The AI Infrastructure Hub:
The business case, the players and the economics
What it takes for a 100 MW AI data center to earn its cost of capital.
The market is a barbell: frontier training concentrates in a few gigawatt-class U.S. campuses, while inference drives regional hubs from 2027 to 2032 — in U.S. regions and in nearly every nation. Sovereign and regulated demand is among the fastest-growing sources of spend, and the gap between announcements and steel in the ground rewards those who can execute.
What the business case and feasibility look like
The anchor decides feasibility.
A creditworthy multi-year take-or-pay off-take for 40–70% of capacity, with ~20– 25% prepaid.
Social license is a first-order risk.
71% of Americans oppose local builds; ~$85B of projects have been cancelled.
Illustrative 100 MW U.S. facility · 25-year base case · new-operator financing (WACC ≈ 13%)
Only current-generation silicon clears the hurdle
Power, facility and staffing costs are largely fixed per MW, so revenue per MW decides returns.
Cost of capital is a structural divide
A 300–600 bps gap separates hyperscalers from new operators — a large swing in returns.
The broader conclusion
Like airlines, AI will bring unprecedented improvements to people’s lives and business productivity.
However, just as airline operators as a combined industry have generated low to returns since inception, we expect AI infrastructure operators’ returns cumulatively to be low over the long term unless they bring disciplined execution, rapidly get to scale, achieve low cost of capital and establish sustainable differentiation.
AI infrastructure providers & neoclouds: bare GPU
rental is fragile.
Move up the stack into sovereign niches and higher-value services, and run a disciplined capital- allocation playbook.
Sources: IDC, JLL, Accenture, Sightline Climate, Goldman Sachs, Evolvance, MarketsandMarkets (2026); JKay Advisors analysis
To request a detailed report, Click here!