$76.61
Open on report
$71.88
Close on report
-6.17%
% since report
CoreWeave: Caught Between Paper and Silicon — Abelian Analysis
Caught Between
Abelian Analysis
February 27, 2026
▼ Short Position
CRWV
NASDAQ
Short-Selling Research · AI Infrastructure / Credit Analysis
Caught Between
Paper and Silicon
CoreWeave's Race Against Physical Reality
"CoreWeave is not the villain in this story. It is the victim. The problem is not the strategy. The problem is the physics of the conversion — the irreducible time between when a financial commitment is made and when a GPU starts earning revenue. Every day in that gap costs money. And the gap is growing."
— Abelian Analysis, February 27, 2026
Price Target
$45–80
vs. ~$90 current · Sep 2026 bear put spread
Days Sales Outstanding
182 days
Q4 2025 — 2+ quarters of revenue owed
Construction in Progress
$9.4B
Idle hardware earning zero revenue, +194% YoY
Total Debt
$21.4B
6.4× debt-to-equity; first maturity 2029
Q1 2026 Int. Coverage
~1.5×
EBITDA/interest; down from 2.3× in Q4 2025
FY2026 Capex Guided
$30B+
"At least" — on $5.1B in FY2025 revenue
Required Q2–Q4 Avg.
$3.5B/qtr
Best ever reported: $1.57B (Q4 2025)
MSFT Concentration
~67%
Single customer; existential if they pull back
Report Evaluation
Scored across three dimensions: Thesis Strength · Evidence Quality · Catalyst & Downside
A2 / B2 / C3
A2/B2/C3 — Governance and structural mismanagement (leveraged time arbitrage straining under declining capex efficiency, near-1× interest coverage, and a 182-day DSO; no deliberate fraud alleged — the company is doing exactly what the market asked). Documentary evidence from SEC filings cross-referenced across four entities (CoreWeave, Microsoft, NVIDIA, TSMC). Hard catalyst: September 2026 bear put spread, explicit $45–80 price target, datable Q1 (mid-May) and Q2 (mid-August) earnings triggers with specific CIP and revenue/MW thresholds defined.
A — Thesis Strength
A3Fraud / Deception
Accounting fraud, deliberate misrepresentation, knowing concealment
A2Governance / Mismanagement
Leveraged GPU REIT operating at $795K/day per $1B in idle assets; 182-day DSO on Microsoft; declining capex efficiency; ~1.5× interest coverage with $30B+ 2026 capex; market has sold every earnings report
A1Structural / Valuation
Overpricing, business erosion, macro-driven issues
B — Evidence Quality
B3Primary Investigation
FOIA, original interviews, site visits, proprietary data
B2Documentary
SEC filings cross-referenced across CoreWeave (10-Q, S-1), Microsoft (10-Q/K), NVIDIA (10-K FY2026), and TSMC purchase obligations. Earnings call transcripts and management guidance. Analytical ratios (DSO, capex efficiency, coverage, revenue/MW) derived from disclosed financials
B1Analytical
Peer comps, pattern-based inference from public data
C — Catalyst & Downside
C3Hard Catalyst
September 2026 bear put spread; explicit $45–80 price target (~50% downside from ~$90); datable Q1 (May) and Q2 (August) earnings triggers with specific numeric thresholds: CIP levels ($11B / $15B), DSO (150 days), revenue/MW ($1.86M healthy vs. $1.73M warning)
C2Soft Catalyst
Probable but undated trigger events
C1Thesis-Only
No clear trigger; long-duration short
Active Score:
A2 — Governance / Mismanagement
B2 — Documentary Evidence
C3 — Hard Catalyst
Investment Thesis
CoreWeave's business model is a leveraged time arbitrage: borrow money, buy GPUs, deploy them, generate revenue before interest eats the margin. The model has a brutal corollary — every dollar spent on GPUs starts accruing two costs the moment it leaves the building: interest on the debt and depreciation on the asset. Revenue starts only when hardware is racked, powered, connected, and contracted.
Three independent financial indicators confirm the conversion gap is real and widening: Days Sales Outstanding hit 182 days in Q4 2025; Construction in Progress — hardware bought but not earning — reached $9.4B (+194% YoY); and capex efficiency is declining from $0.31 to $0.25 of incremental revenue per dollar invested. Against this, the FY2026 guide requires a violent re-acceleration: Q2–Q4 must average $3.5B/quarter, more than double any quarter ever reported, starting from a decelerating sequential growth trend. Q1 2026 interest coverage is guided to compress to ~1.5× EBITDA — a level where a single quarter of execution miss could push coverage below 1×. A structural headwind compounds the financial strain: the AI compute market is shifting from training to inference, where cost per token falls ~10× per year and hyperscalers are building the purpose-built silicon CoreWeave does not control.
Core Findings
📭
Cash Collection
182-Day DSO: Cash Collection Structurally Broken
In Q4 2025, accounts receivable grew 7.3× faster than revenue — for every $1 of new revenue, $7.30 went uncollected. CoreWeave is extending vendor financing to some of the richest companies on the planet. Microsoft generated $81.3B in revenue last quarter; it is not short of cash. If DSO stays above 150 days through Q1 2026, this is structural, not a timing artifact — and a company servicing $21.4B in debt cannot fund six months of float for customers indefinitely.
🏗️
CIP Buildup
$9.4B in Construction in Progress — Earning Zero Revenue
CIP represents GPUs ordered but not racked, data centers under construction, hardware awaiting power and cooling. It does not generate revenue, is not depreciated — but the debt funding it accrues interest every day. At CoreWeave's blended rate, carrying $9.4B in idle assets costs ~$750M per year, or $2M per day. CIP grew +36% in a single quarter (Q3 to Q4 2025) and is +194% year-over-year. It is the financial embodiment of the conversion gap.
📉
Efficiency Decline
Capex Efficiency Declining: $0.31 → $0.25 Per Dollar
If the conversion gap were purely timing, each dollar of capex would produce at least the same incremental revenue over time. The opposite is happening: FY2025 capex produced $0.31 of incremental revenue per dollar; FY2026 capex is guided to produce only $0.25 — a 19% decline. Revenue per MW of active power capacity dropped ~20% quarter-over-quarter from Q3 to Q4 2025. New capacity is not monetizing at the rate of the existing base.
🎯
Hockey Stick
FY2026 Guide Requires Violent Re-Acceleration
FY2026 guidance of $12–13B against Q1 guidance of $1.95B implies Q2–Q4 must average $3.5B per quarter — nearly 70% of all of FY2025 revenue, every quarter for three quarters. Sequential growth has been decelerating: +81%, +21%, +15%, guided +24% for Q1. The guidance then requires this to more than triple. Q4 EPS missed consensus by 130% — actual -$1.04 vs. -$0.45 expected. The market has sold every post-earnings report.
🔗
Counterparty Risk
The Daisy Chain — Net Working Capital of −$8.95B
CoreWeave sits at the center of a web of mutual obligations where every participant depends on every other continuing to perform. Net working capital is -$8.95B. The $8.5B Meta-backed credit facility — the bull case's answer to every liquidity question — was conspicuously absent from the Q4 2025 earnings discussion. CoreWeave issued $2.6B in convertible notes instead. The Meta facility may yet close. Its absence is a data point worth monitoring.
🧠
Structural Shift
Inference Takes Over — CoreWeave Sells Training at Scale
Inference accounted for ~⅓ of AI compute in 2023, ~½ in 2025, projected ⅔ by 2026 (Deloitte TMT). Inference cost per token falls ~10× per year through quantization, distillation, and purpose-built hardware. Google's Ironwood, Amazon's Inferentia, Meta's MTIA — all hyperscalers are deploying inference silicon they own and operate. CoreWeave's $30B+ FY2026 capex buys general-purpose NVIDIA GPUs into a market where the dominant workload is migrating toward hardware CoreWeave does not control.
Quarterly Revenue: Actual vs. FY2026 Guide Implied
Source: CoreWeave SEC filings and Q4 2025 earnings release. FY2026 midpoint of $12.5B less Q1 guide of $1.95B = $10.55B remaining across Q2–Q4, requiring ~$3.5B average per quarter.
Q2 2025
$1,127M (+81% seq.)
Q3 2025
$1,365M (+21% seq.)
Q4 2025
$1,572M (+15% seq.)
Q1 2026 (guided)
$1,950M (+24% seq.)
Q2–Q4 2026 (required)
~$3,500M avg. — guidance requires +80% from Q1
Sequential growth has been decelerating. The guidance requires it to more than triple from Q1 to Q2 — beginning at a trend line that shows the opposite. Q2 2026 earnings (mid-August) is the decisive test: if it does not show $3B+, the FY2026 guide is mathematically dead.
Construction in Progress — The Conversion Gap Growing
Source: CoreWeave quarterly balance sheets (SEC filings). Q1 and Q2 2025 estimated from trend; Q3 and Q4 from quarterly disclosures. CIP = hardware purchased or in progress but not yet generating revenue.
Q4 2024
$3.2B
Q2 2025
~$4.9B (est.)
Q3 2025
~$6.9B (+41% from Q2)
Q4 2025
$9.4B (+36% QoQ / +194% YoY)
At CoreWeave's blended interest rate, carrying $9.4B in idle assets costs ~$750M per year — $2M per day — against zero revenue. Q1 2026 CIP is the single most important data point at the next earnings release: ≤$11B is healthy; above $15B is critical.
Capex Efficiency: Incremental Revenue per $1 of Capex
Source: CoreWeave SEC filings and FY2026 earnings guidance. Incremental revenue = FY-over-FY revenue growth divided by capex for the same period. FY2026E uses midpoint guidance. Capex timing lags are acknowledged; the declining trend raises questions about incremental monetization.
FY2025 actual
$0.31 incremental revenue per $1.00 capex
FY2026E guided
$0.25 incremental revenue per $1.00 capex (−19%)
Revenue per MW of active power capacity also dropped ~20% from Q3 to Q4 2025. New capacity is not monetizing at the same rate as the existing base — either due to front-loaded capex timing, pricing pressure, or utilization challenges.
The AI Financial Supply Chain — Paper vs. Silicon
Obligation Web: What Every Layer Owes the Layer Below
NVIDIA's quarterly revenue recognition is where paper becomes silicon. Everything above it is a financial claim on GPUs not yet shipped.
| Layer | What It Represents | Amount | Status |
|---|---|---|---|
| MSFT Remaining Performance Obligations | Contracted but undelivered cloud services — up from $242B seven quarters ago | $631B | Paper |
| MSFT Finance Leases | Commitments to GPU capacity providers — up from $22.9B in Q1 2024 | $60.2B | Paper |
| CRWV Revenue Backlog | Contracted future compute delivery; 42% expected within 24 months; $66.8B must pass through physical conversion before it counts | $66.8B | Paper |
| CRWV Total Debt | Borrowed to buy GPUs. $6.7B current (<12 months). $21.4B total. Accruing interest daily. | $21.4B | Paper |
| NVDA Quarterly Revenue (~90% Data Center) | The paper/silicon conversion moment — GPUs actually shipped, received, and beginning their economic life | $68.1B/qtr | Silicon |
| NVDA Purchase Obligations | Wafers ordered from TSMC; ahead of physical fabrication | $45.8B | Paper |
| CRWV Net Working Capital | Current assets less current liabilities — the acute stress point in the chain | −$8.95B | Stress |
CIP at Q1 2026 — The Single Most Important Number to Watch
✅ Healthy Scenario
$9–11B
Deployment keeping pace with $6–7B in Q1 capex. New capacity placed in service roughly as fast as procured. The time arbitrage is working — revenue ramping, gears turning.
⚠️ Warning Scenario
$12–15B
Pipeline backing up. Hardware purchased faster than deployed. Interest accruing on dead weight. Margin trough extends beyond Q1 2026. Q2/Q3 hockey stick at risk.
🚨 Critical Scenario
>$15B
System under strain. CIP would be 35–40%+ of total PP&E. FY2026 revenue guidance for Q2/Q3 in serious doubt. Financial scaffolding starts to wobble.
Company Profile
Company
CoreWeave, Inc.
Ticker / Exchange
CRWV / NASDAQ
IPO Date
March 2025
FY2025 Revenue
~$5.1B
Revenue Backlog
$66.8B
Active Power (Q4 2025)
850 MW
Research Firm
Abelian Analysis
Published
February 27, 2026
Trade Structure — Hard Catalyst (C3)
September 2026 Bear Put Spread · CRWV
Current Price
~$90
Target Range
$45–$80
Mid-May 2026 — Q1 2026 Earnings
Does the hockey stick start? Watch CIP (target ≤$11B), DSO (target <150 days), revenue/MW ($1.86M healthy vs. $1.73M warning). Q1 guide: $1.95B revenue, 0–2% operating margin.
⚡ Mid-August 2026 — Q2 2026 Earnings (Decisive)
If Q2 revenue is not $3B+, the FY2026 guide is mathematically dead. Back half would need $4B+/quarter — strains credulity. September expiry gives ~10 days for repricing after Q2 results.
Close Condition — What Would Prove This Wrong
Q1 at $2.2B+ with DSO below 120 days and CIP declining → the conversion gap is closing → close the position immediately.
The Cost of the Conversion Gap
$795K / day
Per $1B in idle assets — from the day of purchase until the GPU begins generating revenue
Interest (~8% blended)$219K/day per $1B
Depreciation (5-yr straight-line)$548K/day per $1B
Facilities & power (est.)$27K/day per $1B
On $9.4B CIP (Q4 2025):~$2M / day
"CoreWeave is not losing money because demand is weak. It is losing money because the physical conversion — racking GPUs, connecting power, onboarding customers — takes time. And time, at $795K per day per billion dollars, is expensive."
Abelian Analysis — Section III
Post-Earnings Moves — All 3 Quarters as Public Company
Q2 2025 — Aug 12
−16%
Q3 2025 — Nov 10
−21%
Q4 2025 — Feb 26
−8% (AH)
Q4 EPS missed consensus by 130%: actual −$1.04 vs. −$0.45 expected. Revenue growth is real but the cost structure overwhelms it every quarter.
"Keep CIP at $11B and the gears are turning. Let it blow past $15B and the financial scaffolding starts to wobble — and the hockey stick quietly becomes impossible."
Abelian Analysis — Section XV
Credit & Leverage Snapshot
Total debt (Q4 2025)$21.4B
Current debt (<12 months)$6.7B
Debt-to-equity6.4×
Net working capital−$8.95B
Q4 2025 interest coverage2.3× EBITDA
Q1 2026 guided coverage~1.5× EBITDA
Q1 2026 interest expense guide$510–590M
First debt maturity2029
Inference Shift — The Structural Headwind
Inference share of AI compute (2023)~33%
Inference share of AI compute (2025)~50%
Projected inference share (2026)~67%
Inference cost/token annual decline~10×/year
Builders with majority inference (2025)74%
Same metric one year earlier48%
Abelian Analysis — Conclusion
CoreWeave is not the villain. It is the canary. And the canary is singing.
▼ Short CRWV — September 2026 Bear Put Spread · $45–$80 Target
The AI infrastructure buildout is real. The demand for compute is real. The contracts are real. But between the contract and the running GPU sits a conversion gap — a physical, irreducible period during which data centers must be built, hardware must be shipped, power must be connected, and workloads must be onboarded. Across that gap, the financial obligations do not pause: interest on $21.4B in debt, depreciation on assets losing a fifth of their value every year, carrying costs on billions in hardware not yet generating a dollar. Construction in Progress at $9.4B, DSO at 182 days, and capex efficiency declining from $0.31 to $0.25 warrant scrutiny — not because demand is fake, but because the financial structure may not keep pace with the physical deployment required to service it. Q1 2026 earnings (mid-May) sets the stage. Q2 2026 earnings (mid-August) is the decisive test. If the hockey stick does not materialize then, the September put spread expires on a repriced stock.
What Could Prove This Wrong
The Hockey Stick Materializes
Management says they are "virtually sold out in 2026" with "substantially all" FY2026 capex tied to signed contracts. Q1 at $2.2B+ with DSO below 120 days and CIP declining means the conversion gap is closing. Close the position.
The $8.5B Meta Loan Closes
The facility would dramatically improve the liquidity narrative and shift market sentiment. It would not fix DSO or capex efficiency trends, but it would extend runway — and it is the single biggest tail risk to the trade.
GPU Pricing Power Increases
If next-generation hardware (Blackwell Ultra, Rubin) commands meaningfully higher per-unit pricing, revenue per MW could jump at constant deployment pace — closing the conversion gap from the revenue side rather than the cost side.
Strategic Acquisition
Microsoft or another hyperscaler could acquire CoreWeave at a premium. Probability is low given ~$50B+ market cap and $21B in debt, but it is not zero — and it would be the highest-conviction exit for MSFT's 67% dependency.
AI Demand Step-Function
A breakthrough application creating step-function compute demand raises utilization and pricing power across the fleet. This is not a bet against AI demand — it is a bet that the financial structure cannot keep pace with physical deployment at guided scale.
Delayed-Draw Structure Alignment
Credit facilities that fund capital only as data centers come online partially align cash outlays with deployment. No near-term maturity wall before 2029. The interest coverage squeeze is real, but CoreWeave is not facing imminent default — the revenue must arrive on schedule.
Disclosure & Disclaimer
Disclosure: Abelian Analysis is short CRWV via a September 2026 bear put spread. No position in NVDA, MSFT, or TSM. This is not investment advice. All data sourced from public SEC filings and earnings disclosures: CoreWeave 10-Q (Q1–Q3 2025), S-1/A (March 2025), and Q4 2025 earnings release; Microsoft 10-Q and 10-K filings (FY2024–FY2025); NVIDIA 10-K FY2026; TSMC purchase obligations from NVIDIA quarterly commitment footnotes. Earnings call guidance from CoreWeave Q4 2025 earnings call, February 26, 2026. Do your own work.