China’s AI Ambitions: A Shift Away from US Technology
As the global artificial intelligence (AI) race intensifies, China is making significant strides in reducing its reliance on American technology. Two key players in this movement are DeepSeek and Huawei Technologies, both of which are spearheading efforts to develop homegrown solutions that can rival those of their US counterparts.
In late August, DeepSeek, a relatively young AI startup, unveiled an updated foundational model that caught the attention of investors around the world. The release sent shockwaves through the market, particularly for Nvidia, as the company’s shares dipped. This was not just because of the model’s capabilities but also due to DeepSeek’s shift towards using domestically produced chips. Analysts noted that this move signaled a growing confidence in China’s ability to create competitive AI infrastructure without depending on foreign technology.
Huawei, another major player, has also been at the forefront of this transformation. The tech giant recently showcased its latest Ascend chip series, highlighting hardware designed to deliver high-performance computing without the need for Nvidia processors. This announcement marked a significant moment for Huawei, as it was the first time the company had shared details about its chip roadmap since being placed on the US blacklist in 2019 over national security concerns.
The push for self-sufficiency is not just a reaction to external pressures; it is also driven by internal goals. Beijing has encouraged its tech giants to stop purchasing Nvidia’s China-specific chips, which were designed to comply with US export restrictions. As a result, companies like Huawei and DeepSeek have become symbols of China’s resilience and innovation in the face of technological challenges.
Huawei’s Ascend Chips: A New Era in AI Computing
At the heart of Huawei’s strategy is its ambitious blueprint for future Ascend chips. Deputy Chairman Eric Xu Zhijun described these chips as “the foundation of Huawei’s AI computing strategy.” According to the company, future generations of chips will double in compute power every year while supporting more data formats, enhancing usability, and increasing bandwidth.
Several models are set to be released in the coming years. The Ascend 950PR, designed for pre-fill and recommendation tasks, is expected to launch in the first quarter of next year. The Ascend 950DT, optimized for decoding and training, will follow in the fourth quarter. The Ascend 960 and 970 processors are scheduled for release in the fourth quarters of 2027 and 2028, respectively.
Alongside these new chips, Huawei introduced “the world’s most powerful” supernode computing clusters—Atlas 950 and Atlas 960 SuperPoDs, as well as Atlas 950 and Atlas 960 SuperClusters. These systems can aggregate up to a million Ascend neural processing units (NPUs), which are designed to accelerate AI tasks.
According to Kevin Xu, founder of Interconnected Capital, these innovations demonstrate how Huawei is leveraging its networking expertise to close the gap with industry leaders like Nvidia. However, despite these advancements, Huawei still lags behind in single-chip performance, according to Charlie Zheng, chief economist at Samoyed Cloud Technology Group.
DeepSeek’s Transition to Domestic Chips
DeepSeek has also taken steps to integrate domestic chips into its AI models. The Hangzhou-based start-up used more than 2,000 Nvidia chips to train its V3 foundation model, but its recent update, V3.1, was trained using a proprietary UE8M0 FP8 scaling method. This approach maximizes efficiency from available hardware and is tailored for next-generation home-grown chips.
Analysts believe that Huawei could be one of DeepSeek’s primary hardware suppliers. While the two companies have not confirmed whether Ascend chips have been used in V3.1, many industry insiders anticipate that Huawei will play a key role in future developments. Huawei’s Xu highlighted the use of Ascend chips in running DeepSeek’s R1 reasoning model, emphasizing the collaboration between the two companies.
Broader Implications for China’s AI Industry
The shift toward domestic chips is not limited to Huawei and DeepSeek. Major players in China’s AI computing sector are united in their goal of achieving tech self-sufficiency. In July, Huawei joined the Model-Chips Ecosystem Innovation Alliance, a group of Chinese semiconductor and AI companies working together to promote locally developed processors in AI projects.
Another notable player is Cambricon Technologies, an AI chipmaker based in Beijing. Founded in 2016, Cambricon has seen a surge in popularity, with its stock price rising nearly 500% over the past year. The company reported a staggering 4,348% year-on-year increase in first-half revenue to 2.88 billion yuan (US$404 million), fueled by continued market expansion and active support for AI applications.
China’s largest tech companies are also ramping up their efforts to adopt more domestic chips in their AI infrastructure. Tencent Holdings stated that its cloud computing unit had “fully adapted to mainstream domestic chips,” while Alibaba Group Holding’s semiconductor design unit, T-Head, developed a new application-specific AI chip said to be on par with Nvidia’s H20 processors.
Challenges Ahead
Despite these advancements, challenges remain. One of the key issues is the production capacity of China’s foundries, such as Semiconductor Manufacturing International Corporation (SMIC). These facilities are under US tech sanctions and may struggle to mass-produce Huawei’s Ascend chips at yields comparable to those of Taiwan Semiconductor Manufacturing Company (TSMC), which ceased collaboration with Huawei in 2020.
Samoyed’s Zheng estimated that China’s 7-nanometre production lines could supply about 600,000 to 700,000 AI chips in 2026. If Huawei’s million-card cluster comes to fruition, it would consume one-sixth of the annual production capacity, placing immense pressure on domestic fabrication plants.
Additionally, there are questions about whether Huawei’s “supernode” architecture can work seamlessly, allowing many chips to function smoothly as one computer. This is crucial for large-scale AI training, according to Interconnected Capital’s Xu.
Despite these challenges, Huawei remains committed to its vision. “Our chip technology is currently one or two generations behind [Nvidia], and we cannot predict how many generations behind we’ll be in the future,” Xu said in a recent interview. “We have no choice but to find another way out.”
