In a nutshell: China's tech sector is racing to replace foreign AI chips with domestic alternatives as US export controls restrict access to advanced processors. Alibaba, emerging startups like MetaX, and established firms such as Cambricon Technologies lead the push, supported by Beijing's heavy state investment in AI and semiconductors.
China's leading tech companies have long relied on American chipmakers such as Nvidia, but US export restrictions and rising geopolitical tensions have accelerated efforts to develop domestic alternatives. Alibaba, once a top Nvidia customer, now joins local developers in meeting the growing demand for chips that can power emerging AI services.Industry experts stress that China still trails the United States in producing chips that match the cutting edge of American design and manufacturing. Chinese factories face persistent shortages of advanced fabrication equipment, as US policy limits the transfer of leading-edge technology.
Amid these limitations, Alibaba has developed a new AI chip capable of handling a wider range of inference tasks – the operations that let AI models interact with users in real time – rather than being limited to specialized functions. Currently in testing, the chip marks a move away from earlier processors focused on specific applications. Sources familiar with the development told The Wall Street Journal that a Chinese fabrication facility produces the chip, a departure from processors TSMC made before US export controls barred the company from supplying advanced AI chips to Chinese clients.
Shanghai startup MetaX recently launched a chip with higher memory capacity than Nvidia's H20 model. Washington briefly allowed the H20 back into the Chinese market before Beijing ordered companies to halt purchases over security concerns. Although MetaX's chip consumes more power, designers say it suits specific AI workloads, and the firm is preparing to scale production.
Beijing AI chip designer Cambricon Technologies reported strong quarterly growth, earning $247 million in revenue on robust orders for its Siyuan 590 chip. The company's rapid stock gains prompted management to warn investors about overheating risks. Despite a subsequent share price decline, Cambricon's market value remains above its pre-drop level, holding at more than $87 billion.
Beijing's drive to reduce foreign dependence extends beyond individual business initiatives. In January, the government announced an $8.4 billion fund to support AI development and supply chain independence. Huawei Technologies recently unveiled a powerful AI computing system built around 384 proprietary Ascend chips. Some assessments suggest the machine outperforms leading US hardware on select benchmarks, though it carries substantial energy costs.
"To achieve comparable computing results, we can combine chips – and there's no need to worry about the chip problem," Huawei founder Ren Zhengfei told the Communist Party's main newspaper in June.
Tension is growing within China over reliance on domestically produced chips. Many engineers favor Nvidia hardware for its robust ecosystem and established tools, while local options often lack compatibility or ease of use. However, Alibaba's new chip works with the Nvidia platform, making it easier for developers to transition existing software.
Public cloud providers have largely avoided bulk orders of Huawei's chips, even with official encouragement, because they see the company as both a supplier and a direct competitor in the cloud computing business. This cautious stance underscores the broader challenge for domestic chip adoption, showing that state support alone cannot overcome concerns about performance, compatibility, and competition.
Despite the surge in new projects and state investment, China still faces a critical weakness in training high-quality AI models. Advanced AI training relies on processors that Nvidia can no longer supply due to US controls. Alibaba's latest domestic chip focuses on AI inference – the process of running pre-trained models – but insiders say it is not optimized for the initial, energy-intensive training phase.
Chinese engineers working with homegrown chips report frequent technical issues, like overheating and system breakdowns during extended model training. Huawei declined to comment on such challenges when reached for further details.
Nevertheless, optimism persists. Earlier this summer, AI startup DeepSeek sparked a market rally by hinting on social media that its software innovations, combined with improving Chinese-made hardware, could advance domestic training capabilities.
"An entirely 'Made in China' AI stack may emerge in volume to meet both domestic demand and expansion into overseas markets sooner than most people think, credibly challenging Nvidia and the American AI stack both at home and abroad," noted Kevin Xu, head of AI investment firm Interconnected Capital.