We get asked to open somewhere new most weeks. The criteria have not changed since 2016, and they are applied in this order.
1. Real interconnection, not a cross-connect to one carrier
A site is only useful if there is an exchange worth joining and at least three transit options that will sell to a network our size. Single-carrier facilities look cheap and become expensive the first time that carrier has a bad week.
2. Diverse fibre with a documented route
Two paths on a drawing is not diversity. We ask for the actual route, and we have walked away from facilities that could not produce one. Bucharest in 2024 taught us that lesson expensively.
3. A jurisdiction we can explain to a customer in one sentence
If we cannot state plainly which law applies to a disk and who can compel disclosure, we should not be selling servers there.
4. Power we can price for five years
Hardware lives five years. If the power cost is unforecastable over that horizon, the site's economics are a bet rather than a plan.
Latency to a market is deliberately not on the list. It is the reason we start looking, not the reason we sign.