Where could my AI request be processed?
A probabilistic estimate of the cloud region — and data centers in our database — that could plausibly serve a request, from the provider's documented routing scope, estimated network latency and known capacity. Providers don't disclose where an individual request runs; this never claims to know.
Method
- Candidate regions come only from the provider's documented scope for the chosen mode (a fixed region, a US-only setting, or global).
- Estimated round trip = 4 ms + 2 × distance × 1.5 (fiber detour) ÷ 200 km/ms.
- Relative likelihood ∝ exp(−RTT / τ) × (relevant capacity + 50 MW)0.5, with τ = 20 ms when requests go to a chosen or nearest region and 60 ms when the provider routes by available capacity. Capacity counts facilities in our database near the region run by the cloud (half weight) or listing the AI company as a user (full weight); for US/global modes, the company's dedicated sites outside public cloud regions are candidates too.
- Confidence rates the region estimate only: High for a single documented region, Medium for a documented set, Low for undocumented or global routing.
- The optional AI explanation is generated from this estimate alone and rejected if it mentions any region or facility not in it.