Coverage
Capacity is not scarce everywhere. It is scarce where you were going to look.
Every market has a different binding constraint — a queue, a moratorium, a water table, a regulator, a transformer. Knowing which one applies is most of the work of finding capacity that can actually be delivered.
Four regions, four different problems
United States
The deepest market and the longest queues. Northern Virginia, Texas, Phoenix, Atlanta, Ohio — and the secondary markets where power is actually available now.
Open the market →North AmericaCanada
Cheap firm hydro in Québec, the fastest interconnection on the continent in Alberta, and a sovereign compute agenda funding both.
Open the market →Europe, Middle East, AfricaEMEA
FLAP-D is largely closed. The Nordics, Iberia and the Gulf are where megawatts are being delivered — each with its own regulatory price.
Open the market →Asia PacificAPAC
Singapore rations, Johor absorbs, Japan pays a premium for certainty, India is the cheapest and the most conditional.
Open the market →How a market gets ruled in or out
Buyers usually arrive with a market already chosen, and the choice is usually inherited — from where the company is headquartered, where the last deployment went, or where a vendor happens to have space. That is how you end up paying a scarcity premium in a market that is scarce for reasons unrelated to your workload.
- Power position, first and always. Not whether a building exists, but whether it has energised capacity or a firm, documented path to it. Everything else in this list is negotiable; this one is physics and paperwork.
- Latency, honestly assessed. Training runs tolerate tens of milliseconds. Inference serving real users does not. Conflating the two is the single most expensive mistake in site selection, and it eliminates the markets where capacity is cheap and available.
- Regulatory and data exposure. Residency obligations, sectoral rules, export control on the silicon itself. A market can be perfect on power and cost and still be unusable for your data.
- Delivered cost, not headline rate. Power price, currency, tax treatment, cross-border transit and the cost of the people who will have to physically touch the equipment.
- Exit and expansion. Whether the market will still have capacity when you need to double, and what it costs to leave if it does not.
The capacity sourcing mandate runs this comparison properly, and the note on where capacity is actually available covers the current state of it.
Asked about coverage.
01Which markets do you actually cover?
Mandates run across North America, EMEA, APAC and Latin America. In practice the shortlist for any given requirement is three or four markets, not fifteen — the value is in ruling markets out quickly rather than presenting all of them.
02How do you choose a market for a requirement?
Power position first, then latency to the users or data, then regulatory exposure, then cost. For training workloads latency is usually the loosest of the four constraints, which opens up markets most buyers never consider.
03Can a single deployment span regions?
Frequently. Training in a low-cost, power-abundant market and inference close to users is the most common split, and it changes which contracts need to be tied together and which can be independent.
Tell us the workload, not the market.
Give us the megawatts, the density, the latency requirement and the data constraint. We will tell you which three markets are real and which one is a scarcity premium you do not need to pay.