Gulf bets on owning AI infrastructure rather than building frontier models

Key Points

1. Saudi Arabia and the UAE are building gigawatt-scale AI capacity with rival US chip and cloud suppliers
2. Investor Judah Taub argues the Gulf's leverage lies in infrastructure both Washington and Beijing need
3. Compute is becoming strategic infrastructure, raising questions of physical security and sovereign control

The latest:

Saudi Arabia and the UAE are assembling AI computing capacity measured in gigawatts, working simultaneously with competing American chipmakers, cloud providers and model developers. Judah Taub, founder and managing partner of Israeli venture capital firm Hetz Ventures, argued that the Gulf’s advantage may lie in controlling the compute, energy and data centers both technology blocs depend on, rather than in producing a leading frontier model.

Details:

• The thesis: Taub, identified by Semafor as an investor focused on cybersecurity, data and AI infrastructure, argued that Riyadh and Abu Dhabi have enough capital, energy and state-backed infrastructure to build large-scale AI capacity without becoming extensions of Silicon Valley or China’s technology sector. He described the Gulf as potentially one of the few places where both ecosystems can operate.
• The Saudi vehicle: HUMAIN, the AI company backed by the Public Investment Fund, is building relationships with several competing providers rather than committing to a single stack. NVIDIA and HUMAIN announced plans in 2025 for AI factories of up to 500 MW over five years, while AMD separately announced a collaboration targeting a comparable 500 MW.
• From plan to production: AMD said on August 31, 2026 that HUMAIN had begun delivering production AI compute in Saudi Arabia using AMD Instinct MI355X GPUs, EPYC CPUs and Cisco networking. The next phase would add up to 250 MW from 2027, keeping an AMD-Cisco-HUMAIN target of as much as 1 GW by 2030.
• The cloud layer: AWS and HUMAIN announced more than $5 billion in joint investment in 2025 for an AI Zone in the kingdom, later expanded to include as many as 150,000 AI accelerators spanning NVIDIA infrastructure and AWS Trainium chips. AWS says its first Saudi cloud region launches in December 2026, with up to 50 MW of AI Zone capacity planned by 2028.
• The Emirati model: OpenAI announced Stargate UAE in May 2025 as its first international Stargate deployment and first project under OpenAI for Countries, with G42, Oracle, NVIDIA, Cisco and SoftBank participating and US government coordination. G42 says Microsoft is adding 200 MW through Khazna Data Centers, with deployment expected to begin before the end of 2026.
• Sovereign services: Abu Dhabi separately entered a sovereign-cloud agreement with Microsoft and Core42 covering government services, placing state data on infrastructure operated under domestic control rather than on standard commercial cloud regions abroad.
• The security turn: Reuters reported this month that the UAE was revising the design of its planned 5 GW AI campus after Iranian attacks on US-linked technology infrastructure in the Gulf. The revised concept would distribute facilities around the country and consider hardened or underground construction. No date was given for a final design.
• The caveat: Taub has previously warned that much Middle Eastern AI investment could prove wasteful if governments prioritize headline projects over commercially useful infrastructure. In a February 2026 Semafor column, he argued the region would be judged on fundamentals: reliable power, permitting, engineering talent and infrastructure customers actually use.
• The constraint: Advanced US chips and AI systems remain subject to national-security controls, and Washington has repeatedly scrutinized Gulf technology ties with China. The strategy depends on retaining access to restricted American technology while preserving enough autonomy to work with Asian suppliers.

Background:

Analysts increasingly describe the US-China AI contest as multidimensional, covering semiconductors, data centers, standards, deployment and robotics rather than frontier models alone. Carnegie’s discussion of the race similarly argues leadership cannot be reduced to a single technological finish line.

Between the lines:

The UAE’s redesign of a 5 GW campus after attacks on US-linked infrastructure suggests compute is entering the category occupied by pipelines and ports: assets whose physical security and sovereign control matter as much as their economics. Taub’s framing implies leverage without supremacy, though the oil analogy has limits he acknowledges — compute is easier to reproduce geographically, and semiconductor supply remains concentrated outside the Gulf.

What’s next

Watch AWS’s first Saudi cloud region launch, scheduled for December 2026, the start of Microsoft-G42 capacity deployment before year-end 2026, and AMD’s 250 MW phase beginning in 2027.