Geo-DNS based GPT API - Observations on current capabilities

Hello forum,

I’m looking for information about the regional data center that each GPT is running within. What I’ve tested so far is a geo-DNS deployed API using Azure’s Traffic Manager geo-DNS service.

Based on where the API caller is located, this service will either provide an IP into Azure’s North Europe data center; or an IP into Azure’s US East data center. When I test via a browser with VPN, everything is working great. Connect to UK VPN; hit the GET endpoint and it shows me North Europe connection. Similar for US East.

However when ChatGPT GPT is making the API calls this is always coming to US East. So either OpenAI’s GPT infrastructure itself is pinned to USA, or my GPT builder profile is pinned to USA, or some other combination.

Has anyone else had experience with geo-IP based GPT API? Could anyone from OpenAI comment on this?

With the way this is, it means ChatGPT “out of the box” we can’t have the data residency for Europe unless we publish a separate Europe only GPT having a specific server URI. And hence a separate GPT instance for every Region.

If anyone resident outside NA wants to check it out, I’ve put up a temporary GPT that will report if the OpenAI API traffic is originating US or Europe:

Temporary Region Checker GPT

… and by the way, this GPT was created while I am connected to UK VPN. So far I’ve tested that:

  • saving the GPT while on a remote VPN location doesn’t move it outside USA
  • creating a GPT while on a remote VPN location doesn’t create it outside USA

Potential next step is to create a net-new user on my GPT Teams subscription and have this user be completely UK based; to see if that user’s GPTs might be UK based. Not sure I want to incur an extra license for a month just to test this quite yet :slight_smile:

Dan

1 Like