CoreWeave Hits Profitability While Applied Digital Burns Cash Building Data Centers

CoreWeave delivered a strong quarter with $1.36 billion in Q3 revenue (up 134% YoY), positive operating income of $51.9 million, a cash balance of $1.89 billion and a revenue backlog north of $55 billion, bolstered by expanded partnerships (Meta, OpenAI) and a $6.3 billion NVIDIA collaboration. Applied Digital reported $64.2 million in Q1 2026 revenue (up 84% YoY), included $26 million from tenant fit-out services, secured a new 150 MW lease with CoreWeave at Polaris Forge 1 and is building Polaris Forge 2, but remains operating-loss making (negative operating income) with a smaller $286.2 million cash position. Valuation dispersion is notable: Applied Digital trades at roughly 50x trailing sales versus CoreWeave at ~22x, while analyst sentiment skews positive for both (CoreWeave mixed coverage, Applied Digital strong buy-side support).

Analysis

Market structure: CoreWeave (CRWV) is the near-term winner — $1.36B revenue, $55B backlog and positive operating income give it faster scaling and pricing power versus Applied Digital (APLD), which trades at 50x P/S despite smaller revenue and negative operating leverage. Hyperscalers (META, AMZN) and NVIDIA (NVDA) benefit as demand sinks into cloud compute and GPU supply agreements; Equinix/DLR face pricing pressure on new hyperscale business and longer RFP cycles. Commodity and power: rising data-center buildouts support copper, transformers and incremental utility capex; expect upward pressure on industrials and energy offtake contracts over 12–36 months.

Risk assessment: Tail risks include a demand shock (AI model spending fall >20% YoY), GPU export controls, or multi-quarter construction delays at Polaris Forge that would convert forecasted lease revenue into stranded capex. Time horizons: earnings-momentum trades play out in days–weeks, backlog conversion and lease commencements in 3–12 months, and structural profits/ROI in 2–5 years. Hidden dependencies: tenant fit-out execution, PPA/power availability, and contractor labor cost inflation can flip cashflow profiles quickly. Key catalysts: NVDA product cadence (next 90–180 days), hyperscaler capex calls, and APLD lease commencements.

Trade implications: Favor long CRWV exposure via equity or 6–12 month call-spreads sized 2–4% AUM given positive OI and $1.89B cash; avoid valuation-rich APLD equity outright and prefer guarded short or put-spread exposure. Implement a pair trade (long CRWV, short APLD) to express platform vs. builder dispersion; target relative outperformance of 20–40% over 3–9 months. Use volatility strategies around earnings: buy CRWV LEAP call spreads and buy APLD 3–6 month put spreads to define downside.

Contrarian angles: Consensus may be overestimating the security of APLD’s lease-derived revenue — execution and power constraints are underpriced and justify a much lower multiple if build delays exceed 6–12 months. Conversely, CRWV’s backlog is large but revenue depends on spot/pricing for GPU hours; a price collapse in GPU spot markets could compress margins quickly. Historical parallel: hosting booms of 2017–2019 show fast capacity addition can flip pricing in 12–18 months; watch utilization metrics, not just bookings.

Key Decisions for Investors

• Establish a 2–4% long position in CRWV via equity or a 6–12 month call-spread (buy ATM, sell 1.3x strike) within 2 weeks; close or trim if quarterly revenue growth falls below 50% YoY or operating income reverses to negative for two consecutive quarters.
• Initiate a 1.5–3% pair trade: long CRWV / short APLD (dollar-neutral) for a 3–9 month horizon to capture platform vs. builder rerating; tighten stop if APLD announces >300 MW additional firm leases or its share price rises >25% from entry.
• Buy a 3–6 month put spread on APLD sized to 1–2% AUM (20–30% width) after any post-earnings pop >5% to hedge execution and power-risk; exit if APLD confirms on-time commercial operation of Polaris Forge 1/2 within announced timelines.
• Rotate 2–4% from legacy data-center REITs (EQIX/DLR) into select NVDA supply-chain or GPU-exposed names over 3–12 months, or into CRWV, after monitoring NVDA server-GPU SKU launches in the next 90–180 days; reduce if NVDA guidance misses gross-margin thresholds by >200bps.