Sovereign AI as a Pillar of Oman Vision 2040

Read Oman Vision 2040 alongside its implementing programmes and a single argument falls out: the country has committed to lift the digital economy from 2 percent of GDP in 2021 to 10 percent by 2040, and it has named artificial intelligence inside the instruments that deliver that lift. For sovereign buyers, the relevant question is no longer whether AI fits the national plan. It is which slice of the AI estate has to live inside the institution's own walls, and which can sit on a shared national layer. This article maps that split against the Vision text.

1. Where Vision 2040 actually puts digital and AI

The Vision document, published by the Oman Vision 2040 Implementation Follow-up Unit, organises the country's 20-year strategy around four national priorities: People and Society; Economy and Development; Governance, Institutional Performance and Resource Management; and a Sustainable Environment. Inside Economy and Development, "a diversified and sustainable economy" lists ICT as one of the productive sectors expected to drive non-oil GDP. The phrase "knowledge-based economy" recurs across both the Economy axis and the Governance axis.

The explicit AI vocabulary lives one layer down, in the implementing programmes. The National Programme for Artificial Intelligence and Advanced Digital Technologies, run by MTCIT, states its aim as enhancing AI applications "in economic and developmental sectors", localising advanced digital technologies, and achieving effective AI governance with a human-centred lens. It supersedes the 2022 Executive Programme for AI and Advanced Technologies and sits as one component of the National Digital Economy Programme alongside Government Digital Transformation, Infrastructure, Digital Industry, E-Commerce, and Space.

The same documents quantify the economic stake: doubling the digital economy's contribution to GDP from 2 percent in 2021 to 10 percent by 2040. That is the number every Vision-aligned AI procurement is, eventually, asked to defend.

2. The diversification and sovereignty argument for on-prem AI

Two strands of the Vision converge on sovereign on-premise AI. The first is economic diversification: every riyal the country spends on hyperscaler tokens to summarise its own circulars, anonymise its own claims, or draft its own correspondence is a riyal exported. On-premise appliances move the spend onto Omani balance sheets: local integrators, local power, local maintenance contracts, locally trained operators. They convert one-off licence fees into recurring local payroll.

The second strand is institutional sovereignty. The Governance pillar of Vision 2040 emphasises rule of law, anti-corruption, and effective public administration. Several institutions inside that pillar (the Royal Office equivalents, the State Audit Institution, the central bank, the police, the judiciary) work with material whose loss carries constitutional, not commercial, weight. For these bodies, the threshold for a citizen-data prompt leaving Omani jurisdiction is not "is the cloud provider audited?" but "does the law allow this prompt to cross the border at all?" The honest answer is usually no, which leaves on-premise as the only compliant deployment shape. We sketched the architecture in detail under on-premise AI for sovereign institutions in Oman and the GCC.

The 2026 launch of Omantel's Otech sovereign cloud platform, billed in the local press as the first AWS-accredited sovereign cloud in the Middle East, sets the macro context: there is now a sanctioned national landing zone for sovereign cloud, but the most regulated workloads still need a deeper guarantee than a sovereign cloud account. They need an air-gap and a key the institution holds itself.

3. National workforce and skills implications

Vision 2040 is, at heart, a workforce document. The People and Society priority invests in education, scientific research, and human capability; the Economy priority asks the labour market to absorb that capability into productive sectors. AI procurement only earns a Vision-aligned tick when it produces national skill, not just throughput.

• Tooling, not magic. Sovereign deployments expose the full stack to Omani operators: prompt design, retrieval pipelines, evaluation harnesses, drift dashboards, fine-tuning runs. Hyperscaler APIs hide all of that. Skills compound only when the substrate is visible.
• Reusable institutional muscle. A ministry that runs its own AI inventory and its own evaluation suite (see building an AI governance function inside an Omani ministry) builds people who can be redeployed to the next ministry. SaaS-only stacks build vendor-management contracts, which do not transfer.
• Arabic at the centre. Models like Falcon Arabic and Qwen 3.6 already match closed-source quality on Omani-style MSA. Local fine-tuning and evaluation work happens on local hardware or it does not happen at all.

The National Programme funds the supply side directly: a National Centre for AI Research and Development, an AI Studio, scholarships, and partnerships with the World Economic Forum's Centre for the Fourth Industrial Revolution. The demand side is every sovereign procurement that insists on Omani delivery teams, on-premise transfer-of-skill clauses, and Arabic evaluation suites baked into acceptance criteria.

4. Mu'een and on-premise sovereignty co-exist

The 2026 to 2030 digital-economy roadmap names a national AI platform aimed at government decision support, summarisation, and Arabic content generation. Mu'een, Oman's national shared AI platform, is the public-sector productivity layer: a government employee in any ministry can use it to draft, translate, or analyse non-classified content under a single national procurement, a single national governance frame, and shared Arabic-quality investments.

That layer does not replace institutional sovereignty. It complements it. The split, deployed honestly, looks like this:

1. Mu'een handles common workloads. Drafting circulars, polishing reports, summarising long Arabic PDFs at the public-document classification level.
2. Institutional on-premise handles controlled workloads. Citizen records, prudential supervision, criminal intelligence, audit findings, internal correspondence flagged restricted-and-above. Same models, often the same open weights, but inside the institution's own air-gapped rack with its own keys, logs, and decommission rights.
3. The boundary is data classification, not vendor preference. A ministry that runs its own classifier and pushes "Public" prompts to Mu'een and "Restricted+" prompts to its own appliance is doing exactly what the Vision asks: shared infrastructure where it scales, sovereign infrastructure where the law and the threat model demand it.

That is the operational meaning of "knowledge-based diversified economy" once you stop reading it as a slogan: a national platform that maximises productivity, and an institutional substrate that holds anything the country cannot afford to lose. Both are Vision 2040 work.

Briefing. If you are mapping your institution's AI workloads against this two-layer split before your next budget cycle, write to [email protected] for a one-hour briefing. We bring a classification template, a Mu'een-vs-on-prem decision matrix, and a sample procurement clause set tuned to MTCIT and NCSI expectations. By quotation.