How US Export Controls Affect Saudi AI Development

The May 2025 US-Saudi AI cooperation framework, headlined by Humain’s announced procurement of up to 18,000 NVIDIA H100 chips and a pathway to much larger volumes, was the most visible signal that US export control policy and Saudi AI ambition have reached a negotiated equilibrium — for now. But the underlying regulatory architecture remains in place, continues to constrain Saudi AI development in ways that are not always visible from the deal announcements, and could shift significantly depending on US geopolitical calculations.

This guide explains what the current export control framework actually restricts, how Saudi Arabia has navigated it, and what alternative strategies are in play for a sovereign AI buildout that cannot afford to be dependent on a single regulatory permission that could be revoked.

The Current AI Diffusion Framework: Tier-2 Classification

Saudi Arabia sits in Tier 2 under the BIS AI Diffusion Rule finalized in January 2025. This classification determines what AI chips Saudi entities can receive and under what conditions.

What Tier-2 Means

Tier-2 countries can receive advanced AI chips — NVIDIA H100, H200, GB200, and equivalents — but with significant constraints:

  • Individual transactions below the 1,700 H100-equivalent compute threshold proceed with standard export documentation but no individual license review
  • Transactions above this threshold require either a Validated End-User (VEU) agreement or an individual export license from BIS
  • All exports carry end-use monitoring obligations and re-export restrictions
  • Chips exported to Saudi Arabia cannot be re-exported to Tier-3 countries (China, Russia, Iran) without separate BIS authorization

The practical effect for Saudi Arabia: small-scale AI deployments (hundreds of GPUs) face minimal friction. Large-scale sovereign AI compute (tens of thousands of GPUs) requires explicit US government approval through diplomatic and regulatory channels.

What GPUs Are Specifically Restricted

The BIS control list targets chips above specific Total Processing Performance (TPP) thresholds. Restricted hardware for Tier-2 exports above compute caps includes:

  • NVIDIA A100 (both 40GB and 80GB variants)
  • NVIDIA H100 (SXM and PCIe variants)
  • NVIDIA H200
  • NVIDIA B100/B200/GB200
  • AMD MI300X
  • Intel Gaudi 3 (at higher thresholds than NVIDIA equivalents)

Chips below the threshold that are currently freely exportable to Saudi Arabia: NVIDIA L40S, A10G, T4, and older professional GPU series. These are appropriate for inference workloads that don’t require H100-class memory bandwidth, but are not competitive for LLM training or high-throughput inference of 30B+ parameter models.

How Humain Secured Exceptions

Humain’s announcement of 18,000 NVIDIA H100 GPUs in May 2025, with a pathway to hundreds of thousands more chips, was not simply a commercial transaction — it was the result of multi-year diplomatic engagement between the Saudi government and the US executive branch.

The Diplomatic Pathway

Several specific actions facilitated US approval:

  1. Crown Prince Mohammed bin Salman’s personal engagement with the Trump administration: The AI cooperation framework was announced during a presidential state visit, elevating it from a commercial deal to a bilateral foreign policy commitment. This political framing gave the US executive branch more flexibility to enable exports than a purely commercial export license application would have generated.

  2. Saudi commitment to data security governance: Saudi Arabia agreed to data security requirements embedded in the cooperation framework, including commitments about not re-exporting chips to restricted countries and allowing end-use verification by US officials. These commitments gave BIS the assurance it requires to approve large-scale Tier-2 exports.

  3. Separation from Chinese technology supply chains: The US has been systematically concerned about Chinese access to advanced AI chips through indirect pathways. Saudi Arabia’s commitment to not integrate Huawei or other sanctioned Chinese companies into the AI infrastructure receiving US chips was an implicit condition of the cooperation framework.

  4. Commercial incentives for US companies: The framework included commitments to purchase US cloud services, US AI software, and to host US hyperscaler infrastructure in Saudi Arabia — giving US tech companies (NVIDIA, Google, Microsoft, Amazon) commercial incentives to support the export approvals.

What This Means for Non-Sovereign Entities

Humain’s diplomatic pathway is not replicable by private companies or non-PIF entities. A foreign startup building an AI data center in Saudi Arabia cannot access Crown Prince-level diplomatic channels. For non-sovereign Saudi entities and for international companies building Saudi operations, the standard VEU application and export license process applies, with all its timeline and uncertainty costs.

Impact on Saudi Sovereign AI Timeline

Export control constraints have already meaningfully delayed Saudi sovereign AI development:

2023–2024 pipeline: Multiple Saudi hyperscale AI compute projects were stalled waiting for export approvals. SDAIA’s ambition to have operational sovereign AI training clusters by late 2023 slipped to 2025 due to GPU procurement constraints combined with data center construction timelines.

Training vs. inference gap: Saudi entities were able to procure GPU quantities below the license threshold more easily, creating a pattern where inference capacity (lower chip counts, smaller transactions) was deployed faster than training capacity. This means Saudi AI applications have largely been built on inference of models trained abroad, rather than on domestically trained sovereign models — a gap Humain’s procurement is designed to close.

SDAIA and Allam: Allam (the 34B Arabic LLM) was trained using a combination of IBM’s compute resources and limited Saudi-resident compute. The training compute bottleneck constrained training data size and iteration velocity. With Humain’s GPU acquisitions now flowing, future Allam versions are expected to train on significantly more Saudi-resident compute.

US Government Concerns: Data Security and Military End-Use

The US government’s fundamental concern with Tier-2 GPU exports is not Saudi Arabia specifically — it is preventing advanced AI compute from enabling Chinese or adversary AI military capability through indirect channels. Saudi Arabia sits at the intersection of several US concerns:

Concern 1 — Chinese technology entanglement: Saudi Arabia has deep economic relationships with China (China is Saudi Arabia’s largest trading partner). Chinese technology companies (Huawei, ZTE, Alibaba Cloud) are active in Saudi Arabia. The US concern: could Huawei-built network infrastructure or Chinese cloud services create a pathway for advanced US AI chips to benefit Chinese AI development? Saudi Arabia’s diplomatic commitments in the 2025 framework directly address this concern.

Concern 2 — Military AI end-use: The Saudi military modernization program is significant. The US has imposed arms export restrictions on some Saudi purchases historically. BIS applies additional scrutiny to any AI chip export where there is a plausible military end-use pathway, even for ostensibly commercial buyers.

Concern 3 — Re-export risk: The most operationally concerning risk for BIS is chips shipped to Saudi Arabia being re-exported to Iran, China, or Russia. Saudi Arabia’s geography and its trade relationships create a non-trivial re-export pathway that BIS must manage with end-use verification requirements.

Saudi Responses to Export Control Constraints

Saudi Arabia has pursued a multi-track strategy to reduce long-term dependency on US export approvals:

Seeking Tier-1 Upgrade

Saudi Arabia has formally requested Tier-1 classification from the US government, which would eliminate per-transaction compute caps and substantially reduce export control friction. As of mid-2025, this request has not been granted. Tier-1 status typically requires deeper intelligence-sharing relationships and defense integration (formal treaty ally status) that Saudi Arabia does not currently have with the US. The US-Saudi MNNA (Major Non-NATO Ally) designation, previously discussed in Washington policy circles, would be a step toward Tier-1 eligibility.

MNNA Designation

Major Non-NATO Ally status was under active diplomatic discussion between Riyadh and Washington as part of the broader normalization framework. MNNA status does not automatically grant Tier-1 export control status but creates a pathway for preferential treatment and provides political cover for more flexible BIS licensing decisions. This designation would be a significant milestone for the Saudi AI compute program.

Alternative Silicon Strategies

In parallel with diplomatic efforts to secure US chip access, Saudi Arabia has invested in silicon diversity:

  • Huawei Ascend 910B investment: Saudi entities have evaluated (and in some cases piloted) Huawei’s AI chips for workloads that do not involve US-restricted activities. This is a hedge against future US policy tightening.
  • Groq LPU: Groq’s Language Processing Units offer extremely low-latency inference at competitive economics. Groq has established a commercial relationship with Saudi cloud entities.
  • SambaNova and Cerebras: Alternative AI chip architectures from US companies, subject to US export controls, but currently below the Tier-2 threshold restrictions for most configurations.

Domestic Chip Development

SDAIA and King Abdullah University of Science and Technology (KAUST) have launched exploratory programs in chip design, including participation in global academic chip design programs. Meaningful domestic Saudi chip production is a decade or more away — but the policy signal that Saudi Arabia intends to develop indigenous semiconductor capability reduces long-term dependency on US export approvals as a geopolitical lever.

What Happens If US-Saudi Relations Shift

The 2025 AI cooperation framework is politically, not legally, anchored. A change in US administration, a deterioration in bilateral relations (triggered by oil production decisions, regional security incidents, or Israeli-Palestinian diplomacy), or a recalibration of US export control policy could narrow or eliminate the preferential treatment Saudi AI has received.

Downside Scenarios

Scenario 1 — Policy tightening: If BIS determines that leakage of advanced AI chips through Tier-2 countries is occurring at scale, a regulatory ratchet (tighter compute thresholds, stricter VEU requirements) could restrict Saudi procurement without any bilateral diplomatic rupture.

Scenario 2 — Bilateral friction: If US-Saudi relations deteriorate over oil production (OPEC+ decisions that disadvantage US shale producers), Saudi weapons purchases from China or Russia, or regional security conflicts, the political layer supporting Humain’s preferential treatment could evaporate quickly.

Scenario 3 — Technology evolution makes controls less relevant: Counter-intuitively, if alternative silicon (Huawei, AMD, domestic designs) reaches competitive capability faster than US policy anticipated, the leverage of US export controls over Saudi AI development diminishes naturally.

Hedge Strategies for Operators

For companies building Saudi AI infrastructure that depends on US chips:

  1. Maintain a hardware-agnostic software stack where possible (ROCm compatibility alongside CUDA, ONNX runtime)
  2. Qualify alternative inference infrastructure (Groq, Intel Gaudi, AMD) in parallel with NVIDIA deployments
  3. Structure GPU procurement contracts with regulatory contingency clauses and staged payment structures
  4. Monitor US-Saudi diplomatic developments (OPEC+ decisions, US arms sales, normalization with Israel) as leading indicators of export control policy direction
  5. Engage US export control counsel with active BIS relationships for early warning on regulatory changes

The current window of US-Saudi AI cooperation is real and creating genuine infrastructure momentum. It is also contingent on political conditions that can change. Saudi Arabia’s sophisticated planners understand this, which is why the sovereign AI program’s architecture — hardware diversity, diplomatic engagement, alternative silicon investment, indigenous capability development — is structured to survive policy volatility.

Impact on Specific Saudi AI Use Cases

Export controls affect different Saudi AI use cases differently, and understanding those specific impacts helps organizations prioritize their investment and advocacy efforts.

Saudi Government AI Applications

Government ministries deploying AI for domestic public administration — NAFIS (employment), Etimad (procurement), Absher (civil services) — are primarily running inference workloads on relatively small cluster sizes. Most government AI deployments fall below the 1,700 H100-equivalent threshold in any single transaction. The export control impact on day-to-day government AI operations is therefore moderate — smaller GPU allocations can be procured without individual licenses, and the managed SDAIA API provides access to pre-approved compute.

Where government AI faces export control friction: training new foundation models for Arabic language, Arabic speech recognition, or Arabic vision models requires training-scale compute (10,000+ GPUs) that is firmly in the licensed zone. SDAIA’s Allam development required IBM’s compute infrastructure precisely because Saudi-resident training-scale compute was not available. Future Allam versions and successor models depend on Humain’s procurement pipeline delivering licensed compute at scale.

Saudi Oil and Gas AI Applications

Saudi Aramco’s AI applications — seismic interpretation, predictive maintenance, drilling optimization, reservoir simulation — represent some of the most compute-intensive industrial AI workloads in the world. Aramco Digital’s compute footprint is large enough that export controls are directly relevant. Aramco has navigated this as a large, internationally sophisticated entity with established US government relationships — its export control compliance program was mature before the AI Diffusion Rule took effect.

For smaller oil and gas AI vendors serving Aramco or other Saudi operators, the relevant question is whether your hardware sits in Aramco’s pre-approved compute environment or in your own Saudi-based deployment. Using Aramco Digital’s cloud infrastructure eliminates your direct BIS compliance burden (Aramco holds the authorizations) but creates vendor lock-in and data access constraints.

Academic and Research AI in Saudi Arabia

KAUST and KFUPM conduct AI research requiring serious compute for training experiments. KAUST maintains access to US-based compute through international academic collaborations (access to NSF National AI Research Resource, partnerships with MIT and Stanford). Saudi research institutions have been largely successful in maintaining training-scale compute access through academic collaboration channels, which are subject to different (and generally lighter) BIS treatment than commercial deployments.

The research pathway is the least constrained by export controls — and it is one reason Saudi Arabia has invested heavily in KAUST’s international research profile. AI researchers who can publish at NeurIPS with access to adequate compute are a strategic national asset, and their academic compute access is protected by diplomatic norms around international scientific collaboration that the US has historically honored even during periods of geopolitical tension.

The Long Game: Saudi AI Sovereignty Despite Controls

Saudi Arabia’s response to export controls reveals a sophisticated long-term strategy rather than simple dependence on US permission. The Kingdom is simultaneously:

  1. Maximizing access under current rules through Humain’s VEU process and diplomatic frameworks
  2. Diversifying supply chains through investment in Huawei Ascend, Groq, and other alternative silicon vendors
  3. Building domestic capability through KACST chip design programs and engineering education investment
  4. Participating in international AI governance through SDAIA’s engagement with OECD AI Policy Observatory and G20 AI principles, building relationships that give Saudi Arabia influence over future control frameworks

The result is a multi-decade AI sovereignty program designed to be resilient to any single source of supply chain disruption. For companies working within the Saudi AI ecosystem, understanding this strategic context explains why the Kingdom’s AI investments are structured the way they are — and why the opportunity is as large and durable as it appears to be.