Underwriting AI data centres through the cycle

At the peak of every infrastructure cycle the same question is asked: what happens if it turns? With AI capital expenditure at record levels, it is a question every serious counterparty raises, and one that should be answered in the underwriting rather than after it.

Having built data centre platforms from inception to institutional scale three times, I have found that the facilities which kept their value through difficult markets shared characteristics. None depended on a single technology thesis being right.

Demand is several markets

AI training is the most visible demand, but it sits on a large and steadily growing base of cloud, streaming, financial services and enterprise workloads, and a faster-growing layer of inference that runs continuously once models are deployed. ECB President Christine Lagarde observed in September 2026 that Europe already has too little data centre capacity for its own needs before AI growth is counted. A slowdown in training spend would not remove that shortage.

Scarcity is in power, not in buildings

In Europe, the constraint on new capacity is grid connection, with queues in the main hubs averaging seven to ten years according to the IEA. A facility with confirmed power, energised before 2030, remains scarce regardless of any individual tenant's plans. If a tenant's requirements change, a powered, zoned and built facility can be re-let to another cloud provider, AI lab or enterprise. A project whose power is still prospective cannot.

Own the durable layers

Accelerators depreciate within a few years; land, power connections, buildings and cooling infrastructure serve decades and several hardware generations. The platform's model is deliberately on the durable side of that line: customers own the compute, Qore owns the infrastructure.

Flexibility and headroom

A facility designed for one tenant's proprietary layout and one hardware generation is fragile. One with headroom in power and cooling, designed for liquid cooling, divisible into phases and re-lettable across customer types, is resilient. Modular delivery reinforces this by adding capacity as demand is confirmed rather than in a single speculative block.

Commitments that follow certainty

Finally, discipline in commitments: pre-letting before building, and acquisition payments staged to confirmed power milestones. This does not eliminate risk, but it ensures the largest commitments follow certainty rather than precede it.

A correction would test business models built on speculative capacity, single tenants, unconfirmed power and short-lived assets. It would be considerably kinder to infrastructure that is power-secured, well located, flexible and owned for the long term. That is the infrastructure Europe is short of, and the infrastructure this platform is built to own.

Christof Hämmerli

Co-Founder and Chief Investment Officer, Qore Infrastructure Partners. Profile