What is an AI infrastructure and GPU services business worth?

Valuing a GPU hosting or compute infrastructure business starts by setting aside the current month’s utilization numbers, which can look impressive and still say very little about what a buyer is actually acquiring. This is a hardware-and-contracts business dressed in an AI label, and the two things that decide most of the value — what’s owned outright and what customers are actually committed to — rarely show up clearly in a top-line revenue figure. A buyer who stops at revenue and a rough multiple is pricing the business on the wrong variable.

What a buyer is actually paying for

The businesses that command a stronger position in this category combine a few specific things: compute contracted under multi-year or reserved-capacity terms rather than sold spot-market by the hour, hardware that’s owned outright or leased on terms favourable enough to function like ownership, and data-centre, power and cooling relationships that took years to build and can’t be replicated by a new entrant signing a lease. A customer base spread across several buyers of compute, rather than concentrated in one or two large model-training workloads, adds further support, because a single customer’s decision to move workloads elsewhere doesn’t threaten the whole business. None of this shows up as a separate line on a set of financial statements, which is exactly why a buyer has to go looking for it.

Hardware age is a valuation input, not a footnote

GPUs and other accelerators used for AI workloads are a fast-depreciating asset class, and a fleet’s age relative to the current generation of hardware matters as much to a buyer as the fleet’s book value. Equipment nearing the end of its competitive useful life can still generate revenue today while being worth far less to a buyer who has to plan for replacing it soon after closing. The Canada Revenue Agency’s capital cost allowance rules set out the tax mechanics for depreciating this kind of equipment, but the accounting depreciation schedule and the equipment’s actual competitive life rarely move at the same pace — a buyer values the hardware on the second, not the first.

Recasting earnings without hiding the real costs

The usual small-business add-back exercise — stripping out owner compensation, one-time costs and non-recurring items — still applies here, but a compute business has costs that look discretionary and aren’t. Power, cooling and the lease or debt service on hardware are ongoing costs of operating the business, not one-time expenses, and a seller who tries to add them back to inflate normalized earnings is presenting a number a buyer’s advisor will simply reverse. A more useful exercise separates recurring, contracted revenue from spot or short-term resale revenue, since the two carry very different risk and rarely deserve the same multiple.

Why two similar-looking providers price differently

Two businesses with near-identical current revenue can be worth meaningfully different amounts once contract structure and hardware position are accounted for. One provider running reserved-capacity contracts on owned, recent-generation hardware with a spread of customers is a fundamentally more durable business than one reselling leased capacity on a thin margin to a couple of large training customers with no minimum commitment. The second business’s utilization, and therefore its revenue, can fall sharply and quickly if even one customer moves workloads elsewhere, and a buyer prices that fragility directly into the number offered.

What different buyer types actually want

The buyer pool for this category is unusually varied for a small or mid-sized business — data-centre and colocation operators expanding into AI compute, cloud and hosting companies adding GPU capacity, private equity infrastructure funds, and AI labs looking to secure captive capacity for their own use. A strategic buyer already running data-centre infrastructure may value the power and colocation relationships more than a financial buyer would, while an AI lab buying for captive use may care less about customer diversification, since it plans to become the primary customer itself. Knowing which kind of buyer is most likely to be interested changes what a seller should emphasize, though it does not change what the business is actually worth to the market as a whole.

Data residency and privacy posture as a pricing input

A compute provider’s data-residency commitments, and its posture as a data processor under federal privacy law, are a genuine value driver in this category, particularly for the slice of the customer base drawn from regulated industries or government bodies. Those customers often require, contractually, that data stay within Canada and that the provider can demonstrate its processor obligations are properly documented, and a buyer pays more confidently for a business that can substantiate those commitments with real technical and contractual evidence than for one that simply asserts them in a sales deck. Where any cross-border customer relationships exist, a buyer should also expect that certain classes of advanced computing hardware and some international dealings can attract additional regulatory attention, worth understanding early rather than treating as background noise. None of this changes the underlying hardware-and-contracts economics, but it changes how confidently a buyer can rely on the customer base staying put — and that confidence shows up directly in the multiple offered.