The short answer: you can run enterprise AI on Saudi personal data and SAP data today, but the model’s location matters as much as the data’s. In practice, most of the work is in three choices. First, classify the data an AI use case touches. Second, pick a model that processes that data inside the Kingdom, or a transfer that the Personal Data Protection Law (PDPL) permits for that data flow. Third, check that your SAP AI features actually run where your SAP system runs. This guide therefore maps what we checked on official pages in November 2026, and it flags what we could not confirm. It is a practical explainer for CIOs, not legal advice. For that reason, your legal and compliance teams make the final call.
What the PDPL asks of an AI workload
Royal Decree M/19, dated 9/2/1443 AH, issued the PDPL, and it has since been amended. SDAIA, the Saudi Data and AI Authority, oversees it and publishes its regulations on its data governance platform.
What the transfer regulation says
For AI, the key text is SDAIA’s Regulation on Personal Data Transfer Outside the Kingdom. This matters because a prompt that carries personal data to a model hosted abroad can count as a transfer or disclosure. In outline, the regulation works like this:
- A list of countries. Article 3 provides for a list of countries and international organisations with an appropriate level of protection, which SDAIA publishes and reviews.
- Specific cases. Article 4 lists cases that are exempt from some of the Law’s transfer conditions, and each one still needs its own safeguards.
- Risk assessments. Article 7 requires a risk assessment before a transfer in defined situations.
So, the regulation does not rule out a model hosted abroad. However, whether a given AI data flow qualifies depends on its details. For that reason, map each flow against these rules with your privacy counsel before you build it.
Proposed changes to watch
The rules are moving. In 2026, SDAIA opened a public consultation on amendments to the PDPL Implementing Regulation. Proposed Article 23 says the controller “shall store Personal Data within the Kingdom”. However, the same text says personal data may still be transferred or disclosed outside the Kingdom in line with the Law and its Implementing Regulations. For now, it is a proposal, not law. Even so, it is a good reason to prefer in-Kingdom processing for new AI services.
Sector rules sit on top of the PDPL
However, many organisations face stricter rules than the PDPL alone. Three matter most for AI hosting:
- NCA Cloud Cybersecurity Controls. The current version is CCC-2:2024. Its list of updates (Annex D) moved the data localisation controls to the National Data Management Office (NDMO) at SDAIA. It also tells entities to consult the NDMO before they act on localisation.
- CST cloud regulations. CST Decision 506/1445 (October 2023) approved version 4 of the Cloud Computing Service Provisioning Regulations. As a result, these replaced the older Cloud Computing Regulatory Framework.
- Data classification. Government bodies classify data under the NDMO’s framework, and the class of a dataset can affect where you may host it. However, we could not open the official policy text, so confirm the levels and hosting rules with your data office.
In practice, check each dataset’s class against the NDMO’s policy before you choose where a model runs. Also, a design that processes data in the Kingdom from the start keeps future audits simple.
Where models can run in the Kingdom today
A cloud region in the Kingdom does not guarantee that the model runs there. For example, some services keep storage local but process prompts elsewhere. The table below therefore shows what each vendor’s own pages said when we checked.
| Provider | Saudi Arabia | What to check |
|---|---|---|
| Google Cloud (Vertex AI) | Dammam region (me-central2) appears in the Vertex AI generative AI locations list | Google states that “endpoints don’t guarantee data residency or in-region ML processing”. Confirm the exact model and endpoint. |
| Oracle Cloud (OCI Generative AI) | Saudi Arabia Central (Riyadh, me-riyadh-1), listed with local model processing | Which models the region offers, and on-demand or dedicated clusters |
| Microsoft Azure | Microsoft lists a Saudi Arabia East region | We could not confirm that the region is live, or which Foundry models it offers. |
| AWS (Amazon Bedrock) | A Saudi region is “on track to launch in December 2026” | Not live when we checked. Bedrock lists no Saudi region yet. |
| OpenAI API | No Saudi data residency region | The UAE region offers regional storage and limited regional processing |
| Anthropic Claude API | No Saudi option | The direct API offers “us” and “global” inference only. On a cloud platform, the platform’s region decides. |
Finally, open-weight models give you another option. In that case, you run them yourself on in-Kingdom cloud infrastructure or on-premises. As a result, you control the location, but you also own patching, capacity and evaluation.
What this means for SAP data
SAP data often carries the most risk in the business, because it holds employees, customers, prices and payments. As a result, three facts from SAP’s own documentation shape the design:
- SAP BTP runs in the Kingdom. SAP lists two Cloud Foundry regions on Google Cloud in Dammam: sa30 for regulated customers and sa31 for non-regulated customers. SAP’s January 2025 announcement describes the regulated region for government bodies and critical national infrastructure operators.
- Joule has no Saudi data centre row. SAP’s Joule data centre list includes Dubai (AE01) but no Saudi entry. Also, SAP’s document grounding page warns that its large language models “are deployed outside of the Google Cloud sa30 region”.
- Generative AI Hub mixes hosting models. SAP AI Core offers SAP-hosted open-source models and remote models on external platforms. However, we could not confirm availability in sa30 or sa31. Ask SAP for the current region and model list before you plan on it.
In short, a KSA-hosted SAP system can still send prompts abroad through an AI feature. Therefore, check each AI feature separately, not just the SAP landscape.
Four patterns that work
We use four patterns, and most programmes combine them. Our page on sovereign and private AI covers the architecture in more depth.
- In-Kingdom managed model. Use a model that the provider processes in a Saudi region. This suits personal and confidential SAP data, once you confirm the region and model.
- Private or open-weight model. Run the model in your own in-Kingdom cloud tenancy or data centre. This suits regulated data, and it also suits tasks a smaller model handles well.
- Split processing. Keep SAP records in the Kingdom and send a foreign model only masked or derived text. Masked text that can still identify a person indirectly remains personal data, so confirm the result with your privacy team.
- Cross-border for non-personal data. Use a global model for public or non-personal data, such as product documentation. Then record the decision and review it when the rules change.
A model router then sends each task to an approved model for its data class. Sovereign AI and a private AI cloud also differ, and the sovereign and private AI page compares them.
A checklist before the first production release
- List the SAP tables and documents each AI use case reads, and classify them.
- Name the model, the region and the deployment type for each data class.
- Ask each vendor where it processes prompts, logs and fine-tuning data, not only where it stores them.
- Map each AI data flow that leaves the Kingdom against the transfer regulation with your privacy counsel, including whether a risk assessment applies.
- Then test Arabic and domain quality on the in-Kingdom model before you commit to it.
- Set an owner to re-check regions and rules every quarter, because both change quickly.
Our AI governance, evaluation and security work turns this list into controls and tests.
How Altivate helps
Altivate is an SAP Gold Partner, a member of the OpenAI Partner Network and a member of the Claude Partner Network. We hold ISO/IEC 27001 certification for information security and ISO/IEC 42001 certification for AI management.
Our forward-deployed AI engineers work inside your team. First, they map the data boundary. Then they benchmark the candidate models and build the first governed route to production on SAP data.
Frequently asked questions
Can we use ChatGPT or Claude with SAP data in Saudi Arabia?
Yes, with care. Neither direct API offered a Saudi processing region when we checked. So, either send only non-personal data, or confirm with your privacy counsel that the transfer meets the PDPL’s rules. Masking helps, but masked data that can still identify someone indirectly remains personal data. Alternatively, choose a model your cloud platform processes in the Kingdom.
Does a Saudi cloud region mean the AI runs in Saudi Arabia?
Not always, because storage, processing and logging can sit in different places. Google, for example, says its Vertex AI endpoints do not guarantee in-region ML processing. So, check each model and feature.
Would the proposed amendments ban foreign AI models?
Not as drafted. The proposed Article 23 of the Implementing Regulation adds a storage duty, but the same text says personal data may still be transferred or disclosed outside the Kingdom in line with the Law and its Implementing Regulations. So, watch for the final text before you change a live design.
Sources
We checked each source on 18 November 2026. Cloud regions and model lists change often, so re-check them before you decide.
- SDAIA, Regulation on Personal Data Transfer Outside the Kingdom
- SDAIA via Istitlaa, Amendments to the Implementing Regulation of the Personal Data Protection Law (public consultation)
- NCA, Cloud Cybersecurity Controls (CCC-2:2024) and its full text
- CST, Decision 506/1445 on cloud computing regulations
- Google Cloud, Vertex AI generative AI locations
- Oracle, OCI Generative AI regions
- Microsoft, Azure geographies
- Amazon, AWS cloud region in Saudi Arabia
- OpenAI, API data residency
- Anthropic, Claude data residency
- SAP, SAP BTP Cloud Foundry regions and Joule data centres
Map your AI data boundary
Bring one AI use case and the SAP data it needs. We will map where each part can run and what a first production release would take.

