CoreWeave's $66B Backlog Meets A $30B Reality

CoreWeave delivered $5.1B revenue in 2025, up 168% YoY, with Q4 revenue of $1.6B and a 57% adjusted EBITDA margin. Contracted backlog grew to $66.8B — more than quadrupling in one year — driven by take-or-pay contracts that provide unusually high early-stage revenue visibility. Active power capacity is 850 MW and is expected to exceed 1.7 GW by 2026, signaling rapid infrastructure scale-up that could materially affect AI/cloud compute supply dynamics.

Analysis

The dominant dynamic is demand stickiness created by long-term, take-or-pay contracts — that converts what would normally be multi-year buildout uncertainty into near-term cash-flow visibility and bargaining power. This shifts pricing leverage downstream: CoreWeave can sustain higher utilization and extract premium for guaranteed capacity, while spot GPU markets and small cloud incumbents face churn as customers prefer capacity certainty for large training runs.

On the supply side, rapid capacity expansion amplifies second-order stress on the power and interconnect ecosystem: substations, transmission upgrades, and long-lead transformers become gating constraints that create localized scarcity and higher marginal build costs. Equipment vendors and utilities with capacity to fast-track interconnects (NVIDIA/transformer suppliers, specialized EPC contractors, select utility partners) are therefore quasi-levered to CoreWeave’s growth without being direct substitutes.

Key reversal risks cluster around execution and technology. A meaningful delay in permitting or interconnects can push utilization below breakeven on new capacity within quarters; separately, a wave of model-efficiency gains or wholesale migration to proprietary accelerators by hyperscalers could reduce incremental compute demand by a material percentage over 12–36 months. Conversely, if CoreWeave converts a material share of its backlog into long-dated managed services, margins could expand further and create a high-ROI annuity profile that is underappreciated by public markets.

Key Decisions for Investors

• Long CRWV equity (6–18 month horizon): initiate position sized for 3–5% portfolio exposure. Rationale: asymmetric upside if backlog converts and utilization trends continue; set a tactical stop at -20% from entry and trim 30–50% on a 30–40% price move to lock gains.
• CRWV 18–24 month call spread (defined-risk): buy LEAP calls and sell nearer-dated calls to finance premium if available (e.g., buy Jan-2028 calls / sell Jan-2026 calls). Trade objective: capture convexity from backlog conversion while capping upfront cost; ideal risk/reward ~2–4x if runway execution is smooth.
• Pair trade — long CRWV / short EQIX (6–12 months): go overweight CoreWeave and underweight legacy colocation REITs to play structural shift to specialized GPU farms. Target relative outperformance of 15–25%; unwind if data-center fundamentals tighten broadly (macro-driven) or if CRWV margin compression appears in two consecutive quarters.
• Hedge via NVDA options (3–9 months): buy a small NVDA put spread as tail protection against a sudden tech derisk that would hit multiple AI-exposed names, or sell a limited amount of implied vol on NVDA if conviction in continued GPU tightness is high. Position size should be <50% of the CRWV exposure and used only as portfolio insurance.