A provider can promise to keep nothing and still process your prompt on another continent. Data residency is the question of where.
How it works
Residency has two parts that providers often separate:
| Part | What it means | Example |
|---|---|---|
| Storage at rest | Stored content (files, batches, saved conversations) sits in the region | OpenAI: data stored at rest in the selected region “when the endpoint requires persistence” |
| Processing | Inference itself runs on GPUs in the region | OpenAI: regional processing in the US, EU, Australia, Japan, India and UAE; Canada, Singapore, South Korea and the UK offer storage only |
OpenAI’s guide shows the usual caveats. A project’s region is fixed at creation. Residency “does not apply to system data”, and OpenAI “may also process and temporarily store Customer Content outside of the Region to deliver the services” in some cases. Read the exceptions, not just the region list.
The law behind residency is usually a transfer rule, not a storage rule. GDPR Article 44 says a transfer to a third country may take place only if the conditions of Chapter V are met, including for onward transfers. The European Commission’s adequacy decisions let data flow to listed countries “without any further safeguard being necessary”; for the US that covers only companies in the EU-US Data Privacy Framework. In Saudi Arabia, SDAIA’s transfer regulation (issue 02, August 2024) sets out adequacy, safeguards and risk assessments for transfers outside the Kingdom. The Saudi and UAE PDPL checklist walks through it.
Why it matters
- Every prompt with personal data is a potential transfer. If the model runs abroad, so does the personal data in the prompt.
- Partners and failover change the country. A provider with backup capacity elsewhere may process some requests there. Ask about each subprocessor.
- Minimisation still comes first. A prompt with no personal data in it raises fewer transfer questions. Redact before you send.
On BroadNet Inference
BroadNet’s Terms (§2) and privacy statement say its own models and the open-weight models it hosts run on BroadNet-owned or leased hardware in the United States, with partner capacity as backup. Some models are partner-hosted: the Terms (§7.3) name DeepSeek V4.1 Flash as partner-hosted, and partner-served models run under zero-data-retention terms with US inference partners. The Cohere Transcribe Arabic model page lists “NVIDIA GPUs operated by BroadNet in the United States”.
There is no self-serve region selector. §7.3a says: “Contact us to agree processing locations for your deployment before sending data with residency requirements.” BroadNet’s Inference overview page says “Regional processing in MENA or Europe is available on request for customers with data-residency requirements”; like private endpoints, it is an enquiry agreed in writing, not a self-serve option. See private endpoint planning and the contact page. Inference content is not retained, which helps, but it is not in-country processing. This page is not legal advice.