The Pre-2024 Reality
Before 2024, the majority of large Saudi data center projects involved Chinese partnership in some form — Huawei equipment in commercial data centers, Chinese contractors building infrastructure, Chinese capital co-investing in projects, and Chinese cloud providers (primarily Tencent Cloud and Alibaba Cloud) operating regional capacity. The Saudi-China technology relationship reflected the broader pre-Trump-administration dynamic: when the US restricted access to advanced technology, China stepped in with capable alternatives at competitive terms.
The relationship had deep structural roots. China is Saudi Arabia’s largest trading partner by volume, Huawei held significant Saudi telecom contracts across the 5G buildout despite sustained US pressure to exclude it, and the Kingdom’s stated technology posture was explicitly multi-polar — engagement with both American and Chinese ecosystems rather than alignment with either. Specific examples are visible in the historical data. Center3, the major Saudi colocation and connectivity provider, operated significant Chinese-equipment-based infrastructure. Several of the 22 active Saudi data centers as of end-2023 used Huawei networking and server systems. The shift toward 62 active and pipeline data centers by 2030 was originally projected with substantial Chinese supplier participation.
The multi-polar posture was not naivete; it was leverage. Saudi Arabia understood that the credible option of Chinese suppliers improved its negotiating position with American ones. Through 2024, the Biden administration’s export-control architecture validated the hedge: the AI Diffusion Rule formalized in January 2025 would have placed Saudi Arabia in a Tier 2 classification with sharp caps on advanced GPU access — treating the Kingdom as a diversion risk rather than a partner. Under that framework, Chinese equipment was not just a bargaining chip but a plausible primary path. Had the Diffusion Rule survived, the Saudi compute buildout would likely have proceeded on a mixed stack with Huawei Ascend playing a material role.
The 2025 Reversal
Beginning in early 2025 and accelerating through November 2025, the Saudi-Chinese technology relationship in AI infrastructure was systematically reduced. The driver was the US export-control framework: when the Trump administration approved Saudi Arabia for 35,000 NVIDIA GB300 systems in November 2025, the approval included explicit prohibition of Chinese-manufactured equipment in any approved AI facility. The trade-off was structural — Chinese equipment ban in exchange for American chip access.
The reversal was embedded in a much larger alignment. The November 2025 approval came during Crown Prince Mohammed bin Salman’s Washington visit, alongside Saudi Arabia’s designation as a major non-NATO ally and a Saudi investment pledge to the United States of up to $1 trillion. The Trump administration had replaced the Biden-era containment logic with what officials describe as “strategic diffusion” — pushing American silicon into allied jurisdictions specifically to forestall Huawei Ascend penetration. Within that framework, the Chinese-equipment exclusion is not an ancillary condition; it is the entire point. American chips flow precisely so that Chinese chips do not.
Saudi Arabia accepted the trade-off. The policy decision was not without internal debate — Chinese capacity is real, Chinese equipment is competitive, and the Saudi-China relationship has broader dimensions beyond AI infrastructure — but the calculation favored alignment with the American AI alliance over technical agnosticism. Notably, the Saudis drew a line inside the negotiation: compliance terms cover storage at pre-approved deployment sites, personnel access authorization, resale and relocation restrictions, and BIS reporting and audit mechanisms, but Saudi negotiators reportedly rejected anything resembling kill-switches or remote-disable capabilities invocable by US authorities. The formula was oversight yes, control no. The Kingdom traded its Chinese supplier option for American frontier access, but it did not trade away operational sovereignty over the systems it bought.
What Got Replaced
Three categories of pre-2024 Chinese involvement were replaced. First, networking and server hardware: Huawei equipment was substituted with Cisco (through the AMD-Cisco-Humain joint venture, which commits 1 GW of AI infrastructure over five years) and other US and Western vendors. Second, AI accelerator silicon: the path that would have run through Huawei Ascend was redirected to NVIDIA (18,000 GB300 systems initially, a 600,000-GPU pipeline over three years), AMD (MI-series accelerators through the Cisco JV), Qualcomm (200 MW of AI200/AI250 inference capacity), Groq (the LPU inference cluster with Aramco Digital), and Intel in an exploratory role. Third, cloud regional presence: while Tencent Cloud’s announced regional cloud is still proceeding, its scope is constrained and its market position is dominated by AWS ($5.3 billion Saudi region), Google Cloud ($10 billion Dammam hub), and Microsoft (Saudi region in Q4 2026).
The replacement happened across procurement cycles 2025-2026. Existing Chinese-equipment installations are being phased out as they reach end-of-life. New deployments default to US and Western vendors. The transition is operationally complex — replacing infrastructure mid-life is expensive — but strategically straightforward: the policy decision is settled, and every approved AI facility is contractually bound to it.
There is an important nuance in how the substitution preserved the Kingdom’s diversification instinct. Saudi Arabia did not exchange a multi-vendor Chinese-inclusive stack for an NVIDIA monoculture; it rebuilt the multi-vendor architecture entirely inside the American alliance. NVIDIA dominates training, AMD competes on price-performance, Qualcomm handles rack-density inference, Groq optimizes sequential-token throughput, SambaNova serves SDAIA’s specialized training workloads. The hedge against single-vendor dependency survived the pivot — it simply no longer crosses the US-China divide.
Inside the Approved-Facility Perimeter
The operative unit of the ban is the approved AI facility — the pre-approved deployment sites where export-cleared NVIDIA systems may be stored and operated under the November 2025 framework. In practice, that perimeter encloses essentially all of the Kingdom’s frontier compute: Humain’s Riyadh campus (200 MW Phase 1, absorbing a significant portion of the initial 18,000 GB300 allocation), the planned Dammam campus (300 MW), the Hexagon government data center (480 MW, hosting SDAIA’s sovereign AI factory of up to 5,000 Blackwell GPUs), and the Humain-xAI joint venture facility (500 MW, running Blackwell GPUs procured through Humain’s NVIDIA master agreement). Every facility on that list is new construction, which makes the Chinese-equipment exclusion cheap to enforce: nothing needs to be ripped out of a greenfield site. The ban binds hardest not on the frontier fleet but on the legacy estate — the pre-2024 colocation base where Huawei networking and server systems were the installed norm.
That is where the phase-out economics play out. The 22 active data centers of end-2023 were built for enterprise workloads at conventional rack densities; the AI-era buildout toward 62 sites by 2030 involves fundamentally different power and cooling engineering. Much of the legacy Chinese-equipped capacity would have required substantial retrofit for AI-adjacent workloads regardless of policy, which means the equipment transition rides on top of an infrastructure transition that was happening anyway. The ban accelerated and directionalized a refresh cycle rather than forcing a purely political write-off — one reason the Saudi side could absorb the condition without visible commercial disruption.
The compliance architecture also changes who operates the market. Under the BIS framework, deployment-site reporting and audit access make every approved facility a jointly legible asset — visible to Saudi operators and US regulators simultaneously. Vendors with existing US compliance machinery (Cisco on networking, the American chip suppliers, the hyperscalers) fit natively into that regime; Chinese vendors cannot participate even as subcontractors within the perimeter. Procurement, in other words, is no longer a purely commercial decision anywhere near the AI stack. Every purchase order inside an approved facility is simultaneously a compliance event, and the vendor list is the enforcement mechanism.
What Remains
Some Chinese commercial presence remains. Tencent Cloud’s $150 million Middle East cloud region, announced at LEAP 2025, is proceeding — Saudi Arabia did not explicitly exclude Chinese cloud providers from operating regions, only from operating in approved AI facilities. Tencent’s Saudi rationale was never primarily an AI infrastructure play: it is anchored in gaming and entertainment, where Saudi Arabia ranks among the world’s top ten markets for gaming engagement, roughly 70% of the population is under 35, and Tencent titles command large user bases requiring low-latency edge infrastructure. Alibaba Cloud maintains a Saudi presence. Various smaller Chinese technology partnerships in non-AI domains continue, and Huawei’s legacy telecom footprint persists in the 5G layer. The Saudi-China economic relationship in energy, trade, and broader investment remains substantial.
The export-control architecture produces a structural asymmetry for the remaining Chinese players. If Tencent Cloud deploys NVIDIA GPUs in its Saudi infrastructure, those GPUs remain subject to US export jurisdiction regardless of the operator’s nationality — Chinese cloud providers in the Kingdom gain no exemption from BIS rules. And the Chinese domestic alternatives (Cambricon, Biren, and the broader non-NVIDIA accelerator field) currently lag NVIDIA’s Blackwell generation significantly for AI training workloads. The practical result: Chinese cloud presence in Saudi Arabia is confined to workload categories — gaming, CDN, enterprise applications — where frontier accelerators are not the binding input. Sensitive government AI, defense-related compute, and anything touching the approved-facility perimeter is structurally out of reach.
But in the AI-specific stack — chips, data centers, models, frontier infrastructure — the alignment is now decisively American. The pivot is operationally complete; what remains is the trailing operational work of phasing out legacy Chinese-equipment installations. The residual Chinese footprint is best understood as deliberately preserved optionality rather than incomplete execution: Riyadh kept every relationship that does not conflict with the approved-facility perimeter, and surrendered only those that did. A state that wanted a clean break would have excluded Tencent’s cloud region entirely; a state preserving a hedge lets it proceed at $150 million scale in workload categories the export framework does not touch.
Strategic Implications
The pivot reflects a deeper strategic reality. As of 2025, the United States offers Saudi Arabia frontier AI capability that China cannot match: Blackwell-class GPUs, the leading foundation models, the leading cloud platforms. The American chip ecosystem, even with export controls, provides better technology than Huawei Ascend at this generational frontier. Aligning with the American stack maximizes Saudi capability; aligning with the Chinese stack would have meant accepting trailing-edge capability at the precise moment the Kingdom committed $77 billion to building at the frontier.
The pivot also repriced Saudi Arabia’s position in the regional competition. The UAE’s parallel buildout — G42, restructured around Microsoft’s equity position and the Stargate program — had already chosen the American side; a Saudi hedge toward China would have handed Abu Dhabi the uncontested claim to being Washington’s Gulf AI partner. By accepting the Chinese-equipment ban, Riyadh neutralized that differentiation and converted the US relationship into the anchor of its own program: the major non-NATO ally designation, the GB300 volumes, and the xAI, Google Cloud, AWS, and Microsoft partnership stack all followed within the same alignment window. The precedent now cascades outward — the Saudi terms are the template being studied for the UAE’s next tranche, Egypt, Indonesia, and every jurisdiction where the realistic choice is American or Chinese rather than American or domestic.
For Beijing, the loss is larger than one market. Pre-2024 Saudi Arabia was the showcase for Chinese digital infrastructure exports at sovereign scale — proof that Huawei-built, Chinese-financed capacity could anchor a G20 economy’s digital transformation. Post-2025 Saudi Arabia is the showcase for the opposite proposition: that when frontier capability is on the table, even a deliberately multi-polar state chooses the American stack and accepts exclusionary conditions to get it. The demonstration effect operates on every government watching from the sidelines of the US-China technology competition.
Reversibility and What to Watch
The pivot is reversible in principle. If Huawei Ascend or other Chinese alternatives close the capability gap, and if US export terms tighten in ways that compromise Saudi sovereignty, the pivot could reverse. The ban on Chinese equipment exists today because Saudi Arabia chose American chips; it could be revisited if the cost of choosing American becomes too high. Saudi Arabia retains the underlying relationships — trade, energy, the Huawei telecom layer, the Tencent and Alibaba commercial presence — that would make a re-hedge operationally feasible, which is precisely why the Kingdom preserved them.
But reversal requires both technical convergence and policy divergence — Chinese silicon reaching frontier parity and American terms deteriorating simultaneously. Both are possible; neither is imminent. For the foreseeable horizon (2026-2028), the Saudi-American alignment in AI infrastructure is durable, held in place by sunk investment on both sides: American vendors have committed tens of billions to Saudi infrastructure, and Saudi Arabia has committed its compute program to American silicon under conditions it negotiated hard to make tolerable.
The indicators to watch are specific. First, the Ascend trajectory: each Huawei accelerator generation narrows or fails to narrow the gap with NVIDIA’s current architecture, and the gap size is the price of the American alignment. Second, US policy continuity: any move by a future administration toward kill-switch-style conditions or approval revocations would test the oversight-not-control bargain that made the pivot acceptable in Riyadh. Third, the treatment of the remaining Chinese presence: if Tencent Cloud’s Saudi scope expands into AI-adjacent workloads without objection, the ban’s perimeter is softening; if it stays confined to gaming and enterprise cloud, the perimeter is holding. The pivot was the single most consequential supplier realignment in the global AI buildout — and its durability is now a leading indicator for how the entire US-led chip alliance holds together.