Constitutional Position: The Authority That Governs AI Governance
Saudi Data and Artificial Intelligence Authority occupies a position in the Saudi state architecture that has no precise Western equivalent. Established by royal order in August 2019 — a year when most governments were still debating whether they needed an AI strategy — SDAIA was granted horizontal authority across every ministry and public body in the Kingdom. It does not serve a single sector. It does not report to a line ministry. It reports, through its governance structure, to the Crown Prince’s office, and its mandate is to coordinate data and AI policy across the entire apparatus of the Saudi state.
That constitutional position is the first thing to understand about SDAIA. In most countries, AI policy is fragmented: the finance ministry handles data in banking, the health ministry handles health data, the telecom regulator handles data localization. SDAIA collapses that fragmentation into a single authority. When SDAIA sets a standard, it sets it for everyone. When SDAIA decides that a particular AI application requires an ethics review, that decision applies government-wide. The institutional design reflects a deliberate choice: Saudi Arabia would not allow AI governance to be as siloed as its pre-2019 digital infrastructure was.
The royal order establishing SDAIA also embedded it in the Vision 2030 architecture from the outset. Vision 2030 was already three years old when SDAIA was created, but the Kingdom had come to recognize that data and AI were not peripheral to Vision 2030’s diversification ambitions — they were the enabling layer. SDAIA was the institutional expression of that recognition. The choice to establish SDAIA as a standalone authority rather than a department within an existing ministry was deliberate and consequential: it gave SDAIA budget autonomy, staffing flexibility, and the political standing to direct other ministries rather than coordinate with them as a peer. See Vision 2030 Framework for the broader strategic context within which SDAIA operates.
The National Data Bank: Saudi Arabia’s Sovereign Training Corpus
The most strategically important asset SDAIA controls is not a model or a data center. It is the National Data Bank — an integrated repository connecting more than 430 government systems as of 2025. This makes the National Data Bank almost certainly the largest consolidated sovereign Arabic-language administrative dataset in the world.
To understand why this matters, consider the economics of large language model training. The quality and breadth of training data determines what a model can do. Arabic-language data has historically been underrepresented in global model training pipelines: most foundational LLMs were trained on English-dominant corpora, with Arabic as a secondary language that the models handled adequately but not natively. The National Data Bank changes the calculus entirely. When SDAIA trains Allam, it is drawing on government records, administrative data, healthcare records (with appropriate anonymization), legal documents, and public service interactions — all in Arabic, all reflecting the specific linguistic and cultural register of Saudi society, spanning decades of digitized institutional memory.
The 430+ integrated systems span every major government function: civil registration, tax authority, social insurance, education, healthcare, judicial records, and more. This is not raw web scrape data of uncertain provenance; it is structured, validated, administratively significant data that reflects how 35 million people actually interact with their government. For an Arabic-first AI platform, this is a foundational advantage that cannot be replicated by any private company and would take a decade for any other Gulf state to assemble at equivalent depth.
The data integration effort was not trivial. Saudi Arabia’s pre-2019 government IT landscape was fragmented, with different ministries running incompatible systems on different infrastructure. SDAIA’s National Information Center (NIC) — one of its five subsidiary units — handled the technical integration work: building APIs, establishing data standards, negotiating data-sharing agreements across ministries, and implementing the governance framework that made cross-ministry data sharing legally and procedurally possible under the Saudi Personal Data Protection Law (PDPL). The NIC effort amounted to a wholesale modernization of Saudi government data infrastructure, executed in parallel with building the AI capabilities that would consume that data.
KSA-RoD and Data Residency
A critical aspect of the National Data Bank’s legal architecture is the KSA-RoD (Kingdom of Saudi Arabia — Residency of Data) requirement. Government data in Saudi Arabia must, under this framework, be stored on infrastructure physically located in the Kingdom. This requirement shapes every international cloud provider’s Saudi strategy: Microsoft, Google, AWS, and others have all invested in Saudi-based data center infrastructure specifically to comply with KSA-RoD and serve Saudi government customers. The data residency requirement also reinforces SDAIA’s sovereign compute logic — government AI workloads cannot simply be offloaded to foreign cloud infrastructure, which is part of the justification for SDAIA’s 5,000-GPU Blackwell deployment.
NDMO and PDPL Implementation
The National Data Management Office (NDMO), another SDAIA subsidiary, serves as the regulatory implementation arm for Saudi Arabia’s data governance framework. The PDPL, which came into force in 2022 with implementing regulations following in subsequent years, established the legal framework for how personal data can be collected, processed, shared, and transferred outside the Kingdom. NDMO operationalizes that framework: issuing guidance, reviewing compliance, and in certain cases granting the cross-border data transfer authorizations that international cloud providers need to serve Saudi customers.
SDAIA’s regulatory posture on AI is notably different from the EU AI Act approach. Where the EU has adopted a risk-based, prescriptive framework with significant pre-market conformity requirements, SDAIA has issued Generative AI Guidelines and AI Ethics Principles that are principles-based and outcome-oriented. The approach defers to sector regulators — the Saudi Central Bank handles AI in financial services, the Saudi Food and Drug Authority handles AI in healthcare — rather than creating a separate AI regulator. This permissive-but-supervised approach has been attractive to international technology companies considering Saudi deployment, as it avoids the compliance overhead that has made EU AI Act conformance expensive for global deployments. International companies frequently describe SDAIA’s regulatory approach as enabling rather than constraining — a framing that SDAIA has actively cultivated to position Saudi Arabia as an AI investment destination.
Allam: The National Large Language Model
Allam is SDAIA’s most visible product and the clearest demonstration of what the National Data Bank enables. The model carries 34 billion parameters — comfortably in the range of capable large-scale LLMs — and was trained on a corpus of 8 petabytes of Arabic-weighted data. The training corpus drew on the National Data Bank’s government data alongside curated Arabic text corpora from media, literature, religious texts, and digital communications.
The IBM partnership for Allam deployment is strategically significant. SDAIA launched Allam on IBM’s Watsonx platform, which gave the model an enterprise-grade deployment infrastructure and a path to integration with IBM’s broader enterprise software customer base. The Watsonx deployment also gave SDAIA access to IBM’s model governance tooling — important for a government entity that needed to demonstrate responsible AI deployment, not just raw capability. The Watsonx relationship makes Allam available to IBM’s global enterprise software customers as a component, potentially expanding the model’s reach well beyond Saudi Arabia without requiring SDAIA to build a global distribution infrastructure.
Allam is deployed across multiple government use cases: administrative document processing, citizen services chatbots, and as the underlying model for Humain Chat — the Arabic-first consumer AI assistant that Humain launched as part of its consumer product stack. The Allam-to-Humain-Chat pipeline represents the division of labor between SDAIA and Humain in practice: SDAIA develops and maintains the foundational Arabic model; Humain productizes it for consumer and commercial applications.
The “Saudi Arabia leads in Arabic AI with launch of HUMAIN Chat and ALLaM 34B model” framing from Economy Middle East in 2026 captures the positioning SDAIA has achieved: not merely an Arabic-capable model, but a genuine leader in Arabic-native AI. This matters for the broader export ambition. If Saudi Arabia is to become a token exporter — as Humain CEO Tareq Amin has stated explicitly — the Arabic AI stack needs to be the world’s best. Allam is the foundation of that claim. At 34 billion parameters, Allam is competitive with many capable open-weight models in English; its Arabic-language performance advantage over global models is likely significantly larger, given the training corpus depth.
Compute Infrastructure: 5,000 NVIDIA Blackwell GPUs
SDAIA’s sovereign compute footprint is anchored by a confirmed deployment of 5,000 NVIDIA Blackwell GPUs. This positions SDAIA as one of the largest government-direct AI compute operators in the world, separate from the much larger commercial deployments being assembled under Humain.
The Blackwell GPU selection reflects the same silicon pipeline that runs through the entire Saudi AI buildout: NVIDIA as the primary hardware vendor, Grace Blackwell architecture as the preferred compute substrate. The 5,000-GPU deployment is sized for government workloads — training and fine-tuning government-specific models, running Allam inference at national scale, supporting the AI applications that 430+ integrated government systems will eventually expose to the public.
This compute infrastructure sits within SDAIA’s direct operational control, which is important for the sovereignty logic. Government workloads — citizen data processing, national security applications, administrative AI — cannot run on foreign-controlled infrastructure without creating data sovereignty risks. The SDAIA compute cluster provides the sovereign envelope: Saudi-controlled hardware, Saudi-controlled operating environment, Saudi-controlled access permissions. The scale of 5,000 GPUs is substantial for government-direct operation: most national AI compute clusters operated directly by government authorities are significantly smaller. SDAIA’s deployment signals that Saudi Arabia treats government AI compute as a matter of national security infrastructure rather than simply an IT procurement decision.
The Five Subsidiaries: Division of Labor
SDAIA’s five subsidiary units each address a distinct dimension of the data and AI mandate:
The National Center for Artificial Intelligence (NCAI) is the research and capability development arm, responsible for building Saudi AI technical talent, funding AI research, and developing the capabilities that feed back into SDAIA’s model development and policy work. NCAI operates in partnership with Saudi universities and international research institutions, building the academic AI ecosystem that will eventually produce the PhD-level researchers and engineers that sophisticated AI development requires.
The National Data Management Office (NDMO) is the regulatory implementation arm, handling PDPL compliance, data governance standards, and cross-border data transfer authorizations. NDMO’s decisions directly affect every international cloud provider operating in Saudi Arabia, making it one of the most practically consequential SDAIA units from an industry perspective.
The National Information Center (NIC) is the technical infrastructure arm, managing the government IT systems, the National Data Bank integrations, and the SDAIA compute infrastructure. NIC is the operational backbone of the entire SDAIA enterprise — without its technical capacity, the National Data Bank would be a policy concept rather than a functional asset.
The Research arm (distinct from NCAI in operational focus) handles applied research partnerships with universities and international institutions, focusing on near-term applications of AI to Saudi-specific challenges rather than foundational AI research.
The National Center for Digital Acceleration (NCDAI) drives AI adoption across government ministries — the implementation arm that converts SDAIA’s policies and models into deployed government services. NCDAI is the interface between SDAIA’s AI capabilities and the 100+ government ministries, agencies, and public bodies that are expected to leverage them.
Key Bilateral Relationships
SDAIA’s partnership architecture reflects a deliberate multi-vendor strategy that serves both functional and geopolitical purposes. The IBM relationship anchors the Allam deployment on Watsonx and provides enterprise software integration. The Microsoft Azure relationship gives SDAIA access to global cloud infrastructure for workloads that do not require sovereign compute, and Microsoft’s investment in Saudi-based Azure infrastructure addresses KSA-RoD requirements. The Databricks partnership — worth $500 million — provides the data lakehouse and MLOps infrastructure that makes the National Data Bank operationally useful for AI training pipelines. Databricks’ data engineering and governance tools are the operational plumbing that converts 430+ integrated government systems from a theoretical asset into an actual training corpus. The NVIDIA relationship for the 5,000 Blackwell GPU deployment is foundational for compute; SDAIA has been careful not to make it exclusive, maintaining vendor diversification across the software and cloud layers even while concentrating on Blackwell for raw compute.
The multi-vendor approach serves two strategic purposes simultaneously: it avoids single-vendor lock-in and the pricing leverage that comes with it, and it positions Saudi Arabia as a sophisticated technology buyer capable of managing complex multi-partner relationships. This sophistication is itself a strategic asset — international technology companies engaging with SDAIA encounter an organization capable of nuanced commercial negotiation, which is not uniformly true of government technology bodies in the region.
Division of Labor: SDAIA vs. Humain
The relationship between SDAIA and Humain is the defining structural feature of Saudi Arabia’s AI governance architecture. The division of labor is conceptually clean: SDAIA owns policy, regulation, and government AI workloads; Humain owns commercial AI infrastructure and products. In practice, the lines are blurrier and the coordination challenge is ongoing.
Allam sits precisely at the intersection: SDAIA develops and owns it; Humain productizes it through Humain Chat. The National Data Bank trained Allam; Humain is the primary commercial deployer. SDAIA sets the regulatory framework within which Humain operates; Humain’s commercial activities generate the economic outcomes that Vision 2030 requires. This mutual dependency creates a relationship that is simultaneously collaborative and potentially competitive — as Humain’s commercial ambitions grow, it may develop requirements for model capabilities that diverge from SDAIA’s policy-focused priorities.
The governance interface between the two organizations is still being established. SDAIA is a government authority with a regulatory mandate; Humain is a PIF-backed commercial entity with a revenue mandate. These different institutional identities create different decision-making rhythms, accountability structures, and performance metrics. Managing the boundary between them in a way that preserves SDAIA’s regulatory independence while enabling Humain’s commercial agility will require institutional discipline that is being developed in real time.
What to Watch
The key indicators for SDAIA’s execution trajectory: the rate of government AI service deployment (how many of the 430+ National Data Bank systems are actively feeding AI applications rather than just technically integrated); the pace of PDPL implementation and its practical impact on international cloud providers’ Saudi government contracts; and the Allam roadmap — specifically whether SDAIA releases Allam 2.0 with expanded parameters and whether it pursues open-source release to establish Saudi Arabia as a contributor to global Arabic AI infrastructure. Watch also for SDAIA’s posture in international AI governance forums, where it is increasingly positioning as the authoritative voice of Arabic-language AI interests against English-dominant global standards bodies. The degree to which SDAIA succeeds in that positioning will determine whether Allam becomes a regional standard or remains a national asset.
Key relationships: Humain, NVIDIA, Vision 2030, PIF, Mohammed bin Salman. See also Silicon Pipeline, Infrastructure.