For roughly two years the binding constraint on artificial intelligence deployment was the supply of accelerators. Capital chased allocation, allocation chased fabrication capacity and valuations across the semiconductor supply chain repriced accordingly. That constraint has not disappeared, though it has been substantially relieved by capacity coming online and by a broadening field of credible parts.
The constraint that has replaced it is electrical, and it will not relieve on the same timetable. Silicon fabrication responds to capital within eighteen to thirty months. Transmission infrastructure, generation capacity and the queue for grid interconnection respond over five to twelve years, and in several major markets they are not responding at all.
What This Does to the Asset
A data centre is customarily underwritten as a real estate asset with a technology tenant. The land is valued, the shell is costed, the fit-out is amortised and the covenant of the operator is assessed. On that basis a number of facilities have been financed on assumptions that no longer describe the business.
The scarce input is not the building. In the markets Sapna Ventures has examined, a contracted power supply with a firm energisation date is now worth a substantial multiple of the site it will serve. Two adjacent parcels of land, identical in every physical respect, can differ in value by an order of magnitude on the basis of where each one sits in the utility's queue.
This has an unhappy implication for anyone holding a development pipeline. A pipeline consisting of optioned land with indicative connection offers is a pipeline of intentions. A pipeline consisting of energisation agreements is an asset. The two are frequently presented in the same slide, under the same heading, expressed in the same megawatts.
Where the Gulf Sits
The Emirates and Saudi Arabia hold a structural advantage here that has little to do with policy enthusiasm and a great deal to do with physics and sovereign balance sheets. Generation capacity can be commissioned quickly because the fuel is domestic and the permitting authority and the utility answer to the same government. Solar irradiance is among the highest available anywhere, and the land beneath it has no competing use.
The offsetting difficulty is thermal. Ambient temperatures across the Gulf summer push cooling loads well above the assumptions embedded in most published efficiency figures, and the obvious remedy – evaporative cooling – consumes water in a region where water is produced by desalination and therefore by more electricity. The facilities Sapna Ventures regards as underwritable in this climate are those designed around closed-loop liquid cooling from the outset, instead of those retrofitted towards it.
How Sapna Ventures Underwrites It
Four tests are applied before capital is committed to a digital infrastructure asset.
The power agreement is examined before the building. An energisation date that is contractual, with damages attached, counts. An indicative offer from a utility does not, irrespective of how encouraging the correspondence around it has been.
Tenant concentration is treated as the principal credit risk. A single-tenant facility leased to a well-capitalised hyperscaler is a good asset and a bad diversification, and the covenant should be priced as what it is, which is corporate credit exposure dressed as property.
Escalation is stress-tested against power price as an alternative to inflation generally. In several of these markets the electricity input is the majority of operating cost and moves independently of any index the lease is tied to.
Residual value is written down aggressively. The building will outlive several generations of the equipment inside it, and the specification demanded by a tenant in 2035 is not knowable from here. Underwriting that leans on terminal value is underwriting a guess about hardware nobody has designed.
The Position Taken
Sapna Ventures holds digital infrastructure exposure at the operating end of the spectrum, in assets already energised and already let. The firm has declined several development opportunities over the past year on power grounds alone, some of them at prices that would have looked attractive had the megawatts been contracted. They were not, and megawatts that exist only in a letter of intent have a way of remaining there.