Executive Summary
AI represents 19% of total data center capital stock.
Non-AI infrastructure accounts for 81% ($2.29T vs $530B).
Markets mispricing based on 2025 annual flow (58% AI) rather than cumulative capital deployed.
Annual vs Cumulative: The Critical Distinction
2025 annual spend: AI captured 58% share.
2015-2025 cumulative capital: AI represents 19% of total deployed infrastructure.
The valuation error: Markets anchor to annual flow metrics while ignoring the $2.29T capital stock already in operation.
Financial Implications
| Metric | Non-AI Infrastructure | AI Infrastructure |
|---|---|---|
| Capital Deployed (2015-2025) | $2.29T | $530B |
| Annual Revenue Generated | $900B – $1.1T | $150B – $200B |
| Revenue per $ Deployed | $0.40 – $0.48 | $0.28 – $0.38 |
| Gross Margin | 25% – 30% | 15% – 22% |
| Utilization Rate | 60% – 75% | 35% – 50% |
| Power Consumption (per rack) | Baseline | 3x – 5x higher |
| Refresh Cycle | 5 – 7 years | 2 – 3 years |
Key finding: Non-AI infrastructure delivers 30-40% higher revenue per deployed dollar. AI requires 1.5-1.7x more capital to generate equivalent gross profit.
AWS Case Study
| Period / Type | Capex Deployed | Annual Revenue (2024-2025) | Gross Margin |
|---|---|---|---|
| Traditional Cloud (2015-2022) | $180B | $85B | 25% – 30% |
| AI Infrastructure (2023-2025) | $60B | $12B | 15% – 22% |
Observation: AI infrastructure generates lower revenue and margins per dollar deployed despite higher growth rates.
ROI on AI Investment
Bottom line: AI infrastructure is profitable but capital-inefficient compared to traditional cloud.
AI generates 15-22% gross margins vs 25-30% for traditional infrastructure.
Requires 1.5-1.7x more deployed capital to produce equivalent gross profit.
Revenue per dollar: $0.28-0.38 vs $0.40-0.48 (30-40% lower efficiency).
Power: 3-5x consumption per rack, grid capacity constrained.
Depreciation: GPU refresh cycles 2-3 years vs 5-7 years traditional.
Pricing: Competitive pressure as providers race to scale.
Utilization: 35-50% vs 60-75% traditional. Variable training workloads reduce billable hours.
Investment implication: AI growth is real but structurally less efficient. Valuation multiples should reflect lower returns per deployed dollar.
Unit Economics
Revenue Efficiency
- Non-AI: $0.40-0.48 revenue per dollar deployed annually
- AI: $0.28-0.38 revenue per dollar deployed annually
- Gap: Non-AI 30-40% more efficient
Gross Margins
- Non-AI: 25-30%
- AI: 15-22%
- Capital required: AI needs 1.5-1.7x more capital for equivalent gross profit
AWS Example
- Traditional cloud: $180B → $85B revenue → 25-30% margins
- AI infrastructure: $60B → $12B revenue → 15-22% margins
Investment View
- $2.29T base generates $900B-1.1T revenue (6x larger than AI)
- Non-AI delivers superior revenue per dollar and margins
- Apply blended multiples reflecting base economics dominance
Discover more from Priory House
Subscribe to get the latest posts sent to your email.