Applied Digital Stock: Valuing The AI Infrastructure Play

Applied Digital has started delivering tangible progress in its AI data-center buildout—its first 100 MW AI facility at Polaris Forge 1 received Ready-for-Service status, helping drive a near 22% five-day gain and roughly 4x YTD move for a company with ~ $9 billion market cap trading at ~33x forward revenue. The firm has a $3 billion, 280-MW Polaris Forge 2 campus under development (operational early 2027) and expanded its CoreWeave leasing to raise contracted revenue to about $11 billion; consensus expects ~38% revenue growth in 2026 and ~85% in 2027. Key upside is strong demand for GPU-optimized capacity and long-term leases, while meaningful downside risks include heavy capital intensity, execution/delay risk, high valuation leaving little margin for error, and rising competition from hyperscalers.

Analysis

Market structure: Applied Digital (APLD) is a direct beneficiary of surging GPU-driven AI demand, capturing high-margin MW leases and GPU-as-a-service revenues; primary winners are purpose-built infra owners (APLD, niche lessees like CRWV) while legacy colo/cloud (Digital Realty, core hyperscalers when building internally) face either stranded capacity or higher capex. The $11B contracted backlog and Polaris Forge pipeline tighten near-term supply for turnkey AI racks, supporting premium pricing for validated RFS campuses through 2026–2027 and creating a multi-year bottleneck in purpose-built capacity.

Risk assessment: Tail risks include (1) large execution delays/cost overruns at Polaris Forge 2 (>6–12 month slips) eroding 2027 revenue, (2) GPU export controls or NVDA supply shocks delaying deployments by >3 months, and (3) hyperscaler vertical integration reducing third-party demand by 20%+ over 2–3 years. Short-term (days–months) price action will be volatility-driven around lease/capex updates; long-term (2026–2028) fundamentals depend on PPA/energy contracts and grid capacity availability.

Trade implications: Construct small, defined-risk exposure to APLD (2–3% portfolio) plus asymmetric option structures (LEAP call spreads) to capture 2027–2028 revenue inflection while capping downside; consider relative trades long APLD vs short legacy data-center REITs (e.g., DLR) to express premium for purpose-built AI infrastructure. Cross-asset: expect incremental corporate issuance (debt) from infra builders, upward pressure on industrial metals and power-hedge instruments, and wider credit spreads for execution-risky developers.

Contrarian angles: Consensus prices rapid growth but understates grid/PPAs and GPU supply as hard constraints—if GPU deliveries slip 10–20% or power PPAs rise 15% cost-on-cost, margin compression will be sharp. The market may be overpaying near-term: favor staged entries with objective triggers (RFS confirmations, incremental lease signings) rather than full-scale buy-and-hold today.

Key Decisions for Investors

• Establish a 2.5% long position in APLD equity now to capture validated RFS momentum; set a hard stop-loss at -20% and take-profit tranche at +100% over 12–24 months tied to Polaris Forge 2 milestones (on-time H1 2027 commissioning = add 1% exposure).
• Buy a defined-risk call spread: APLD Jan 2028 LEAP buy call 30% OTM / sell call 60% OTM (size = 1% portfolio notional) to leverage 2027–2028 revenue growth while limiting premium outlay; close on any NVDA supply delay >12 weeks or if implied vol (IV) rises >40%.
• Run a pair trade: long APLD (1.5%) vs short Digital Realty (DLR) equal notional (or similar legacy REIT), horizon 9–18 months, to capture value shift to purpose-built campuses; unwind if APLD net contracted backlog falls >20% or DLR announces AI-focused capex retooling.
• Monitor these triggers in the next 30–90 days as sell/scale signals: (a) NVDA GPU shipment guidance changes >±10% (negative cut = trim APLD by 50%), (b) any single-customer counterparty (e.g., CoreWeave) covenant/credit deterioration flagged in filings, (c) Polaris Forge 2 cost overrun >15% or >6-month delay—reduce APLD exposure to zero if two occur.