Buying an AI infrastructure and GPU services business in Canada

A GPU hosting or compute infrastructure business can look like a straightforward acquisition — buy the hardware, keep the customers, keep the revenue coming in — but the things that make one of these businesses a good buy or a bad one sit underneath the numbers a seller presents first. Evaluating the opportunity properly means asking about hardware, contracts and agreements before spending much time on price, because those three things determine what the revenue is actually worth.

What a good opportunity looks like

A strong candidate runs on hardware that’s a recent enough generation to remain competitive for a reasonable stretch after closing, generates revenue from customers under multi-year or reserved-capacity contracts rather than spot resale, and holds data-centre, power and colocation agreements that are either already assignable or reasonably likely to be consented to. A spread of customers, rather than dependence on one or two large training workloads, is also a strong sign — it means the business isn’t one customer’s decision away from a very different-looking year.

What a weak opportunity looks like

The businesses that look attractive on paper and disappoint after closing tend to share a pattern: hardware close to the end of its competitive life relative to current accelerators, customer contracts with no minimum commitment that can be cancelled with little notice, thin resale margin over the underlying cloud or lease cost with no proprietary layer of value added on top, and debt or lease obligations on the hardware sized for equipment with a much longer useful life than it actually has left.

What a seller may not volunteer

Ask directly about utilization trends over the past several periods, not just the current snapshot, since a business can be running at a healthy utilization rate today on contracts that expire soon with no renewal commitment behind them. Ask about the true generation and remaining competitive life of the hardware fleet, since a seller has an incentive to describe equipment in terms of its working condition rather than its competitive position against newer accelerators. And ask specifically whether every power, colocation and data-centre agreement can be assigned on a change of ownership, or whether any require a third party’s consent that hasn’t yet been sought. It’s also worth asking whether the business, or any of its customers, has ever received a regulatory inquiry related to data handling, security screening or hardware sourcing — there’s no dedicated licence for this activity in Canada today, so a seller focused on operations may not think to mention a compliance conversation that never resulted in a formal finding, even though a buyer would want to know about it.

Look past the hardware — check what software actually differentiates the business

Some providers in this category have built real orchestration or scheduling software in-house — the layer that decides how compute gets allocated, monitored and billed — and that layer, where it genuinely exists, is often the difference between a business worth a premium and one that’s simply reselling access to someone else’s infrastructure at a thin margin. A buyer evaluating the opportunity should ask directly whether any such software exists, and if it does, whether the company actually owns the intellectual property in it outright. It’s common in a fast-growing technical business for part of that software to have been built by a contractor or an early technical hire who never signed an agreement assigning the resulting rights to the company, leaving a gap in the ownership chain a buyer inherits on closing unless it’s fixed beforehand. A provider with no proprietary software layer at all isn’t necessarily a bad purchase, but it should be evaluated — and priced — as a hardware resale business, not as a technology company that happens to sell compute.

What the buyer needs to qualify for

There’s no professional licence standing between a buyer and this kind of purchase in Canada today, but there are real gatekeepers all the same. A lender financing hardware already in place will want to approve, or at least be informed of, a change in the counterparty on that financing. A data-centre or colocation provider will typically want to vet a new operator before consenting to an assignment, even informally, and a customer with a large committed contract may have its own approval process before accepting a change of provider. A buyer who assumes these approvals are a formality is usually the buyer who discovers, partway through closing, that they aren’t.

Structuring around what you find

Where hardware is ageing, contracts are thin or an agreement’s assignability is genuinely uncertain, a buyer isn’t necessarily walking away — these are commonly addressed through price adjustment, a holdback tied to confirming consents after signing, or an earnout structured around retained utilization. What matters is identifying these issues before agreeing on price, since they’re far easier to negotiate into the deal than to raise after a purchase agreement is signed.