CoreWeave: Why I Am Reiterating A Buy

Reiterate Buy on CoreWeave driven by a $66.8B committed backlog and accelerating revenue conversion. Management/analyst guidance targets $12–13B in 2026 revenue, a $17–19B 2026 exit run-rate, and >$30B projected for 2027. Robust attach rates — including $100M storage ARR and ~800% attach among large customers — are cited as supporting long-term margin expansion.

Analysis

CoreWeave’s narrative creates a two-speed market: platform-native GPU clouds and the legacy colo/hyperscaler model. The non-obvious winners are upstream hardware and power-infrastructure suppliers (HBM/PSU vendors, transformer and substation contractors) because sustained GPU intensity forces multi-quarter lead times on capacity expansion; the losers are undifferentiated colo players whose pricing power is tied to rack-level commoditization rather than workload specialization.

Margin expansion is plausibly real but timing is the key variable — attach-rate economics will flow to EBITDA only as utilization, yield on specialized storage, and managed services scale; that implies meaningful margin inflection likely occurs over 4–12 quarters rather than in a single fiscal year. If customers continue to prefer Opex GPU consumption over building on-prem, the addressable margin pool shifts away from hyperscalers to specialist operators, pressuring long-cycle cloud contracts and creating durable recurring revenue.

Second-order supply-chain pressure can act as both accelerator and governor: tight GPU supply or HBM shortages can lift spot pricing and accelerate vendor FCF, but they also force customers to sign longer-term committed contracts which slow reported conversion in near-term revenue prints. Key catalysts to watch are measurable changes in GPU spot pricing, multi-quarter backlog conversion cadence, and incremental power-capacity announcements from utilities; each moves the thesis materially on a 3–12 month horizon.

The contrarian risk is that the market is underpricing the capex and operational cadence required to convert commitments into sustainable margins — if GPU commoditization or hyperscaler vertical integration accelerates, the multiple assigned to growth could re-rate sharply. Conversely, if enterprise AI workloads prove stickier and attach economics persist, upside to consensus is larger than models imply because operating leverage compounds across storage, networking, and managed services.

Key Decisions for Investors

• Long CRWV via 18–30 month LEAPS call spread (size ~2–3% NAV): buy a 24‑month slightly‑ITM call and sell a 50–70% OTM call to fund premium. R/R: asymmetric upside (2x–4x if content conversion and margins materialize) with defined downside limited to premium; roll every 6–9 months on positive catalysts.
• Pair trade — long CRWV / short EQIX (equal notional, rebalanced quarterly): duration 6–12 months. Rationale: capture GPU-native share shift vs legacy colo exposure; hedge macro data-center cyclicity. Target: 20–30% gross return if conversion outpaces commoditized colo, stop-loss if CRWV underperforms by 15% vs EQIX.
• Long NVDA call spread (6–12 months) as a hardware hedge: buy a 1–2 month OTM call and sell a farther OTM call to reduce cost. This expresses the upside if GPU pricing tightness persists; use as hedge against execution risk in CoreWeave thesis. Exit or trim on a >40% move in NVDA or if GPU spot price normalizes.
• Event hedge: buy short‑dated puts on CRWV (3–6 months) sized to cover 25–30% of the LEAPS position ahead of quarterly backlog/conversion prints. Protects against near-term execution misses that could reset multiple and gives optionality to add into weakness.