Huawei Technologies is accelerating its efforts to develop alternatives to Nvidia’s artificial intelligence (AI) computing chips as part of China’s broader strategy to reduce dependence on U.S. technology. The company announced on Thursday plans to launch two new AI chips in 2027, signaling a push to expand its AI hardware portfolio amid ongoing U.S. restrictions.
Huawei stated it will release the Ascend 960DT chip in the first quarter and the Ascend 960PR in the third quarter of next year, both ahead of previously expected schedules. Additional chips are also slated for release in 2028 and 2029. Despite these advances, the computing power of Huawei’s forthcoming chips continues to lag behind that of Nvidia’s top-tier products, based on publicly available performance specifications.
Since the Biden administration imposed restrictions barring Chinese firms from purchasing Nvidia’s most advanced AI chips, Huawei has faced significant challenges in accessing leading-edge technology. In addition to the chip purchase ban, U.S. authorities have blocked Chinese companies, including Huawei, from acquiring state-of-the-art manufacturing equipment. Huawei itself has been on a U.S. government blacklist since 2019.
Guo Ping, chairman of Huawei’s supervisory board, recently expressed the company's ambition to rival Nvidia in the computing and communications sector. Huawei has supplied over 1,000 AI-computing systems—integrating hundreds of AI chips—to more than 370 customers to date.
One of Huawei’s key strategies to compensate for less powerful individual chips involves bundling hundreds of AI chips into substantial clusters that operate collectively as a single computing unit. This approach is intended to emulate the performance of more advanced single processors.
Huawei is also focusing on improving data transmission efficiency within such multi-chip systems. Traditionally, servers rely on various processors and memory units that communicate inefficiently, creating bottlenecks when translating data between components. Huawei’s proprietary technology, dubbed UnifiedBus, seeks to create a universal communication protocol that enables chips, even across different server racks, to exchange data directly.
The increased power consumption involved in linking numerous AI chips prompted Huawei to develop "near-packaged optics," a technology that places data transmission components close to the chips to conserve energy. This approach bears resemblance to co-packaged optics technologies utilized by Nvidia and Broadcom, which similarly aim to enhance data transmission efficiency.
Despite these innovations, Huawei acknowledges that its manufacturing capacity remains limited. The company also notes that integrating large clusters of chips requires highly precise communication systems and sophisticated software, complexities that some data-center operators identify as significant obstacles to adoption.
To refine its products, Huawei collaborates with domestic AI developers such as DeepSeek to gather user feedback and improve its AI chip offerings. The company’s continued investments reflect China’s broader ambition to achieve technological self-sufficiency amid escalating U.S. export controls and geopolitical tensions.