Separate the layers, price the risk.
An AI campus is not one asset but a stack of three, each with a different risk profile and a different natural owner. At the bottom: land, interconnected power, and the built shell — slow to create, durable for decades, and generating contracted cash flow. In the middle: the IT layer — GPUs and network gear that depreciate in three to five years and belong on the balance sheet of whoever runs the compute. On top: the model and product layer, which is genuine venture risk.
When one balance sheet carries all three, everything is mispriced: the venture layer drags the infrastructure's cost of capital up, and the infrastructure's capital intensity drags the venture's returns down. The discipline that fixes this is old and unglamorous — it is the discipline of institutional real estate.
Why triple-net fits AI.
Under a triple-net lease, the tenant pays power, operating costs, and maintenance; the owner's rent arrives as a clean, contracted cash flow. Applied to AI campuses, the structure does three things at once:
- It matches tenor to asset life. A 15–20 year lease with rent escalators mirrors the useful life of the shell and its electrical backbone — not the refresh cycle of the chips inside it.
- It allocates power-price risk correctly. The party that chooses and runs the workloads carries the energy bill; the owner is insulated from the most volatile line item.
- It makes the margin structural. A target operating margin above 90 percent on a triple-net platform is not operational heroics — it is the arithmetic of the structure itself.
An AI campus on a 15-year triple-net lease to a credit tenant is closer to core infrastructure than to venture capital.
The debt market is arriving.
Capital markets have started to treat data-centre leases the way they treat other contracted infrastructure: as securitisable cash flows. Asset-backed issuance against data-centre portfolios has grown from a niche to a mainstream financing channel, and lenders underwrite the lease and the counterparty — not the resale value of GPUs. For operators with real leases and real power positions, that means a materially lower cost of capital; for everyone else, it means the gap between platforms and projects is widening.
The land-and-power layer compounds this: as interconnection queues lengthen, a secured power position appreciates in a way no building does. The bottom of the stack is where the moat lives.
How Castellan is structured.
Castellan is architected for exactly this logic from day one: a Swiss Aktiengesellschaft with institutional governance and IFRS-ready reporting, developing campuses leased on long-tenor triple-net terms to hyperscale, sovereign-cloud, and regulated tenants — a phased build-out toward a minimum of 800 MW by 2032, with the group's access to more than 5 GW of power beneath it.