Saudi Arabia has spent the past year trying to buy its way out of a single point of failure. HUMAIN, the AI company the Public Investment Fund launched in 2025 to anchor Vision 2030's AI ambitions, has signed a roughly $10 billion, 500-megawatt compute deal with AMD, a $2 billion inference agreement with Groq, and an October 2025 memorandum of understanding with Qualcomm to fill 200 megawatts of data-center capacity with its AI200 chips. On June 1, 2026, at Nvidia's GTC Taipei event, HUMAIN also became a launch partner for Nvidia's DRIVE Hyperion robotaxi platform, agreeing to bring Level 4-ready autonomous vehicles to the Kingdom. On paper, this looks like genuine supplier diversification.
It isn't, and a July 14, 2026 Rest of World analysis lays out why: every one of HUMAIN's alternatives to Nvidia is itself a U.S. company, which means every one of them is subject to the same export licensing regime as Nvidia. "Diversification away from Nvidia reduces the Gulf's dependence on any one commercial vendor, but it doesn't change the political risk of dependence on the U.S. at all," Carnegie Endowment fellow Sam Winter-Levy told the outlet. Nvidia's data center division alone generated $51.2 billion in the third quarter of fiscal 2026, and its CUDA software ecosystem has roughly 4 million developers locked in — a moat AMD, Groq, and Qualcomm are only beginning to erode, and only for inference, not the training workloads Vision 2030's most ambitious AI goals depend on.
The licensing regime is real, and its logic isn't crazy
Before arguing that this system is dysfunctional, it's worth stating plainly why Washington built it. Advanced AI accelerators are dual-use technology in the fullest sense: the same Blackwell-class silicon that trains a Saudi Arabic-language model can train a weapons-targeting system, and the same chip that sits in a Riyadh data center today can be re-exported tomorrow to an entity on the Commerce Department's restricted list. The U.S. has legitimate reasons to want assurance that Gulf compute capacity isn't quietly becoming a backdoor for Chinese firms cut off from advanced nodes by their own export controls. When the Commerce Department approved chip sales to HUMAIN and the UAE's G42 on November 19, 2025 — each cleared for up to 35,000 Nvidia GB300-equivalent systems — it attached cybersecurity standards, usage monitoring, and re-export restrictions specifically to prevent diversion to sanctioned firms like Huawei. That is a defensible design for a genuine national-security exposure, not regulatory overreach for its own sake.
But the regime is arbitrary, not rules-based — and that's the actual problem
The trouble is how unevenly and personality-dependently this approval process has run. The November 2025 Humain and G42 approvals followed a Saudi Crown Prince Mohammed bin Salman visit to Washington; under the Biden administration's earlier framework, Saudi Arabia and the UAE had been capped at roughly 1,700 advanced chips a year before the Trump administration scrapped country-tiered restrictions in May 2025. Then, on July 10, 2026, the Commerce Department's Bureau of Industry and Security moved the UAE out of Export Administration Regulations Country Groups D:3 and D:4 entirely and into Country Group A:5 — the same trust tier as the UK, India, and South Korea — granting the UAE government and approved entities license-free access to advanced computing items under the May 2025 U.S.-UAE AI Cooperation framework. Saudi Arabia was not included in that reclassification. It remains one licensing decision, one bilateral summit, one diplomatic reassessment away from either expanded access or a sudden freeze.
That asymmetry is the real story here, more than the chip counts themselves. A rules-based export framework — publish the security and monitoring conditions once, let any Gulf partner that meets them graduate into a stable tier — would achieve Washington's actual security goal (preventing diversion to China) without making a $600 billion national AI strategy hostage to the optics of a state visit. Instead, Saudi Arabia is left negotiating case by case while its neighbor operates under a durable, published rule. If Vision 2030's technology sector is going to attract the private capital SDAIA's National Strategy for Data and AI needs to hit its 2030 targets, investors need to be able to price policy risk. An opaque, relationship-driven approval process is close to the worst way to deliver certainty — and it doesn't even fully serve the stated security rationale, since a Kingdom uncertain about its long-term U.S. chip access has every incentive to hedge by keeping its options with Chinese suppliers open, the exact outcome the licensing regime exists to prevent.
The innovation case for predictability
None of this is an argument against export controls on frontier compute — the underlying national-security logic holds up. It's an argument that the Trump administration's own July 23, 2025 AI Action Plan, which frames "secure, full-stack AI export packages" to allies as a tool for extending American AI dominance, undercuts its own goal by leaving a close security partner in policy limbo. Extending the UAE's rules-based A:5 treatment to Saudi Arabia, conditioned on the same anti-diversion safeguards already imposed on HUMAIN, would let Riyadh's chip diversification actually mean something — rather than functioning as a hedge against the same government that authorized the diversification in the first place.