Executive summary
AI infrastructure businesses resemble traditional infrastructure: high upfront capital, long-lived assets and returns that depend on contracts, utilisation and the cost of capital. The most attractive ones combine scarce inputs — power and permitted land — with long-term, creditworthy customers.
Key findings
- 1Access to reliable power and suitable land can be a bigger constraint on data-center growth than demand.
- 2Long-term contracts with creditworthy customers make returns more predictable than short-term, speculative leasing.
- 3Because assets are long-lived and expensive, the cost of financing has a large effect on returns.
Infrastructure, not just technology
The AI boom is often discussed as a software story. Underneath it sits something much older: buildings, power lines, cooling systems, fibre and land. These are infrastructure assets, and they follow infrastructure economics.
Five questions that matter
- Is the key input scarce? Grid connections, permitted land and cooling water can be harder to secure than customers.
- Who is the customer, and for how long? Long-term contracts with creditworthy tenants make cash flows more predictable.
- How is pricing set? Fixed contracts, pass-through power costs and escalators protect margins differently.
- What does the capital cost? Long-lived, expensive assets are highly sensitive to interest rates and financing terms.
- How fast does the technology change? Buildings last decades, but the equipment inside them may need replacing far sooner.
Durable versus temporary
Booms in infrastructure spending have happened before — in railways, telecom fibre and others. Some of the businesses built in those booms created lasting value; others were built on optimistic forecasts. The difference usually came down to the questions above.
The investor's lens
An attractive AI infrastructure business looks less like a startup and more like a utility or toll road: scarce assets, contracted demand and disciplined financing.
Risks and limitations
This is an educational framework, not a recommendation. AI infrastructure demand forecasts are uncertain and spending cycles can overshoot.
Facts, assumptions, estimates and opinions
TIC labels the nature of each material claim so you can weigh it appropriately.
- Verified factData centers require continuous electrical power and cooling to operate.
- OpinionScarce inputs such as grid connections are likely to matter more to long-term returns than the headline size of a data-center campus.
Sources and references
This piece relies on arithmetic and general principles; no external data sources are cited.
Downloads
A downloadable PDF edition of this research has not been published yet.