Neocloud Top Customers and Counterparty Risk

CoreWeave, Nebius, Crusoe, Lambda, and Nscale (the leading neoclouds) exhibit high customer concentration in long-term GPU capacity contracts with a handful of hyperscalers and AI labs, primarily Microsoft, Meta, OpenAI, Anthropic, and increasingly trading firms like Jane Street. These take-or-pay or similar committed-revenue structures provide revenue visibility but create material counterparty risk tied to a small number of names. Public filings (CoreWeave 10-K/10-Q, Nebius 6-K equivalents, Nscale S-1) and press reporting on private entities reveal the details; private companies disclose far less.[1][2]

CoreWeave (Nasdaq: CRWV) has the most transparent and relatively diversified profile among peers, with Microsoft’s share of revenue declining from 67% in 2025 to top-three quarterly shares of 36%/26%/10% in Q2 2026, while >75% of committed capex through 2027 ties to investment-grade or enterprise customers. Its ~$104B Q2 backlog (rising to $132.5B by end-Q3) and RPO of ~$103.7B rest on multi-year take-or-pay contracts (tenors to 2031–2032) with named deals including OpenAI (up to $22.4B total across tranches through ~2029–2031, plus $350M equity investment), Meta ($14.2B initial + $21B expansion through 2032), Anthropic (multiyear, undisclosed value), and Jane Street (~$6B). Contracts emphasize committed payments regardless of utilization, with increasing customer prepayments and shorter-term elements to accelerate cash flow; termination or step-down provisions are not heavily detailed publicly but appear tied to delivery milestones or material breach rather than easy walkaways. Microsoft and Meta are investment-grade; OpenAI carries high valuation support ($1.25T post recent $11B tranche in an up-to-$20B round) but remains a cash-burning private entity whose commitments are backed by equity raises and partner financing.[3][2][4][5]

This structure gives CoreWeave strong lender confidence (e.g., investment-grade-rated facilities) and utilization visibility that pure on-demand players lack, though heavy initial Microsoft dependence created volatility. For new entrants, the bar is replicating these anchor contracts with IG names while scaling delivery without the same leverage.

Nebius (Nasdaq: NBIS) mirrors CoreWeave’s public-disclosure advantages but with even tighter concentration—roughly 80% of its ~$40B+ customer commitments (and the bulk of its RPO/backlog) from Microsoft (~$17.4B–$19.4B through 2031) and Meta (~$3B initial + $12B+ expansion, up to ~$27B total possible). Deals feature dedicated capacity with take-or-pay-like elements (Microsoft pays whether or not capacity is used) and long tenors; only ~36% of backlog converts in the next 24 months, reflecting front-loaded buildout. Microsoft tranches have been delivered on schedule; Meta capacity ramps in 2027. Both counterparties are investment-grade, reducing credit risk relative to labs, though execution on $20–25B 2026 capex remains the binding constraint.[6][7]

Nebius benefits from the same visibility as CoreWeave but less diversification so far. Competitors must match this hyperscaler anchor quality while proving faster or cheaper delivery to win share.

Crusoe (private) reports the largest headline contracted value (> $140B) and capacity (6GW+ gross contracted, 1GW operational) across data-center development and its Crusoe Cloud platform, with notable exposure to Oracle/OpenAI (Stargate Abilene campus), Microsoft (new 900MW Abilene campus), Google (Texas and prior Wyoming site work), Meta, and Jane Street ($13B five-year cloud contract for GPU clusters). Revenue and per-customer splits are undisclosed; the model integrates power generation/procurement with compute, shortening timelines versus pure developers. Contracts appear long-term and committed (supporting project finance), with cloud elements adding Jane Street (a sophisticated, well-capitalized trading firm) as a high-profile non-lab/hyperscaler customer. Credit quality is mixed—hyperscalers strong, OpenAI/Oracle ecosystem supported by massive partner financing—but private status limits visibility into termination clauses or exact take-or-pay strength.[8][9]

Crusoe’s vertical integration (power + sites + cloud) differentiates it and supports higher contracted value, but reliance on a few marquee projects amplifies single-site or single-tenant risk. New players need comparable power advantages or must accept higher financing costs.

Lambda (private) shows the highest reported single-customer concentration, with an estimated ~$50B backlog where ~70% (~$35B) ties to Anthropic; it also holds a multibillion-dollar Microsoft deployment agreement. On-demand pricing is competitive, but the shift to large reserved clusters for labs and hyperscalers drives the backlog. Financing (e.g., $1B+ delayed-draw facilities backed by IG offtakers) mirrors peers, but extreme Anthropic dependence creates outsized risk if that relationship shifts. Contract details (tenor, take-or-pay strength) are sparse.[10][11]

Lambda illustrates the risk of betting heavily on one frontier lab; entrants should prioritize diversification or ironclad multi-year commitments from multiple IG names.

Nscale (S-1 filed for NYSE listing) carries the most extreme concentration and conditionality: $103.4B total contract value (TCV) at end-Aug 2026 ($2.6B active), with Microsoft (~$43.8B through 2033) and Anthropic (~$44.6B) comprising ~85%. Weighted-average contract life ~5.7 years; agreements are long-term take-or-pay style but explicitly subject to delivery, service availability, and (for Anthropic) Nscale securing financing—Anthropic can cancel on stringent milestones. Only a small fraction is live; H1 2026 revenue was $140.6M (largest customer 52%). Nvidia participation in financing adds another layer.[12][13]

This profile signals the highest near-term execution and financing risk among the group.

OpenAI’s own funding position (recent $11B tranche at $1.25T valuation in an up-to-$20B round, following $122B at $852B) underpins its commitments but highlights reliance on equity raises and ecosystem financing rather than standalone credit. Its major deals include Microsoft (hundreds of billions incremental Azure), Oracle ($300B over ~5 years), CoreWeave ($22.4B), Amazon ($38B + $100B), and others—spread across providers but executed via partner balance sheets and debt (e.g., Oracle/CoreWeave/Crusoe facilities). High valuation and backers (SoftBank, MGX/UAE, Microsoft, Nvidia, etc.) support continued spending, yet cash burn means commitments depend on successful raises and utilization.[14][5]

Counterparty risk ranking (lowest to highest risk, based on concentration, disclosure, customer credit quality, contract enforceability/conditionality, and diversification):

1. CoreWeave — Public filings, declining top-customer share, heavy IG exposure, strong take-or-pay visibility.

2. Nebius — Similar public transparency and IG anchors (Microsoft/Meta), though less diversified.

3. Crusoe — Broad customer mix (including Jane Street) and vertical integration, but private with less granular data.

4. Lambda — High single-lab (Anthropic) concentration.

5. Nscale — Extreme two-customer dominance (~85%), many contracts conditional on unproven financing/delivery.

Overall, neoclouds mitigate utilization risk via committed contracts but inherit concentrated credit and execution risk; success hinges on delivering capacity on time to IG or well-funded counterparties while broadening the base. New entrants face a high bar: winning similar anchors requires proven scale, power access, and relationships that the current leaders already hold. Data is current as of November 2026 filings and reporting; private-company details remain estimates from secondary sources.

---

Recent Findings Supplement (November 2026)

CoreWeave shows the clearest recent improvement in customer diversification among public neoclouds, with filings indicating top-customer revenue share dropping sharply while maintaining heavy reliance on take-or-pay contracts from Microsoft, direct OpenAI deals, and Meta.[1][2]

• In Q2 2026 (ended June 30), the top three customers accounted for 36%, 26%, and 10% of revenue respectively (down from 45% for the largest in Q1 and 67% for Microsoft in FY2025); the top two combined for ~63% of H1 revenue. No single customer exceeds 35% of the ~$104B backlog (improved from 50% prior quarter and 85% earlier).[1][2]
• Key disclosed commitments (post-May updates in filings): OpenAI cumulative up to ~$22.4B across multiple agreements (including ~$6.5B through May 2031); Meta ~$21B through December 2032 (updated March 2026 order form); Jane Street ~$6B (April 2026). Microsoft remains a major anchor but is no longer the dominant single share.[1][1]
• Contracts are predominantly long-term take-or-pay with committed capacity; >75% of committed capex through 2027 tied to investment-grade or enterprise customers. Backlog expected to recognize ~41% in the first 24 months post-Q2 2026.[2][1]
• ~$35B debt is linked to subsidiary collateral, parent guarantees, and contractual cash flows from these customers. Q3 2026 results (reported around Nov 10) showed continued growth with raised full-year guidance.[3][4]

This reduces single-counterparty exposure compared to prior periods but leaves material concentration risk; competitors or new entrants face similar execution hurdles in securing equivalent long-tenor IG-backed deals.

Nebius remains heavily anchored to two hyperscale customers (Microsoft and Meta), representing the bulk of its ~$40B+ commitments, with recent Microsoft scrutiny introducing new execution and renewal risks.[5]

• Microsoft contract (~$17.4–19.4B through 2031, expandable) tranches fully delivered by Q2 2026; Meta deals total ~$3B (Nov 2025) + ~$12B dedicated (from early 2027) with up to $15B optional, for ~$32B firm (~80% of total commitments).[5][6]
• Q2 2026 revenue $582.3M (+454% YoY), AI cloud ARR $3.0B; backlog/RPO heavily weighted to these two (e.g., ~$20B of end-2025 RPO from MSFT/Meta alone). Additional >$1B deals signed in Q2 with others (e.g., Reflection, Cohere).[7][8]
• Planned 2026 capex $20–25B; Microsoft reportedly tightening oversight of neocloud projects (including Nebius sites) and considering fewer new partnerships amid delays and internal build priorities (Nov 2026 reports). Meta capacity ramps in 2027.[9][10]

High dependence on these two IG counterparties provides visibility but amplifies risk from any Microsoft pullback; entrants would need similar hyperscaler-scale contracts to match.

Crusoe (private) centers on large-scale Oracle/OpenAI Stargate infrastructure alongside Microsoft and reported Meta/Jane Street deals, with >$140B claimed contracted value but execution tied to power and build delivery.[11]

• As of Sep 17, 2026: >6 GW gross contracted capacity (1 GW operational); flagship 1.2 GW Abilene campus for Oracle/OpenAI (618 MW delivered by Sep 2026 across buildings). Separate 900 MW Microsoft campus in Abilene; reported Meta and ~$13B Jane Street cloud contracts.[12][13][14]
• $3.9B Series F at $30.9B post-money valuation (Sep 2026). Recent sale of Wyoming site to Google after failing to secure anchor tenants there (Nov 2026).[7][15]

Oracle/OpenAI exposure via Stargate creates intermediary risk; direct hyperscaler or lab diversification is narrower than CoreWeave/Nebius based on public details.

OpenAI’s funding position strengthened with a November 2026 close but carries revenue-run-rate revision and flexible contract structures across providers.[16]

• Closed first ~$11B tranche of ongoing round at $1.25T valuation (Nov 6, 2026; potential total ~$20B); prior $122B round closed March 2026 at $852B. Annualized revenue run-rate revised to ~$50B (Sep 2026, down from $68B figure).[16][17]
• Major commitments (estimates/aggregates): Oracle ~$300B/5-year Stargate (multiple tranches); Microsoft incremental ~$250B Azure; CoreWeave up to $22.4B; Amazon $38B + $100B; others (Broadcom custom chips, etc.). Total infrastructure estimates vary widely ($400B–$750B+ range cited) with noted flexibility, cancellations, or delays possible.[18][19][20]
• Providers like Oracle and Microsoft have disclosed OpenAI as material but not dominant in their backlogs/RPO; CoreWeave’s share is smaller (~20% of its backlog at most).[18]

Strong equity raises support credit quality for counterparties, but private status, revenue adjustments, and contract optionality add uncertainty versus pure IG hyperscalers.

Sector-wide, neocloud counterparty risk ranks lower for those with demonstrated diversification and IG hyperscaler anchors (CoreWeave edging ahead on filings transparency and concentration improvement) versus heavier OpenAI-intermediary or single-site exposure (e.g., Crusoe); Microsoft’s reported neocloud review adds broad renewal/expansion risk.[21]

• Take-or-pay terms and multi-year tenors (often to 2031–2032) provide backlog visibility across players, but capex intensity (e.g., 3x+ sales for CoreWeave) and power/permitting execution remain key variables.[2]
• Miners pivoting (IREN, APLD, etc.) often act as landlords or hosts with contracts flowing to CoreWeave or hyperscalers, layering additional intermediary risk.[22]
• New or smaller entrants face high barriers: securing equivalent long-tenor, take-or-pay deals from Microsoft/Meta/OpenAI-scale counterparties is rare, with recent data showing concentration even among leaders.

These developments (primarily Q2–Q3 2026 filings, Sep–Nov 2026 announcements and funding) highlight improving but still concentrated visibility, with OpenAI’s position bolstered by capital raises yet subject to scrutiny on spend sustainability.