Applied Digital returned 34% year-to-date through March while Nvidia gained 26%, marking the first sustained stretch where a pure-infrastructure play outran the chipmaker in a bull tape. The shift reflects a structural funding rebalancing: infrastructure deployment capital now exceeds foundation model R&D investment across the sector, reversing the 2022-2024 ratio.
Applied Digital operates 400 megawatts of AI-optimized data center capacity across North Dakota and Texas, selling compute by the rack to hyperscalers and national labs. The company signed 180 megawatts in new lease commitments in Q4 2024, double the 90 megawatts added in Q3, and management guided to 600 megawatts operational by end-2026. Revenue per megawatt stands at $8.2 million annually, roughly 40% above legacy colocation benchmarks due to liquid cooling and higher power density configurations.
The funding crossover matters because model companies historically commanded premium multiples on projected inference revenue, while infrastructure traded closer to industrial real estate comps. That spread collapsed in January when Anthropic delayed its Series D and three midsized model shops merged to share training costs. Meanwhile, Oracle, Microsoft, and Amazon collectively announced $210 billion in data center capex for 2025, 18% above 2024 spend. The capital is flowing to shovels, not prospectors.
Applied Digital's outperformance also reflects power procurement advantages that competitors cannot easily replicate. The company locked 200 megawatts of North Dakota wind power at $0.031 per kilowatt-hour through 2032, roughly half the $0.065 spot rate paid by peers in Virginia or Oregon. Lower power costs translate to 22% higher EBITDA margins on identical lease rates, a gap that widens as inference workloads push utilization above 85%. One hyperscaler client renewed early at $9.4 million per megawatt annually, a 15% premium to initial terms, because relocating 40 megawatts of live production traffic would cost $60 million in downtime and migration labor.
Allocators should track three follow-on signals over the next 90 days. First, Applied Digital's Q1 earnings in mid-April will clarify whether lease conversion rates hold above 70% as the pipeline matures. Second, watch whether Nvidia's Blackwell chip delays push hyperscalers to extend current H100 deployments, which would increase Applied Digital's utilization without requiring new hardware capex. Third, monitor whether any model company announces a training cluster consolidation or partnership that confirms the funding rebalancing is permanent rather than cyclical.
The sector rotation leaves $140 billion in hyperscaler-committed data center capex chasing roughly 3,200 megawatts of AI-ready capacity across all U.S. operators, a 12-month supply-demand imbalance that pricing power typically resolves in the landlord's favor.