31Views
CoreWeave (CRWV) — AI Cloud Infrastructure Deep Dive
CoreWeave is the leading purpose-built AI cloud, renting GPU compute to hyperscalers and AI labs under multi-year take-or-pay contracts. Roughly 98% of Q1 2026 revenue is committed by Microsoft, OpenAI, Meta, and Anthropic. The company IPO'd on Nasdaq on March 28, 2025 and joined the Nasdaq-100 on June 22, 2026. Revenue grew +168% from 2024 to 2025 ($1.9B → $5.1B) with 2026 guidance of $12–13B — implying another ~140%+ growth year. The core debate: does the $99.4B contracted backlog justify $35B in debt at 738× debt/equity?
📈 Price Action: IPO to Present
CRWV IPO'd at ~$40 in late March 2025, rallied violently to a $187 all-time high in late June 2025 — a 4.7× move in 90 days. The stock then entered a 5-month decay to a $63.80 low in Dec 2025 as earnings missed badly (Q1 & Q2 2025). A recovery phase took hold in early 2026, consolidating in the $75–$110 range. Key technical signal: the SMA-50 (orange) just staged a golden cross above SMA-200 (green) in June 2026 — the first bullish MA crossover since the IPO. RSI is cooling at ~40 after a mid-June spike, suggesting consolidation before the next move. Volume surged on the June Nasdaq-100 inclusion. Current price ($95.51) sits near the 200-day MA (~$100.7), a critical battleground.
💰 Revenue Streams & P&L
CoreWeave's sole revenue stream is GPU compute-as-a-service, sold under multi-year take-or-pay contracts. Revenue exploded from $229M (2023) → $1.9B (2024) → $5.1B (2025). Gross margins are exceptional at 65.5% — matching hyperscalers — because CoreWeave has vertically integrated its infrastructure. EBITDA turned strongly positive ($2.4B in 2025, 47% margin), proving the unit economics are sound. The problem: $2.45B in depreciation (GPU refresh cycles) plus $1.3B in interest expense is what pushes net income to –$1.17B. As the debt matures and capex moderates post-build-out, the path to GAAP profitability becomes clearer — but the timeline remains uncertain.
🏦 Debt, Capital Structure & Cash Flows
CoreWeave's balance sheet is the most contentious aspect of the bull/bear debate. Total debt hit $35.1B as of Q1 2026 (MRQ), with a 738× debt/equity ratio — extreme by any standard. However, all debt is backed by contractually committed, investment-grade customer obligations. The DDTL (delayed draw term loan) cost of carry has fallen from 15% to ~5.9%, and new unsecured notes price at ~8.9%. Operating cash flow of $3.1B (2025) demonstrates the business generates real cash; the problem is CapEx devoured $10.3B in 2025 to build out GPU clusters for contracted demand — creating –$7.25B free cash flow. New long-term debt of $11.8B funded the gap. As build-out matures toward contracted capacity, CapEx should normalize and FCF turn positive.
📊 Valuation & P/E Analysis
Traditional P/E is meaningless for CRWV — trailing P/E is –34.7× and forward P/E is –98× because the company is intentionally pre-profit. The relevant multiples are EV/Revenue (13.7×), P/S (8.4×), and EV/EBITDA (28.3×). On EV/EBITDA, CRWV trades at a discount to AI cloud peers (~35×). On EV/Revenue (13.7× TTM), it is slightly expensive vs peers (~11×) — but the 2026 guide of $12–13B implies a forward EV/Revenue of ~6.6–7.2×, which is a material re-rate if management delivers. Cantor Fitzgerald carries a $167 price target. One bull cited a forward EV/S of 6.93× vs peer Nebius at 17.34×, implying ~150% upside. The EPS trend has shown improving accuracy — Q1 2026 delivered its first beat (+$1.22 actual vs $1.11 estimate).
🟢 Bullish Thesis
1. $99.4B Revenue Backlog is the Moat. 98% of revenue is contracted take-or-pay. $21B from Meta alone. 10 customers committed to $1B+. 75%+ of the $30B+ 2027 run rate is already under contract — this is not speculative demand.
2. Nasdaq-100 Inclusion = Forced Buying. Added June 22, 2026. Passive index funds must hold it. Mechanically reduces volatility and expands institutional ownership.
3. NVIDIA Partnership is a Structural Advantage. NVIDIA holds 11% equity stake. CRWV was first to deploy Blackwell and Blackwell Ultra commercially. Priority access to Vera Rubin (H2 2026) could extend the moat by 18+ months over non-partner neoclouds.
4. Debt Cost Improving. DDTL carry down from 15% → 5.9%. Investment-grade ratings allow re-financing. As interest expense normalizes, the path to GAAP profit clears rapidly.
5. Technical Golden Cross. SMA-50 crossed above SMA-200 in June 2026 for the first time — historically a strong medium-term bullish signal. Price near 200-day MA offers a favorable risk/reward entry zone.
6. 2026 Guidance = Re-rating Catalyst. If $12–13B revenue is achieved, forward EV/Revenue compresses to ~6.6×, well below peers. Every beat resets the multiple conversation.
🔴 Bearish Thesis
1. $35B Debt at 738× D/E is Existential Risk. If AI capex demand softens or contracts don't renew, the debt-funded GPU fleet generates no revenue but full interest cost. One large customer non-renewal could cascade.
2. Circular Financing. NVIDIA's 11% stake + hardware supply relationship creates a co-dependency. If NVIDIA raises GPU prices or reallocates supply, CRWV's cost structure breaks. Bulls' 'moat' is partly the vendor's moat.
3. Hyperscaler Capex Risk. Microsoft, Meta, Google, and Amazon are racing to build their own GPU infrastructure. Hyperscalers historically offload capex to neoclouds temporarily while scaling — CoreWeave's backlog could represent exactly this phase, not a permanent relationship.
4. Negative FCF Into Perpetuity. –$7.25B FCF in 2025, –$5.95B in 2024, –$1.1B in 2023. The trend is worsening, not improving. Each new GPU cluster requires fresh debt.
5. EPS History is Volatile. Q1 & Q2 2025 missed by –577% and –186% respectively. The company's own guidance proved wildly unreliable. Q4 2025 also missed. Only Q1 2026 beat. Pattern-reversal too early to confirm.
6. Price Already Down –49% from ATH. Despite positive narratives, the stock collapsed from $187 to $64 in 5 months. The $99.4B backlog was public during that collapse. Sentiment can overwhelm fundamentals at this valuation.
⚖️ Bottom Line
CoreWeave is a high-conviction binary bet on AI infrastructure permanence. The bull case requires: (1) hyperscaler AI spending remaining elevated through 2027+, (2) debt refinancing at lower rates as revenue scales, (3) CapEx normalizing as contracted clusters are built out. The bear case only needs one of: customer non-renewal, NVIDIA supply disruption, or a credit market freeze. At $95.51 (~6.6× forward EV/Sales on 2026 guidance mid-point), the market is pricing in partial backlog execution with meaningful execution risk — not full backlog conversion. The first Q2 2026 earnings print (expected ~August 2026) will be the single most important data point: does revenue track toward the $12–13B guide or does the miss pattern re-emerge? NFA, DYOR.