Major Chinese companies like e-commerce giant Alibaba and telecommunications titan Huawei Technologies have begun competing to develop their own artificial intelligence semiconductors.
Beijing has ordered that Chinese development of semiconductors, and AI chips in particular, become less reliant on the U.S. amid ongoing trade tensions with Washington.
But even as it develops proprietary technologies, Alibaba is working together with U.S.-based Nvidia on AI applications such as software for robots. The company appears to be mixing hardline and cooperative strategies in a bid to broaden the uses of AI.
At an annual Alibaba Cloud conference that began Wednesday in the city of Hangzhou in Zhejiang province, Alibaba Group CEO Eddie Wu said, “The pace of development in the AI industry and the demand for infrastructure far exceed our expectations.”
Alibaba in February announced a three-year plan to invest 380 billion yuan ($53.4 billion) in AI and cloud computing. Wu said the company would “further increase investment.”
Speaking at an event on Wednesday, Alibaba Group CEO Eddie Wu said the pace of development in the AI industry and demand for infrastructure “far exceed our expectations.” (Photo obtained by Nikkei)
He noted that global AI investments already exceed $400 billion annually, and are projected to surpass $4 trillion within five years. Wu pointed to AI foundation models as the next generation of operating systems, and likened the AI cloud to the next form of the computer, signaling an aggressive investment position to take on these challenges.
A highlight of the event was Alibaba’s latest AI server featuring its proprietary semiconductor. “It can be used for AI training, and it is currently used for such tasks as in-house development,” an Alibaba staff member said.
Alibaba’s AI development environment is notable for its easy migration from Nvidia’s CUDA platform for inference processes, said Toshiya Suzuki, principal consultant at research firm Omdia. “It looks like the Chinese government won’t run into any issues [with AI policy], even without [Nvidia’s] H20,” he added, referring to the U.S. company’s China-only AI chip.
Also on Wednesday, Alibaba unveiled its latest foundational model, Qwen3-Max. The company said its new model outperforms those of rivals like OpenAI in the U.S. Following the announcement, Alibaba’s Hong Kong-listed shares jumped 9% from their close the previous day, reaching their highest price in about four years.
Because of export restrictions imposed by the U.S. government, Nvidia supplies Chinese companies with its less powerful products. Even so, China has remained heavily dependent on U.S. players such as Nvidia and AMD for AI semiconductors.
But in April, Chinese President Xi Jinping declared that China must “respond to the rapid evolution of new-generation AI, leveraging the strengths of our new national system to achieve self-reliance.” He signaled that the government would back the development and production of AI semiconductors.
Beijing has also taken steps to pressure Washington. In July, the Chinese authorities said that Nvidia’s H20 chip had security vulnerabilities. In mid-September, Beijing banned major Chinese tech companies from buying Nvidia-made AI semiconductors, according to the Financial Times.
Alibaba is not alone in its pursuit of self-sufficiency.
Huawei announced at an event last Thursday that it would roll out four AI semiconductors by 2028. It is unusual for the company to publicly announce such a roadmap.
Designed by Huawei semiconductor subsidiary HiSilicon and manufactured by chipmakers including Semiconductor Manufacturing International Corporation (SMIC), the Ascend series will expand in 2026 to add two products with high-bandwidth memory, a component that had been considered difficult for China to produce domestically.
At the event, Eric Xu, the company’s current rotating chairman, said, “The enhancement of computing power [through high-performance semiconductors] has been, and always will be, the key to China’s AI.”
Although the performance of individual Huawei products may lag behind that of Nvidia’s top-of-the-line offerings, the Chinese company plans to compensate with server clusters using massive amounts of chips. Products slated for release in 2026 will use about 8,200 of the latest chips, while those put out in 2027 will use about 15,500. These server clusters will reportedly require sites of about 1,000 and 2,200 square meters, respectively.
Other companies are also lining up with the government’s policy. Internet giant Tencent has said it will adopt domestically produced AI semiconductors. “We will balance domestic and overseas products based on costs and other client needs,” said Tencent Executive Vice President Dowson Tong on Sept. 15.
Tencent provides cloud AI services to bolster corporate operations. It is expected to adopt China-made products for its servers, and is reportedly purchasing semiconductors from companies such as Huawei. Tong noted that the capabilities of Chinese products are improving, saying, “the number of products we can use is increasing.”
Nvidia’s share of China’s AI semiconductor market is projected to fall to 54% in 2025 from 66% in 2024, according to Chinese research firm Central China Securities, which used data from Bernstein in the U.S. Huawei’s share is expected to rise from 23% to 28%.
Excluding the big three of Nvidia, Huawei and AMD, the share of other firms is expected to rise to 14% in 2025 from 6% in 2024, with startups such as China’s Cambricon Technologies gaining traction.
Chinese companies still face restrictions on acquiring advanced design and production tools from U.S. and European businesses. While contract manufacturers like SMIC are boosting their capacity, there seem to be constraints on improving production efficiency, raising the possibility of industry consolidation and shakeouts.
