Chinese technology companies have purchased over one million Nvidia gaming and workstation chips this year as they grapple with a shortage of advanced AI processors domestically. Despite U.S. export restrictions imposed four years ago on the sale of cutting-edge AI chips to China—and Beijing’s directive two years ago for tech firms to reduce reliance on Nvidia—Nvidia’s products remain widely used within China’s technology sector.

China has made artificial intelligence a national priority, with companies such as Huawei, Alibaba, Baidu, and Cambricon more than doubling domestic AI chip production in 2026. However, demand for AI computing power continues to outstrip supply. Industry insiders estimate that domestic production, supplemented by about one million chips made overseas, could reach around 3.5 million units this year. Yet demand is believed to exceed 5 million chips, creating a significant shortfall.

This gap poses challenges for China’s rapid expansion of AI infrastructure, as Beijing aims to advance self-sufficiency and reduce dependence on foreign technology amid competition with the United States. According to Gary Yu, a Morgan Stanley analyst based in Hong Kong, China’s expansion of AI computing capacity has been limited more by the availability of high-end hardware than by market demand.

To mitigate the shortfall, Chinese companies have turned to gaming and workstation GPUs from Nvidia that conform to U.S. export controls but were not originally intended for intensive AI use. Models such as the RTX Pro 6000D, RTX Pro 5000, and RTX 5090 offer lower performance compared with flagship Chinese processors like Huawei’s Ascend 910C but remain useful for simpler AI tasks, including early stages of AI inference.

One factor sustaining Nvidia’s position is its Cuda software ecosystem, widely adopted by Chinese tech groups before efforts to transition to domestic alternatives like Huawei’s CANN programming environment, which is reportedly less user-friendly. Some firms, including DeepSeek and Huawei, are collaborating to improve hardware and software integration and accelerate the shift from Cuda to local platforms.

The rapid acceleration in demand for AI capabilities by major Chinese companies—Alibaba, ByteDance, and Tencent among them—has further intensified the shortage. These firms are collectively expected to double their capital expenditures on AI-related computing power to roughly 700 billion yuan ($104 billion) this year. Morgan Stanley projects total Chinese spending on IT infrastructure at $193 billion in 2026, representing about 21 percent of U.S. levels. Alibaba’s chief executive, Eddie Wu, has indicated the current compute deficit will likely persist until at least 2030.

In response, domestic firms are scaling production. Huawei aims to produce 1.6 million AI chips this year, with Alibaba and Baidu combined targeting about one million units, and Cambricon planning 300,000. Alibaba recently launched its most advanced domestic AI processor, the Zhenwu V900, while Huawei plans to debut its AI training chip, Ascend 960DT, in early 2027. To offset individual chip limitations, some companies are deploying advanced networking technologies to connect large numbers of processors, with Huawei setting sights on data centers housing up to one million chips.

Some Chinese players are still developing their own AI chips; ByteDance expects to test advanced prototypes by year’s end and aims for mass production by 2027. Semiconductor Manufacturing International Corp is ramping up fabrication to support this growth, but domestic chip output struggles to match surging demand from AI startups and established companies alike.

To further supplement domestic capacity, Chinese tech firms are increasingly leasing computing power from overseas data centers, particularly in Southeast Asia. Tencent has contracted with Oracle for 100,000 GPUs, while ByteDance and Alibaba are pursuing similar deals. These foreign facilities are typically equipped with Nvidia’s higher-end chips, such as the B300, used for AI training workloads beyond domestic chip capabilities.

The persistent discrepancy between supply and demand continues to underscore broader challenges not only for individual companies aiming to compete globally but also for China’s government seeking technological self-reliance. Meanwhile, Nvidia is advancing its product offerings with the upcoming RTX Pro 5500, expected to improve upon current models permitted under export restrictions. Industry sources say Chinese buyers have already placed orders exceeding two million units of this new chip, slated to begin shipping late this year.

Regulatory discussions are reportedly underway in China to assess whether demand for the RTX 5500 aligns with the country’s strategic goals, with insiders acknowledging that meeting AI computing needs will likely require continued reliance on Nvidia’s products. Representatives from leading Chinese firms including Huawei, Cambricon, Baidu, ByteDance, Alibaba, DeepSeek, and Zhipu have not publicly commented on these developments.