Coreweave’s McBee sells $12.1m in shares

CoreWeave CDO Brannin McBee sold roughly $12.1M of Class A stock on Mar 16, 2026 under a pre-arranged 10b5-1 plan (111,734 shares directly; 21,988 via trust; 13,395 via spouse) while the stock trades at $82.82 (up 107% Y/Y, down 34% over 6M). McBee also converted 100,000 Class B to Class A shares (plus 27,085 and 16,665 via trust/spouse). Company announced a 300 MW AI data center in Regina (first phase launching next year), added NVIDIA HGX B300 chips to its cloud, and secured integrations (Cline); Oppenheimer initiated coverage with Outperform while Bernstein started with Underperform.

Analysis

The issuer sits at the intersection of specialized GPU infrastructure and developer-facing distribution — a structural moat versus generic server vendors but a narrow moat versus deep-pocketed hyperscalers. That positioning amplifies returns if enterprise AI spend re-accelerates, because incremental revenue per GPU node can be 2-4x higher than commodity cloud GPU rentals, but it also creates concentration risk if a handful of large customers re-price or internalize workloads.

Power and supply dynamics are the hidden margin lever. Large new data-center builds convert capital into a recurring power bill: every 100 MW of continuous capacity implies roughly $45–90M/year in energy expense across a $0.05–0.10/kWh range, so energy price shocks translate quickly into margin compression unless hedged with long-term PPAs or colocated cheap supply. Similarly, GPU supply cycles — memory-heavy accelerators versus older architectures — create discrete refresh points where providers with balance-sheet flexibility capture share.

Insider liquidity activity combined with governance-preserving share conversions reads as portfolio rebalancing more than a fundamental capitulation; 10b5-1 plans mute signaling. The divergence between bullish and bearish sell-side views implies a binary outcome window tied to product integrations, enterprise bookings, and next 12–18 month GPU availability — an ideal backdrop for asymmetric option structures rather than a blind long.

Short-term (days–months) catalysts include quarterly bookings and capacity-availability disclosures; medium-term (6–18 months) drivers are customer retention on new platform features and realized power costs. Binary downside is an enterprise AI demand slowdown or a sudden NVDA supply surge that collapses pricing, both of which would re-rate multiples quickly.

Key Decisions for Investors

• Tactical long (1–2% NAV) in CRWV on a pullback of 15–25% within the next 3 months; target +100% in 12 months, hard stop -30%. Trade rationale: asymmetric upside from re-acceleration of enterprise AI spend and platform wins; risk is a cyclical GPU pricing shock.
• Pair trade: long CRWV (0.75% NAV) / short SMCI (0.75% NAV), 6–12 month horizon. Expect premium capture by the specialized infra provider vs. share loss for commodity server OEMs if demand firms; cut both legs if pair diverges >25%.
• Directional NVIDIA exposure via a low-cost LEAP call spread (~0.6 delta long / ~0.3 delta short), 12–18 months, allocation 0.5% NAV. Rationale: benefits from next-gen accelerator adoption without paying for short-term vol; payoff multiple 3x+ if HGX cycle accelerates.
• Volatility-structure trade on CRWV: sell 30–45 day implied vol when IV exceeds realized vol by >20% via a defined-risk iron condor sized to 0.5% NAV. This harvests premium during the binary discovery phase while limiting tail exposure if bookings miss materially.