Huawei’s goal is to become Nvidia, supervisory board chairman tells employees
Huawei Technologies said Thursday it would release two new artificial-intelligence chips next year — the Ascend 960DT in the first quarter and the 960PR in the third quarter — with both running ahead of their initial timelines, as part of a broader push to offer Chinese customers an alternative to Nvidia’s computing systems. Additional chips are planned for 2028 and 2029, according to the company.
The computing power Huawei’s planned chips can individually deliver still trails the best Nvidia chips available today, according to specifications released by the two companies.
Huawei supervisory board chairman Guo Ping told new employees recently, “Huawei’s goal is to become Nvidia,” according to an official company transcript.
The Chinese company is moving aggressively to keep up with Nvidia, with support from Beijing and other domestic companies, and is taking advantage of its expertise in telecommunications equipment.
Huawei has shipped more than 1,000 AI-computing systems — each bundling hundreds of chips to boost computing power — to more than 370 customers. The company also describes a separate effort involving massive clusters that bundle hundreds of thousands of AI chips to function as a single computer. In chaining many weaker chips together, the company aims to mimic cutting-edge models. Huawei works with domestic AI chip users including AI developer DeepSeek to gather feedback on its designs.
Huawei is pursuing three approaches to overcome U.S. barriers: developing alternative chip designs, bundling chips into massive clusters, and improving data transmission.
On chip design, Huawei is developing workarounds and new designs to squeeze more computing power out of less advanced fabrication equipment. Traditional fabrication relies on ultraprecision machinery to squeeze more circuits onto a single chip; Huawei is trying to improve computing efficiency through techniques such as layering circuits within a single chip.
On clustering, Huawei’s technology, called UnifiedBus, aims to establish a universal language so different chips, even across separate server racks, can communicate directly. In a traditional server, various processors and memory units operate like different countries speaking distinct languages, with data requiring translation between them before it can be processed; that translation can create bottlenecks.
Coordinating many AI chips uses substantial power. Huawei has rolled out a technology called near-packaged optics designed to save energy, using components to transmit data placed close to the chips to increase efficiency. Nvidia and Broadcom in their computing systems use something similar aimed at improving transmission called co-packaged optics.
Huawei has acknowledged that its chip-manufacturing capacity remains constrained and that these new techniques bring their own challenges. Bundling chips together requires more precise and reliable communications and highly sophisticated software, hurdles some data-center operators have warned are difficult to clear.
Huawei rotating chairman Eric Xu addressed the strategic reasoning on Thursday. “We can’t be at the mercy of others forever — one day they say they’ll sell you a chip, and the next day they say they won’t,” Xu said. “So even if the technology is a bit less advanced, at least we have something and aren’t living in constant uncertainty.”
Since the Biden administration, the U.S. has barred Chinese companies from buying Nvidia’s best AI chips and blocked China’s chip makers, including Huawei, from buying top-of-the-line equipment. Huawei is subject to further curbs because it has been blacklisted by the U.S. government since 2019.
According to Morgan Stanley, China’s AI chip market is projected to grow to $67 billion by 2030, with domestic companies taking an 86% share, compared with less than half in 2025.