Nvidia H200 chips reach China in limited quantities as US tightens export controls

Jul 18, 2026

Nvidia has delivered initial batches of roughly 10,000 H200 AI chips to ByteDance and Tencent, marking the first shipments since US export approval in late 2025. However, these deliveries represent just 13% of the authorized quota. Beijing's National Development and Reform Commission acts as a strict gatekeeper, directing firms to deploy these chips in infrastructure-constrained Hong Kong to protect domestic chipmakers like Huawei. Meanwhile, severe computing shortages persist in China, forcing local AI firms to optimize software for domestic hardware while US lawmakers move to close cloud-access loopholes in Southeast Asia.

H200 chip deliveries to China

▪Nvidia denied a report by The Information claiming it plans to ship a language processing unit tailored for Chinese customers by the end of 2026, stating it has no such product on its roadmap.
▪ByteDance and Tencent each received approximately 10,000 Nvidia H200 artificial intelligence processors in August 2026, marking the first meaningful deliveries since President Trump approved their export in December 2025.
▪The delivery of 10,000 Nvidia H200 chips to ByteDance and Tencent represents only 13% of the 75,000-unit legal ceiling authorized for each approved Chinese buyer under United States export licenses.
▪Nvidia holds an inventory of approximately 500,000 H200 processors set aside primarily for Chinese customers, which had previously stalled under regulatory limits.

Beijing quota approval process

▪Deploying Nvidia H200 processors in Hong Kong faces logistical bottlenecks due to a local deficit of data center facilities driven by severe power supply limitations.
▪Chinese buyers of Nvidia H200 chips must clear a distinct review and authorization process managed by China's National Development and Reform Commission to receive their remaining licensed allocations.
▪Beijing is directing Chinese technology firms to deploy their acquired Nvidia H200 processors in Hong Kong rather than mainland China to protect and promote homegrown semiconductor alternatives.

Overseas cloud access loopholes

▪The proposed Remote Access Security Act, which passed the United States House of Representatives in January 2026, seeks to expand export controls to cover remote cloud-based access to critical hardware.
▪White House official Michael Kratsios accused Moonshot AI of using restricted Nvidia GB300 chips via a data center facility in Thailand shortly after Moonshot AI released its Kimi K3 model in July 2026.
▪Chinese firms including ByteDance, Alibaba, and Tencent have accessed restricted Nvidia chip compute power remotely via cloud providers in Southeast Asian nations, including Thailand and Malaysia.

Chinese domestic chip alternatives

▪Baidu executive vice president Dou Shen stated that Chinese buyers are increasingly seeking domestic AI chips because the supply of foreign AI chips is likely to remain constrained.
▪Alibaba's chip design unit T-Head open-sourced SAIL, the software stack for its Zhenwu AI chips, at the World AI Conference in Shanghai to lower migration barriers from Nvidia's proprietary CUDA ecosystem.
▪Huawei open-sourced its CANN software platform for Ascend AI processors in 2025, competing alongside Alibaba and Moore Threads to transition developers away from Nvidia's CUDA software stack.

Computing capacity shortage

▪China's average daily token calls exceeded 140 trillion in March 2026, representing a 1,000-fold increase from the beginning of 2024, according to the National Data Administration.
▪Severe computing capacity shortages in China have forced Moonshot AI to freeze new user sign-ups and prompted DeepSeek to raise its developer fees.
▪Chinese AI firms face acute compute constraints because complex tasks like coding still require Nvidia chips, whereas domestic hardware can reliably handle only lower-tier inference tasks.