Huawei, backed by Beijing, unveils AI vision to challenge Nvidia’s global grip

Huawei Technologies Co. has outlined an ambitious strategy aimed at challenging Nvidia Corp.'s dominant position in the artificial intelligence hardware sector. At the Huawei Connect conference in Shenzhen, rotating chairman Eric Xu presented a comprehensive three-year plan that seeks to leverage China's state-backed push for technological autonomy to reduce Nvidia's significant market lead. While Huawei acknowledged its silicon currently lacks the raw computational power and efficiency of Nvidia's offerings, the company is banking on its historical strengths, such as large-scale clustering, advanced networking capabilities, and strong government support. This unveiling marks a notable shift towards transparency for Huawei, typically known for its secretive approach to hardware strategies. The detailed presentation coincided with sensitive discussions between Washington and Beijing. During his presentation, Xu delivered a performance reminiscent of an Nvidia product launch, introducing the next generation of Ascend processors and upgraded "SuperPod" data center designs, which mirror Nvidia's terminology for expansive computing platforms. A key highlight was the introduction of Huawei's "UnifiedBus" interconnect protocol, aimed at connecting up to 15,488 Ascend chips in one system, significantly outpacing Nvidia's upcoming NVLink144. Huawei claims that its self-designed interconnect system enables chip-to-chip data transfers that could be 62 times faster than Nvidia's existing technology, a bold assertion that underscores the company's confidence in its local semiconductor ecosystem. Bernstein Research analysts noted that Huawei's willingness to disclose its AI strategies indicates it has secured reliable manufacturing capabilities within China, even after being cut off from Taiwan Semiconductor Manufacturing Co. (TSMC). The ambitious roadmap comes at a time when other major Chinese technology firms, including Alibaba Group and Baidu, are also making strides in AI chip design, marking a shift in the landscape that previously saw companies remain tight-lipped to avoid scrutiny from U.S. authorities. With U.S. chip policies now central to U.S.-China relations, Beijing has encouraged its tech companies to accelerate innovation and lessen reliance on foreign suppliers. Despite facing restrictions that have limited access to advanced manufacturing tools, Xu believes that Huawei can close the performance gap by creating large clusters of its processors. The company has developed a high-bandwidth memory architecture designed for rapid data exchange between chips. Looking ahead to 2028, Huawei anticipates its Ascend 970 chip could achieve interconnect speeds of 4 terabits per second, far surpassing Nvidia's current capabilities. However, significant challenges remain. Analysts from Jefferies have raised concerns about Huawei’s earlier plans for the Ascend 910D, which have reportedly stalled due to production issues, illustrating the complexities associated with restricted access to cutting-edge manufacturing technology. Performance comparisons also highlight the hurdles ahead, as Huawei's upcoming Ascend 950 is expected to deliver only about 6% of the computing power projected for Nvidia's next-generation superchip. Even as Huawei asserts that massive clusters could theoretically match the performance of Western rivals, doubts linger regarding whether such a "brute force" approach can effectively address the limitations imposed by existing process technology. The Chinese government has prioritized the semiconductor industry in its broader industrial strategy, with President Xi Jinping emphasizing the need for breakthroughs in critical sectors. As Huawei aims to position itself as a national frontrunner in the AI race, the company's latest announcements reflect both its determination and the support from Beijing to foster a self-sufficient technological future. However, the success of these initiatives will ultimately depend on Huawei's ability to scale production of its chips in the face of ongoing challenges.