Saudi Arabia vs UAE AI Race Timeline 2016–2026: Who Is Building the Gulf’s AI Hub?

Arabic is spoken by roughly 400 million people across widely different dialects — Gulf, Levantine, Egyptian, Maghrebi — alongside Modern Standard Arabic used in formal writing, media and government. Global general-purpose models have historically performed less consistently on Arabic than on English, particularly on dialectal text, code-switching (mixing Arabic and English mid-sentence, common in Gulf business contexts) and culturally specific reasoning. That gap is not fixed, and it narrows with every model generation — but it is real enough that both Riyadh and Abu Dhabi treat homegrown Arabic-first models as strategically distinct from simply licensing a Western model.

This is why Jais was trained on 1.6 trillion Arabic-and-English tokens specifically, and why ALLaM’s benchmark win was reported as an Arabic-MMLU result, not a general-purpose one. Government services, religious and legal text, and regional customer service are the workloads where a model tuned for regional Arabic and cultural context has a real, measurable edge over a generic multilingual model — independent of which country built it.

Building a data center and stocking it with the world’s most advanced AI accelerators are two different problems with two different gatekeepers. The first is a construction and capital problem. The second runs through U.S. export-control policy, because Nvidia’s and AMD’s most advanced chips — and the high-bandwidth memory and advanced packaging that go into them — are treated as dual-use technology with potential military applications.

Both Saudi Arabia and the UAE hit real restrictions here for years. That changed unevenly through 2025–2026: in November 2025 the U.S. approved initial advanced-chip export volumes for each country’s designated “AI champion” (HUMAIN for Saudi Arabia, G42 for the UAE). Then, in July 2026, the U.S. Commerce Department moved the UAE specifically into Country Group A:5 — its top export tier — letting approved UAE entities import advanced chips without an individual license each time. Saudi Arabia did not receive the same upgrade and, as of this update, still needs a license for every shipment.

Both countries have large hydrocarbon-funded state budgets, but a data center runs on electricity, not crude oil directly. Saudi Arabia’s grid leans heavily on natural gas with a growing solar buildout tied to NEOM and other renewable-energy projects; the UAE’s 5-GW campus is designed around what G42 calls a “tri-fuel” mix — dedicated nuclear power from the Barakah plant, utility-scale solar, and natural gas backup for 24/7 reliability. Barakah gives the UAE a source of firm, low-carbon baseload power Saudi Arabia’s current grid does not have an equivalent to yet, though neither country has published a facility-specific power-mix breakdown for its AI campuses.

Heat is the more universal constraint: AI accelerators are dense heat sources, and Gulf ambient temperatures regularly exceed 45°C in summer. Neither country’s cooling design has been detailed publicly at a per-facility level, but the general options are the same ones used globally — air cooling (simplest, least efficient at scale), chilled water and hybrid systems, and increasingly direct-to-chip liquid cooling for the newest, densest racks. Water availability, not heat alone, is often the harder constraint in a desert environment; it is a genuine design variable, not an argument that Gulf data centers are inherently inefficient.

Access to the most advanced U.S.-origin AI chips, cloud technology and HBM/advanced-packaging supply chains now depends on export licensing, end-use controls and government-to-government security agreements — not simply a purchase order. The Microsoft-G42 Intergovernmental Assurance Agreement (2024) and the U.S.-UAE AI Acceleration Partnership (May 2025) are the diplomatic scaffolding that made the UAE’s July 2026 license-free upgrade possible; Saudi Arabia’s parallel relationship with Washington has produced comparable initial approvals but not yet the same blanket regulatory status.

Both Saudi Arabia and the UAE also maintain significant, longstanding economic relationships with China — in telecoms, trade and infrastructure generally, though neither country’s AI-specific infrastructure (HUMAIN, G42, Stargate UAE) is built on Chinese chip technology at present, given U.S. export-control conditions attached to the American partnerships above. This is not a simple “choose a side” story: it is two governments managing overlapping, sometimes competing, relationships with both Washington and Beijing simultaneously, with U.S. technology-security policy setting real constraints specifically around advanced-chip access.

Three things will most shape how this comparison reads a year from now. First, whether Saudi Arabia receives its own Country Group A:5-equivalent chip-access upgrade, closing the regulatory gap the UAE opened in July 2026. Second, whether Stargate UAE’s 200-MW first phase actually completes on its Q3 2026 target, and whether the UAE’s reported move toward distributed, hardened sites becomes an official redesign or stays a set of options under review. Third, whether HUMAIN converts any of its 6.6-GW decade target into a dated, under-construction milestone as specific as Stargate UAE’s — the single clearest gap between the two countries’ public communications right now.

Saudi Arabia and the UAE are not simply competing to spend the most money on AI. They are trying to turn capital, electricity, chips, data centers, research, talent and international partnerships into durable computing capability. Measure those things separately — and let the evidence, not a manufactured winner, decide who is actually ahead.