Analysis
30 of 42 providers tracked by StackTicker are headquartered in the US
Sovereign AI is discussed everywhere. We resolved the home country of every provider on our board from structured public records: 30 of 42 are American, two are Chinese, and the whole of Europe accounts for eight.
StackTicker structured report methodology
See where this value came from and when it was checked.
- Source type
- official
- Source retrieved
- Sep 5, 2026 · 16:20 UTC
- Effective date
- Not stated
- Verification
- Source verified
We resolved the headquarters country of all 42 providers we track - GPU hosts and inference APIs - from structured public records rather than from domain suffixes. 30 are American. Two are Chinese. Europe, taken together, is eight.
Every government in the world is currently discussing sovereign AI capacity. This is what the buyable market looks like while they do.
30 of the 42 compute providers on our board are American. Europe accounts for eight, and one provider is Chinese.
Provider headquarters, by country
Home country of the operating company, resolved from structured public records. Not the location of the datacentre - a US company renting European capacity is counted as American. Recounted 42 providers on 2026-09-03 after eight were added.
| Country | Kind | Providers |
|---|---|---|
| United States | API providers | 15 |
| United States | GPU hosts | 15 |
| France | GPU hosts | 3 |
| United Kingdom | GPU hosts | 3 |
| People's Republic of China | API providers | 2 |
| Finland | GPU hosts | 1 |
| India | GPU hosts | 1 |
| Netherlands | GPU hosts | 1 |
| Not established | API providers | 1 |
What this does and does not say
It is a count of companies, not of capacity or of where machines physically sit. A single American provider with capacity in Frankfurt serves European data from Europe while remaining an American company under American law. For latency, headquarters tells you little. For jurisdiction it is one part of the picture: the data-centre location, the contracting entity and the provider's terms can matter as well.
It is also a count of what a normal buyer can actually rent today with a card and an account. Sovereign programmes exist and national champions are being funded; almost none of it is on a public price page, which is the only thing we can measure.
Our view
The interesting number is not the thirty. It is the two. Two Chinese providers appear in a market where Chinese open-weights models are near the top of several public benchmarks - the models travel, and the serving does not follow them. Weights cross borders as files; inference is a business with a jurisdiction, and the two have separated.
For a European or Asian team, the practical consequence is that choosing a model and choosing a jurisdiction are now separate decisions. You can run a Chinese open model on French hardware, or an American model on Finnish hardware, and the constraint that used to bind them - only the maker could serve it - is gone.
What follows from it
If jurisdiction is a requirement rather than a preference, the shortlist is short and worth knowing precisely. Renting the accelerator directly is the version of this where you choose the country outright, because you are choosing the machine.
The same model can cost several times more depending on who serves it. See every provider we track for a model, with the price each one publishes and the page it came from.
Open the comparison →Run it yourselfCheapest GPUs by the hourRenting the hardware keeps the weights, the prompts and the logs on infrastructure you choose. See what an hour of each accelerator costs across every host we track.
Open the GPU market →