Short answer. Data residency is a commitment that specified data is stored and processed in a named jurisdiction. It is narrower than sovereignty and stronger than a region dropdown. Prompts, outputs, logs, backups, support access and subprocessors all need to be in scope, and the commitment needs a contract, an architecture diagram and an audit trail.
| Question | One-sentence answer |
|---|---|
| What is data residency? | A contractual and technical commitment to store and process defined data in a named jurisdiction. |
| Is residency the same as sovereignty? | No. Residency is about location; sovereignty adds legal control, ownership and jurisdiction. |
| Does choosing a region guarantee residency? | No. Logs, backups, support access and subprocessors may leave the region. |
| Which regions are live for Plugsky? | eu-central-1 Frankfurt, me-central-1 UAE, ap-southeast-1 Singapore and us-east-1 Virginia. Riyadh is Enterprise-only. |
| Does GDPR require EU residency? | No. GDPR governs transfers and safeguards; residency is one way to reduce transfer risk, not the legal test. |
| What is a transfer impact assessment? | A documented review of the law and safeguards around a restricted transfer of personal data. |
| How long are prompts retained? | It varies by product and contract; ask for the retention schedule and the deletion mechanism in writing. |
| What evidence proves residency? | A data-flow diagram, subprocessor list, audit records, deletion certificates and contract terms, not a marketing badge. |
Data residency questions usually arrive with a simple brief: keep our data in the country. The simple version rarely survives procurement. Prompts are only one data type, and the region that serves them may not hold the logs, backups or support sessions around them.
This guide separates residency from sovereignty, maps the places AI data actually goes, and sets out the evidence a buyer should demand. It is a procurement guide, not legal advice; qualified counsel should review the final position.
Residency, sovereignty and localisation
The three terms describe different obligations. Mixing them is how a compliant-looking deployment ends up with an uncomfortable transfer memo.
What is data residency in an AI API?
Data residency is the commitment that specified data is stored and processed within a named jurisdiction. The commitment should name the data types in scope: prompts, outputs, request metadata, logs, backups and any derived data. It should also name the regions and the legal entity that receives the data. A provider that offers an EU endpoint has not automatically committed every log line to the EU. Residency is a scope statement, not a switch. Ask for the data-flow diagram and the contract schedule that make the scope enforceable. See the residency explainer for the underlying concepts.
How does data sovereignty differ from residency?
Sovereignty adds legal control to location: who owns the operator, which courts can compel access, where keys and metadata live, and whether the service is subject to foreign law. A data centre in Frankfurt does not make a service sovereign if the operator, control plane or key custody sit elsewhere. Sovereign AI is therefore a procurement and legal question, not only an infrastructure one. Ask who owns the entity, who can authorise access, and what happens under a government request. The sovereign AI page describes the deployment options without claiming local infrastructure that is not live.
Where AI data actually goes
A region selection covers the inference path. It may not cover the observability, support and backup paths that surround it.
| Data type | Typical location | Residency question |
|---|---|---|
| Prompts and completions | Inference region | Is this data in scope, and which region serves it? |
| Request metadata | Control plane | Can metadata be region-pinned or minimised? |
| Logs and traces | Observability stack | Where are they stored, and for how long? |
| Backups | Provider-selected region | Are backups in-region, excluded or encrypted with your key? |
| Support access | Global or regional staff | Who can see data, from where, and under what approval? |
| Subprocessors | Varies by service | Are they listed, notified and subject to objection? |
Which Plugsky regions are live?
Four self-serve regions are live: Frankfurt (eu-central-1), the UAE (me-central-1), Singapore (ap-southeast-1) and Virginia (us-east-1). Riyadh is available on Enterprise and is not a self-serve default region. A live region is a location, not a residency guarantee: the contract still has to say which data types are in scope and which paths are excluded. The table below is the honest starting point for a region discussion, not a substitute for the DPA.
| Region | Location | Access | Residency note |
|---|---|---|---|
| eu-central-1 | Frankfurt, Germany | Self-serve | EU region; contract terms govern which data is in scope |
| me-central-1 | United Arab Emirates | Self-serve | GCC region; check PDPL and sector rules |
| ap-southeast-1 | Singapore | Self-serve | APAC region; check local transfer rules |
| us-east-1 | Virginia, United States | Self-serve | US region; check state and sector rules |
| Riyadh | Saudi Arabia | Enterprise | Not a self-serve default; scope in the contract |
What if my country has no local region?
The honest options are the nearest live region, a private VPC, on-premises deployment or an air-gapped environment. Those are different answers with different legal and operational consequences. A nearby region may reduce latency but does not create a local data centre or satisfy an in-country requirement. A private VPC can pin the inference path but may still leave logs or support access outside the country. On-premises and air-gapped remove more paths but shift the operating burden to you. Say which option you are relying on, and put it in the contract. See private VPC and air-gapped deployment for the scoping questions.
The legal layer: GDPR, PDPL and transfers
Residency is one control inside a wider legal analysis. It can reduce transfer risk, but it does not replace a lawful basis, a processor contract or a transfer mechanism.
What does GDPR require for an AI API?
At a minimum: a lawful basis for processing, an Article 28 processor contract, and a valid transfer mechanism for any personal data leaving the EEA. The contract should cover purpose limitation, subprocessors, security measures, breach notification, deletion and assistance with data subject rights. If data leaves the EEA, standard contractual clauses may be part of the answer, supported by a transfer impact assessment. For the UK, the equivalent analysis runs under the UK GDPR and the ICO's international transfer guidance. Residency in the EU can reduce the number of transfers but does not remove the need for the other terms. The DPA guide covers the contract layer, and counsel should review the final position.
How do Saudi PDPL and DIFC rules change the analysis?
They add separate transfer rules, registration or notification duties, and local-law obligations, so an EU-only analysis is not enough for GCC deployments. PDPL is administered by SDAIA; DIFC and ADGM have their own data protection regimes; sector regulators may add more. The correct approach is to map each data set to the regime that governs it, then verify the contract terms and any transfer mechanism with local counsel. Plugsky documents PDPL, DIFC and NSD alignment in its DPA, but the current version and the specific deployment still need review.
Deletion, logs and audit evidence
Residency claims fail at the edges: a backup in another region, a support session from another country, or a log line that outlives the contract.
What should a deletion commitment include?
A named data set, a trigger, a deadline, a technical mechanism and a certificate. The data set should include prompts, outputs, logs, backups and derived data, and the trigger should cover both user deletion and contract termination. Ask how deletion propagates to subprocessors and backups, and whether the provider returns data before deleting it. “Deleted from the primary store” is not the same as “deleted everywhere.” Put the deadline and the evidence in the DPA or the security schedule, and test the export path before you need it.
What should audit logs show?
Who accessed what, when, from where and under which authorisation. Separate administrative changes from inference requests, because they carry different privacy and retention requirements. Ask whether logs are immutable, how long they are retained, whether prompt content is captured, and how records reach your SIEM. Time synchronisation matters for reconstructing an incident across regions. A residency review should include a sample audit record, not only a description of the logging system. If the provider cannot export the record, you cannot verify the claim.
How do you verify residency during procurement?
Ask for a data-flow diagram, the current subprocessor list, the retention schedule and a sample audit record. Then test: place canary data, monitor egress, confirm the region that serves the request, and check where the logs appear. Request the current attestation under NDA and verify its period; do not accept a certification claim as a statement of fact. The contract should name the regions, the excluded paths and the notice period for changes. Verification is a stack of evidence, not a single certificate.
What to verify in a data residency contract
The schedule should be specific enough that an auditor can test it without interviewing the sales team.
- Defined data. Prompts, outputs, metadata, logs, backups, derived data and support recordings.
- Named regions. The exact region IDs and the legal entity that receives each data type.
- Excluded paths. Anything that may leave the region, stated explicitly rather than left to inference.
- Subprocessors. A current list, change notice and a right to object.
- Transfer mechanism. SCCs or another valid tool, plus the transfer impact assessment where required.
- Support access. Who can access data, from which country, under what approval and with what logging.
- Deletion. Deadlines for primary data, backups, logs and subprocessor copies, with a certificate.
- Audit rights. The right to request records, test evidence and a current attestation under NDA.
- Change notice. Advance notice of a new region, subprocessor or data path, with a termination right.
- Exit. Export format, transition assistance and deletion after termination.
Residency is proven by the paths you can name and the records you can export, not by the flag on the signup page.
What we did not verify. We did not audit Plugsky's infrastructure, review a signed DPA or test a live deployment for this page. Region facts are the published deployment map checked on 20 November 2026. Legal statements are procurement guidance, not legal advice; counsel should review the contract and the transfer analysis for your jurisdiction.
Frequently asked questions
Does data residency apply to logs and backups?
Only if the contract says so. Many residency commitments cover the primary inference path while logs, backups and support access follow a different route. Ask for each data type to be named, with the region and retention period. If a log store or backup sits outside the region, the commitment is partial. Put the excluded paths in the schedule so an auditor can see the boundary rather than infer it.
Can I choose a region per request?
Region selection is normally account-level or deployment-level, not a per-request header. Ask how the region is assigned, whether it can be changed, and whether a failover can route a request to another region. If failover crosses a border, the residency commitment needs to say so. A per-request choice sounds flexible but is harder to audit; a pinned deployment is easier to evidence. Document the pinning method in the contract schedule itself.
Is EU residency enough for GDPR?
No. GDPR requires a lawful basis, an Article 28 processor contract, security measures and a valid transfer mechanism for any personal data leaving the EEA. EU residency can reduce transfer risk, but it does not replace those obligations. It also does not cover data that support staff access from outside the EU. Treat residency as one control in the analysis, not the analysis itself, and have counsel review the final position.
What is a subprocessor?
A subprocessor is a third party that processes personal data on behalf of the processor. In an AI service that can include cloud hosting, observability, support tooling, content moderation and email delivery. The DPA should list them, require notice of changes and give you a right to object. Ask which subprocessors touch prompts, logs and backups, because a long list is less useful than a mapped data path. Ask for the mapped list.
How fast must a provider delete data?
There is no universal deadline; the contract and the applicable law set it. Ask for a specific number of days for primary data, backups and logs, plus the mechanism and the certificate. A short deadline for the primary store can coexist with a long backup cycle, so get both. If deletion must propagate to subprocessors, name them and the evidence you will receive. Test the deletion report before termination, not after the contract ends.
Do AI providers train on my data?
It depends on the product and the contract. Ask whether prompts and outputs are used for model training, whether there is an opt-out, and how long data is retained for abuse monitoring or quality review. Do not assume that a paid plan or a private deployment changes the answer. Put the permitted purposes in the DPA and verify the current retention schedule, because these terms change. Verify the current terms, not a blog post.
What is a transfer impact assessment?
A transfer impact assessment is a documented review of the law and safeguards around a restricted transfer of personal data. It considers the destination's legal framework, the nature of the data, the transfer mechanism and any supplementary measures. It is not a one-off template; it should be updated when the destination, subprocessor or legal landscape changes. Counsel usually owns the assessment, with input from security and procurement. Keep it with the contract.
Can support staff access my data from another country?
They may be able to, even when inference runs in-region. Ask for the support access model: which roles, which countries, what approval, what logging and how access is revoked. A regional support team is a stronger answer than a promise that nobody looks. If support access crosses a border, include it in the transfer analysis and the residency schedule rather than treating it as an operational detail. Name the countries in the schedule.
Does encryption solve residency?
No. Encryption reduces the impact of unauthorised access, but it does not change where ciphertext, keys or metadata are stored. If the provider holds the key, it can decrypt; if the key is in another jurisdiction, key custody becomes part of the analysis. Ask where keys live, who can use them and whether encryption is applied to logs and backups as well as the primary store. Key custody is part of the boundary.
How do I test a residency claim?
Use canary data, monitor egress, confirm the serving region and inspect where logs and backups appear. Ask for a sample audit record and a data-flow diagram before the test, then compare the observed path with the documented one. A test that only checks the public endpoint proves very little. Record the date and repeat the test after any region, model or subprocessor change. Evidence beats a sales assurance every time.
What happens if a provider changes regions?
The contract should require advance notice, a migration plan and a termination right if the change breaks your residency requirement. Ask whether the change applies to existing data or only new data, and how long the old copy remains. A provider that can move data without notice has not made a residency commitment; it has described a current configuration. Notice, migration and exit are the three clauses that matter most.
Sources
Primary sources and official documentation for this 2027 planning guide — evidence checked 20 November 2026.
- Regulation (EU) 2016/679 (GDPR), EUR-Lex — checked 20 November 2026.
- EDPB Guidelines 05/2021 on the interplay between Article 3 and Chapter V — checked 20 November 2026.
- European Commission, Standard Contractual Clauses — checked 20 November 2026.
- SDAIA, Saudi Personal Data Protection Law — checked 20 November 2026.
- DIFC, Data Protection Law No. 5 of 2020 — checked 20 November 2026.
- UK ICO, International Transfers — checked 20 November 2026.
Last checked 20 November 2026.
Related reading
Scope the data path, not just the region
Bring your residency requirement and we will map prompts, logs, backups and support access to the deployment options.
Request security pack Read the DPA guide