "AI factory" & GPU neocloud · vertically integrated power + data center + cloud
Priced primary rounds only; Seed/Series A/B valuations never disclosed (omitted). Series E is post-money on the initial close.
Earnings, margins, COGS & capex
No audited public financials (no S-1 filed as of mid-2026). Revenue is energy + data-center lease + GPU cloud; the GPU/compute line sells on-demand (~$2-3/GPU-hr) and multi-year reserved. Capital intensity is extreme: Abilene flagship carries ~$11.6B of financing including a ~$7.1B JPMorgan Phase-2 construction loan, plus Brookfield ($750M), Upper90 ($225M), Victory Park ($175M) facilities; ~$3.9B+ equity raised lifetime. The model is a levered infrastructure-developer one — returns depend on sustained GPU utilization, lease/PPA durability, and turbine + build execution, not software margin. GPU rental rates have fallen (~$8→~$2/hr historically), a structural margin risk on uncontracted capacity.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~96¢ is cost of goods and ~4¢ operating expense, leaving ~0¢ of operating profit.
Revenue trend
Margins
COGS structure
Not disclosed (private company).
Capex
Not disclosed.
Growth drivers
- Demand is contracted, not speculative — OpenAI/Oracle Stargate + Microsoft 900MW underwrite ~5GW; the question is build-out, not whether anyone shows up
- Vertical integration (power generation + data center + GPU cloud) is the moat rivals lack — Crusoe can energize a site faster than competitors can find megawatts
- Capital + talent flywheel — NVIDIA/Mubadala/Valor backing, JPMorgan/Brookfield debt, and an IPO-proven CFO position it to keep funding gigawatt campuses
- Re-rating has fundamental support — ~$2B FY2026 revenue target growing toward $3.6B-$5.5B implies a sub-20x forward sales multiple at the lower end of the pre-IPO range — rich but not absurd for the growth + asset base
Bull & bear
Crusoe owns the scarcest asset in AI — sited, powered, gigawatt-scale capacity contracted to the most important AI buyers — and its energy-first vertical integration gives it a structural speed-to-power edge that justifies the re-rating toward $30B-$40B.
- Demand is contracted, not speculative: OpenAI/Oracle Stargate + Microsoft 900MW underwrite ~5GW; the question is build-out, not whether anyone shows up
- Vertical integration (power generation + data center + GPU cloud) is the moat rivals lack — Crusoe can energize a site faster than competitors can find megawatts
- Capital + talent flywheel: NVIDIA/Mubadala/Valor backing, JPMorgan/Brookfield debt, and an IPO-proven CFO position it to keep funding gigawatt campuses
- Re-rating has fundamental support: ~$2B FY2026 revenue target growing toward $3.6B-$5.5B implies a sub-20x forward sales multiple at the lower end of the pre-IPO range — rich but not absurd for the growth + asset base
This is a hyper-levered infrastructure developer wearing a software-multiple valuation: $7.1B+ of construction debt sits against a young, concentrated revenue base, execution is already slipping, and GPU-rate deflation plus an AI-capex pause could strand capacity.
- Leverage + concentration is the kill switch: anchor delay/loss or a utilization shortfall turns the debt stack from accretive to existential
- Execution is visibly cracking — phase-2 Abilene delays past permits and a winter liquid-cooling outage undercut the 'speed-to-power, reliable-at-scale' narrative
- $30B-$40B is ~15-20x forward sales for an asset-heavy, low-software-margin business — neocloud comps (CoreWeave) trade well below that on EV/sales and carry the same rate/utilization risk
- Hyperscaler self-build is the secular threat: Microsoft/Oracle/AWS in-sourcing power + DCs erodes Crusoe's addressable demand exactly as it adds the most fixed cost
What it is worth
Pre-IPO private mark cross-checked against neocloud EV/sales comps + a reverse read on implied growth. At a reported $30B-$40B against ~$2B FY2026 revenue, the round prices ~15-20x forward sales — rich vs. the public neocloud comp set (CoreWeave/Nebius) but supported only if the ~$3.6B (2027)/~$5.5B (2028) projection and full utilization of contracted ~5GW materialize. This is an asset-heavy, levered infrastructure business, so EV/sales overstates quality vs. software; the durable question is contracted-capacity MOIC, not multiple.
$8B-$12B
at/below the Series E mark if execution delays persist, GPU rates deflate further, an anchor slips, or the credit/rate backdrop tightens, stranding levered capacity. A down-round IPO or a delayed listing is the realistic downside.
$20B-$30B
re-rated above the Oct-2025 >$10B Series E on real Stargate/Microsoft contracted demand, but discounted for leverage, concentration, and visible execution slippage; clears the prior round, lands below the top pre-IPO whisper.
$40B+
pre-IPO range holds and IPO prices a premium if anchors energize on schedule, utilization is high, and new offtakers convert the 45+GW pipeline; revenue compounds toward $5B+ by 2028.
SWOT
Strengths
- Anchor demand locked — lead developer of OpenAI's flagship Stargate Abilene campus (on Oracle Cloud) and a new ~900MW Microsoft campus — blue-chip, multi-year offtake
- Vertically integrated energy→data center→cloud — owns/sources power (gas turbines, 45+GW pipeline) giving speed-to-power that is the binding constraint for rivals
- ~5GW of contracted AI capacity and a large identified power pipeline — scarce, sited, permitted megawatts
- Top-tier capital access — NVIDIA, Mubadala, Valor, Fidelity, T. Rowe, Founders Fund equity + JPMorgan/Brookfield debt; IPO-grade CFO/COO (ex-MongoDB Michael Gordon) hired Dec 2025
Weaknesses
- Extreme leverage — $7.1B+ construction debt against a thin, young AI-only revenue base — debt service needs near-full utilization
- Severe customer concentration in OpenAI/Oracle/Microsoft — loss or delay of one anchor is existential to a campus
- Execution slipping — Abilene phase-2 buildings behind their permitted March-2026 schedule; a winter liquid-cooling failure knocked buildings offline (mid-2026) — a public reliability black eye
- Young as an AI-only company (mining was ~55% of 2024 rev, sold 2025); limited operating history at gigawatt scale; no audited public financials
Opportunities
- Power is the bottleneck of the entire AI buildout — Crusoe's energy-first integration is the differentiator hyperscalers and pure neoclouds are racing to copy
- 45+GW pipeline → optionality to add anchors (more hyperscalers, sovereign AI, frontier labs) beyond the current three
- Public listing at $30B-$40B would crystallize a premium vs. neocloud comps and fund the next campuses
- Vertical integration lets it capture power + real estate + compute margin layers in one stack rather than renting each
Threats
- GPU rental price deflation + a possible AI-capex air-pocket would strand uncontracted capacity against fixed debt
- Hyperscalers (Microsoft, Oracle, AWS, Google) building their own power + DCs could in-source what they now buy from Crusoe
- Rate/credit-cycle risk — a financing model this levered is acutely exposed to higher-for-longer rates or a credit tightening
- Power/interconnect, turbine-delivery (GE Vernova, PROENERGY), permitting, and grid/gas constraints could delay revenue while debt accrues
Moats, dependencies & bottlenecks
Moats
owns/controls generation (gas turbines, 45+GW pipeline) + data center + cloud, compressing the energize-a-site timeline that gates every rival
permitted, contracted gigawatt capacity (~5GW) — physically scarce and slow to replicate
Anchor relationships with OpenAI/Oracle/Microsoft + NVIDIA equity stake (preferential GPU + ecosystem alignment)
proven access to multi-billion equity + project debt that smaller neoclouds can't match
Dependencies
Oracle (OCI), Microsoft for contracted revenue
Brookfield, Upper90, Victory Park; equity from Mubadala/Valor/others
PROENERGY ~650MW) and gas/grid interconnection
permitting, and West-Texas power/water/cooling infrastructure
Advantages
- Energy-first DNA — started by monetizing stranded/flared gas, so power sourcing is a core competency, not an afterthought
- First-mover at true gigawatt scale via Stargate Abilene — reference site for future anchors
- Vertically captures power + real-estate + compute margin layers in one stack
- Backed by the GPU supplier itself (NVIDIA), aligning supply with demand
Weaknesses
- Capital intensity + leverage dwarf the current revenue base; little margin for execution error
- Customer concentration in 3 anchors; no diversified long tail yet
- Commodity-like exposure to GPU rental price deflation on any uncontracted capacity
- No audited financials / no S-1 yet — disclosure and governance still private-company grade
Bottlenecks
- Power generation + grid interconnect timelines — the rate-limiter on every new campus
- Construction/commissioning schedule (Abilene phase-2 already behind permits)
- GPU allocation and delivery cadence from NVIDIA
- Liquid-cooling reliability at gigawatt density (publicly demonstrated failure mode in 2026)
- Debt-service coverage requiring sustained high utilization
Top signals & trends
Top signals
A confirmed close near $30B-$40B (reported by Axios Mar 2026) validates the re-rating and funds the next campuses; a markdown or stalled raise would be a warning.
Already slipping past permitted March-2026 dates; on-time energization of 700MW+ by Dec 2026 is the credibility test.
Michael Gordon (ex-MongoDB CFO/COO) hired Dec 2025 is an explicit IPO-prep signal; an S-1 would convert rumor to fact.
Rates fell ~$8→~$2/GPU-hr historically; further deflation pressures economics on any uncontracted capacity against fixed debt.
De-risks concentration and proves the 45+GW pipeline converts to contracted demand.
Trends
Data-center power demand projected to compound mid-teens % through 2030 (Goldman); Crusoe sells the binding input.
Energy/interconnect — not chips — is increasingly the gating constraint; favors Crusoe's energy-first integration.
Neocloud rental margins compress as supply catches up; raw GPUaaS is becoming a price-competed commodity.
Microsoft/Oracle/AWS/Google in-sourcing erodes the merchant-developer addressable market over time.
A $7.1B+ construction-debt model is acutely exposed to the rate and credit cycle.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GPU supplier (GB200/H100/H200) and equity investor — the critical input + aligned backer.
Gas-turbine / power-generation equipment (~1GW) for onsite power.
Representative supplier of power/thermal/liquid-cooling infrastructure for gigawatt data centers.
Representative electrical-distribution / power-management gear for data-center buildout.
Anchor demand via the Stargate Abilene flagship (private).
Runs/leases the Abilene GPU capacity on OCI; both customer and channel.
Anchor for the new ~900MW Abilene campus (first building mid-2027).
Named GPU-cloud customer (private).
Named enterprise compute customer.
Largest pure-play GPU neocloud; crossed ~$5B revenue run-rate faster than any cloud platform; the closest public comp and EV/sales benchmark (carries similar utilization/leverage risk, less power vertical integration).
Public neocloud with large Microsoft (~$17-19B/5yr) and Meta contracts; direct GPU-cloud rival, lighter on owned power generation.
Both partner (OCI runs the Stargate Abilene GPUs) and competitor (own OCI capacity + Stargate stake) — the most important frenemy.
Anchor customer (900MW Abilene) and a self-build threat as it expands owned AI data-center + power capacity.
Private GPU-cloud competitor (raised ~$480M Series D, Feb 2025); smaller, no comparable power vertical.
Private energy-linked neocloud; direct analog to Crusoe's energy-first positioning.