CRWV IC Memo | IWANNAVY LAB

Executive Summary & Action Plan

Verdict

SELL (Avoid) — A financing company wearing a cloud costume: 98% take-or-pay contracts are real, but the equity is the thin slice between contracted yield and 7–15% debt cost on a fleet that depreciates faster than it re-prices. The market credits this backlog richer than Oracle's ($0.82 vs $0.64 per $1) with a worse balance sheet — that spread should close from CRWV's side.

Loading chart

Price $88.88 | Market cap $48.5B | Target $72 | Upside -19.0% IWANNAVY Fair Value · price as of 2026-07-11 close · Street consensus $142.29 (reference, 187.6% dispersion) — we are 49% below Street because most models multiply 2027 adjusted EBITDA while netting only today's debt, ignoring the ~$26B of new borrowings that must build that EBITDA

Loading chart

Abstract

CoreWeave doubled revenue to $2.08B in Q1 (+112%) on 98% take-or-pay contracts, booked a record >$40B of commitments (Meta's $21B expansion the anchor), and pushed backlog to $99.4B — yet every line below revenue got worse. Gross margin slid 73%→65.5% as data-center rent and power ramp ahead of revenue; D&A ($1.15B, 55% of revenue) and interest ($536M, 26%) turned a 49% EBITDA margin into a −36% net margin. The fleet is funded by $35.1B of debt at 7–15% with $31–35B more capex guided this year, and two-thirds of the backlog monetizes only in 2028+, on hardware two generations old by then, for three counterparties (Microsoft ~45% of revenue, OpenAI, Meta — the last now also a potential competitor). Our $72 fair value — log-median of an honest-capex DCF ($72), 9x pro-forma-debt 2027 EBITDA ($82), and $0.72-per-backlog-dollar ($69) — sits 19% below spot. Rating: SELL/avoid; a genuine beat-and-raise on 8/18 could squeeze it 15–25%, which is a reason for discipline, not ownership.

Action Plan

No margin of safety and negative asymmetry (R/R 0.23:1). Holders reduce now and into any pre-print squeeze; re-entry only at $72/$55 or after the funding-cost ratchet (DDTL 4.0-style IG deals) proves repeatable.

Loading chart

Risk/Reward 0.23:1 at market | Prob-weighted 12M return −24% (Bull 20% × +11.4% + Base 45% × −19.0% + Bear 35% × −49.4%) | Confidence: Medium

Last Four Quarters

Revenue doubled while the P&L inverted: gross margin compressed ~9pts in four quarters (DC rent +$249M, power +$85M, power-system depreciation +$60M in Q1 alone — occupancy costs ramp ahead of revenue), and the operating line flipped from +3.8% to −6.9%. The EBITDA-to-net-loss wedge is fully explained by D&A (55% of revenue) and interest on $25.1B of principal at 7–15% effective rates, with $6.1B due in the remainder of 2026.

Metric	Q2'25	Q3'25	Q4'25	Q1'26
Revenue ($B)	1.21	1.36	1.57	2.08
Gross margin	74.2%	73.0%	67.6%	65.5%
Op margin	+1.6%	+3.8%	−5.7%	−6.9%
Net margin	−24.0%	−8.1%	−28.7%	−35.6%

Loading chart

Revenue & Profit Mix

No product segments — the mix that matters is counterparty. Top-2 customers are 65% of revenue and 56% of receivables; the backlog is anchored by three names, and the contracts collateralize the debt, so customer risk and collateral risk are the same risk. Q1'26 10-Q.

Loading chart

Business Lines

Committed GPU cloud (~98% of revenue): Multi-year take-or-pay with MSFT/OpenAI/Meta/Anthropic. Must know — pricing locks at signing, so the equity is a spread trade between contracted yield and 7–15% funding; renewal pricing after 4–6yr terms is the great unknown (open-market H100 rents already fell $8→$1.80–3.50/hr).
On-demand/Flex/Spot (~2%): New inference-era SKUs. Must know — strategically important, economically marginal, and the only line directly exposed to spot GPU deflation — kept small by design.
Managed ML stack (W&B, Mission Control, SUNK): Must know — the only genuine switching-cost layer (training goodput, NVIDIA Exemplar Cloud for GB200), yet it reports no standalone revenue.
The balance sheet as a product ($36.4B net PP&E, 1GW active / 3.5GW contracted, 8GW 2030 target): Must know — each customer contract is pledged into a debt vehicle (DDTL 1.0–5.0); DDTL 4.0's $8.5B at ~5.9% (first IG-rated GPU-backed deal) vs DDTL 1.0's 15% is the funding ratchet the whole bull case quietly depends on.

IWANNAVY Fair Value

Log-median lands at $72. The methods agree unusually tightly ($69–82) because they all price the same thing honestly: backlog is gross revenue, not margin, and the EBITDA that bulls capitalize does not exist until ~$26B of additional debt builds it. The entire bull/bear gap is a disagreement about GPU replacement economics, not about demand.

Loading chart

Thesis breaker (for the SELL): a repeatable IG-rated funding ratchet (next DDTL priced near 4.0's ~5.9%, not 5.0's ~8.9%) plus backlog decisively >$100B on 8/18 with gross margin stabilizing ≥65% — that combination re-rates toward $99+ (deep-bull $152 if 58%+ margins and deleveraging compound).

Catalysts & Risks

Catalyst	Timing	Impact	·	Risk	Prob	Impact
Q2'26 print (cons. rev ~$2.55B, EPS −$1.06; backlog & exit-run-rate floor are the real lines)	8/18 AMC	High (two-sided)	·	Refinancing/interest wall: ~$25–30B external funding needed this year; +200bp = ~$700M/yr = 23% of EBITDA	35–45%	−19%+ per turn of de-rating
Meta AI-cloud competition follow-through (7/1 Bloomberg report hit stock and junk bonds; Meta can't re-market CRWV-leased capacity through 2032 per Rosenblatt)	Jul–Sep	Negative overhang	·	Counterparty concentration: top-3 ≈ 70–80% of backlog; contracts collateralize the DDTLs — a customer slip is a covenant event	25–35%	~20–30% of EV
Next debt-market access test (H2 cadence ~1 facility/qtr; spreads re-widened on the Meta report)	30–90 days	Two-sided	·	GPU economics: 6-yr depreciation vs 2-yr silicon cycle; Hopper fleet re-contracts into H1'27 Blackwell supply	40–50%	+$3–4B/yr D&A if life → 4yrs

References

CoreWeave Q1'26 results — $2.1B revenue, >$40B bookings, $99.4B backlog (IR, 2026-05-07) · Q1'26 10-Q (SEC EDGAR) — customer concentration, debt schedule, D&A/interest detail
Bloomberg — CoreWeave junk bonds slide as Meta plans competing AI cloud (2026-07-02) · Benzinga — dip-buyers return; Meta can't re-market leased capacity through 2032 (2026-07-06) · Motley Fool — Meta cloud report hits CRWV (2026-07-01)
CNBC — H100 rental rates deflate to $2–3/hr (2025-11-14)