Infrastructure Economics
The Infrastructure Cycle
Every American infrastructure — electricity, railroads, telephone, broadband, cellular, cloud — has followed the same five-stage economic cycle. Each completes faster than the last.
15
Every American infrastructure — electricity, railroads, telephone, broadband, cellular, cloud — has followed the same five-stage economic cycle. Each completes faster than the last.
This is a base rate, not a law. The telegraph was subsumed before reaching Stage 3. The internet backbone consolidated through bankruptcy, not regulation. Some cycles bypass the bargain entirely. The pattern establishes what to expect. The deviations reveal what's structurally different about each era.
Each cycle completes faster because capital formation is faster, political economy is better understood, and regulatory institutions have precedents from prior cycles. The structure is invariant. The clock accelerates.
Regulatory compression is guaranteed, gradual, and floor-bounded. Regulators set a return ceiling; efficiency gains get confiscated; but returns never reach zero because capital won't flow in at zero return. Allowed ROE: 14% in 1980, 10% today. Predictable. Slow. Durable.
Competitive compression is contingent, rapid, and potentially floor-less. T-Mobile forced wireless ARPU from $65 to $45 in a decade. Long-distance rates fell 50% after the Bell breakup. No regulator set the floor — the market did, and markets don't guarantee survival.
Cloud faces the second, not the first. AWS margins have been stable above 30% — managed services deepening the moat faster than IaaS commoditizes. The treadmill hasn't arrived yet. The evidence says: not soon.
Subadditive cost monopoly. Electricity and broadband. Duplicating the physical infrastructure is genuinely wasteful. One provider serves the market at lower cost than two. The textbook natural monopoly.
Government-allocated scarcity. Cellular. The radio spectrum is finite by physics. Concentration comes from the interaction of physics, policy, and the compounding advantages of early allocation.
Scale economies and ecosystem lock-in. Cloud. No wasteful duplication — concentration comes from accumulated advantages: procurement leverage, data gravity, switching costs. The traditional natural monopoly framework fits poorly here.
All three produce the same policy question: what do you do about a dominant provider of essential infrastructure? But they produce it for different reasons, and the answers differ accordingly.
AI infrastructure is between Stages 1 and 2. The dominant entities own no physical infrastructure — no data centers, no fiber, no spectrum. Their market power is intellectual property and model quality: more fragile than a physical monopoly, harder to regulate. You can't rate-base a neural network.