Middle East GPU Investment: Saudi Arabia and UAE AI Infrastructure Buildout

The Gulf region has emerged as an unexpected heavyweight in global GPU infrastructure, with Saudi Arabia and the UAE collectively committing over $50 billion to AI compute buildout through 2030. Saudi Arabia's Public Investment Fund (PIF) launched the Saudi Arabia AI Infrastructure Fund in 2024 with an initial $25 billion allocation for GPU procurement, data center construction, and AI research. The UAE's G42, in partnership with Microsoft's $1.5 billion investment, is building a pan-regional GPU network spanning Abu Dhabi, Dubai, and expanding into Egypt and Kenya. These investments are driven by a sovereign AI strategy that views domestic GPU capacity as critical infrastructure equivalent to oil refining capacity.

Power is the Gulf's natural advantage in GPU infrastructure. Saudi industrial electricity rates of SAR0.12-0.20 per kWh ($0.03-$0.05) are among the lowest in the world, and the UAE rates of AED0.18-0.28 ($0.05-$0.08) are comparably competitive. For a 1,000-GPU H100 cluster running continuously, the annual power cost in Saudi Arabia is roughly $150,000-$200,000 versus $1.2-$1.5 million in Germany or $800,000-$1.0 million in Singapore. This 5-10x power cost advantage is reshaping global GPU workload distribution, with large-scale training increasingly migrating to Gulf data centers despite latency tradeoffs for inference.

Saudi Arabia's GPU strategy centers on three pillars: the NEOM data center cluster, Aramco's industrial AI compute, and new builds in Riyadh and Jeddah. NEOM's OXAGON campus is designed as a 200 MW AI data center city with direct liquid cooling using Red Sea water at 22-26 degrees Celsius year-round, enabling PUE targets of 1.10-1.15. The initial phase of 40 MW is operational with H100 and H200 clusters deployed by Google Cloud and Oracle, serving both domestic AI workloads and export capacity to European markets via subsea cables through the Red Sea and Mediterranean. A second phase of 80 MW is under construction with completion expected in 2027.

Aramco's AI compute division operates an 8,000-GPU cluster at the Dhahran techno-valley campus, used primarily for oil and gas exploration AI, reservoir modeling, and industrial computer vision. The cluster runs a mix of H100 and AMD MI350X GPUs, with Aramco having signed a $500 million procurement agreement with AMD in 2024 as part of Saudi efforts to diversify GPU supply away from NVIDIA dominance. Riyadh's KACST (King Abdulaziz City for Science and Technology) campus hosts an additional 4,000 H100 GPUs for academic and government AI research. Saudi GPU pricing for reserved H100 capacity runs $2.50-$3.20 per GPU-hour, the lowest in the Gulf, reflecting the subsidized power and the PIF's strategic objective of attracting international AI workloads.

Abu Dhabi has established itself as the Gulf's leading AI infrastructure hub through the G42 ecosystem. The G42-Microsoft partnership, valued at $1.5 billion with an additional corporate structure that transferred $9 billion in assets, includes a commitment to build UAE-based GPU capacity for both sovereign AI applications and Microsoft's global AI workloads. The core facility is the Khazna Data Center campus in Abu Dhabi's Masdar City, a 150 MW AI-optimized facility running H100 and B200 clusters with direct liquid cooling. Total GPU count at Khazna is estimated at 10,000-15,000, making it one of the largest single-campus GPU deployments outside China and the United States.

Dubai's GPU market is smaller but more commercially oriented, with clusters at Dubai Internet City (Equinix DX3/DX4) and Dubai South serving fintech, media, and government AI workloads. The Dubai Multi Commodities Centre (DMCC) launched the DMCC AI Centre in 2025, offering subsidized GPU compute to AI startups at AED5-8 per GPU-hour ($1.36-$2.18). This has attracted 40+ AI companies to the Dubai ecosystem. Cross-border latency between Abu Dhabi and Dubai is 2-4ms, and from UAE to the broader Middle East and North Africa region is 20-50ms, positioning the UAE as the inference hub for the entire MENA region's AI applications.

Both Saudi Arabia and the UAE have export control regulations that require GPU providers to ensure NVIDIA GPUs are not re-exported to China or other restricted destinations. Providers operating in the Gulf must maintain end-user tracking and supply chain compliance as a condition of their operating licenses.

The Gulf's low power costs are its primary competitive advantage in GPU infrastructure, but ambient temperatures of 45-50 degrees Celsius in summer create cooling challenges that partially offset the power cost benefit. Standard air-cooled data centers in the region operate at PUE of 1.4-1.6, versus 1.2-1.3 in the Nordics or US Pacific Northwest. This means 25-35 percent more total power consumption for the same GPU compute load, reducing the effective power cost advantage to 2-4x versus 5-10x on raw electricity rates alone.

The solution being deployed across new Gulf GPU facilities is direct-to-chip liquid cooling and immersion cooling. NEOM's OXAGON uses seawater-cooled direct liquid cooling with inlet temperatures of 28-32 degrees Celsius, achieving PUE of 1.10-1.15. G42's Khazna campus uses rear-door heat exchangers with closed-loop water circulation, achieving PUE of 1.20-1.25 even during July peak temperatures. These cooling systems add 15-20 percent to facility capital costs but reduce operating power by 25-35 percent versus air cooling, making the economics work for large-scale GPU deployments.

Operating GPU infrastructure in the Middle East requires navigating a complex web of export controls and technology transfer regulations. The US Bureau of Industry and Security (BIS) applies enhanced vetting to GPU exports to the Middle East under the 2024 revised Chip Rules, which require validated end-user certificates for any shipment of H100 or H200 GPUs exceeding 500 units to Saudi Arabia or the UAE. NVIDIA has established a dedicated compliance team in Abu Dhabi to manage these licenses, which typically take 4-8 weeks for approval. Providers that violate these terms face export license revocation and potential inclusion on the Entity List.

China's response to US export controls has increased Gulf GPU demand by proxy. Chinese AI companies unable to access H100 directly have established joint ventures with Gulf entities to co-locate GPU clusters in Middle East data centers, with the argument that the GPUs remain physically outside China and under Gulf sovereignty. These structures are under active legal scrutiny by BIS as of 2026, and GPU providers in the Gulf should perform thorough beneficial ownership checks on any China-linked entity seeking GPU capacity. The regulatory uncertainty contributes a 15-25 percent risk premium to Gulf GPU pricing compared to equivalent capacity in the European Union.