In recent years, the AI chip market has become a technological battlefield, with two giants, Nvidia and Huawei, vying for supremacy, particularly amid U.S. trade restrictions. While Nvidia has dominated the global market with its advanced chips, Huawei is making rapid strides in China, launching new AI chips like the Ascend 910C, 910D, and 920 to fill the gap created by bans on Nvidia products. This article examines the competition between the two companies, focusing on Huawei’s new chips and their implications.
The Context of the Competition
Nvidia has been the undisputed leader in the AI chip market, with products like the H100 and B200 widely used for training and running AI models. However, U.S. export restrictions on Huawei, which prohibit the sale of Nvidia’s most advanced chips (including the H100 and B200) to China, have created an opportunity for Huawei. In 2022, the U.S. banned the sale of the H100 to China before its launch, and recently, in April 2025, the Trump administration imposed export licenses on Nvidia’s H20 chip, designed specifically for the Chinese market. These measures have significantly impacted Nvidia’s revenue, with the company reporting a $5.5 billion loss in inventory and contracts in the first quarter due to the restrictions.
Amid these constraints, Huawei has intensified its efforts to develop AI chips that compete with Nvidia’s offerings. The Chinese company recently launched the Ascend 910C and 920 chips, with the Ascend 910D in testing, aiming to directly challenge Nvidia’s H100. These launches come at a critical time as China accelerates its push for technological self-sufficiency.
Huawei’s New Chips: Ascend 910C, 910D, and 920
Ascend 910C
Huawei began mass deliveries of the Ascend 910C chip to Chinese customers in May 2025, according to Reuters. This chip, which combines two 910B processors into a single package, delivers approximately 60% of the inference performance of Nvidia’s H100. While not a major technological breakthrough, the Ascend 910C has become a viable alternative for Chinese companies unable to access Nvidia’s H20 chip. Analysts estimate Huawei could sell over 70,000 units, worth around $2 billion, due to high demand.
Ascend 910D
Currently in testing, the Ascend 910D is considered Huawei’s most advanced AI chip. According to the Wall Street Journal, Huawei hopes this chip will surpass the performance of the H100, with initial samples set to be delivered in May 2025. The chip is being tested by Chinese tech companies to assess its technical feasibility but is still in early development and requires further testing for commercial use.
Ascend 920
Launched at a Huawei partner conference in April 2025, the Ascend 920 is manufactured using a 6 nm process and is estimated to deliver over 900 teraflops (BF16 performance) per board, with a memory bandwidth of 4 Tbps. This chip is positioned as a direct competitor to Nvidia’s H20, offering 30-40% better performance than its predecessor. Mass production is scheduled for the second half of 2025. However, Huawei has denied rumors that it officially unveiled the chip at the April conference.
Additionally, Huawei introduced the AI CloudMatrix 384 Supernode architecture, based on the Ascend 910C chips, which rivals Nvidia’s NVL72 system, offering comparable performance for AI data centers.
Challenges for Huawei
Despite its progress, Huawei faces several hurdles. Its chips, while competitive, are considered a generation behind Nvidia’s latest products, such as the Blackwell series. Huawei’s manufacturing process, primarily handled by SMIC (Semiconductor Manufacturing International Co.), relies on 5 nm and 6 nm nodes, while Nvidia and TSMC are advancing toward 2 nm nodes.
Moreover, Huawei struggles with low production yields, with reports indicating that four out of five manufacturing processes at SMIC are defective. This limits Huawei’s ability to meet demand. Additionally, Huawei’s software ecosystem, CANN, introduced in 2018, is less mature than Nvidia’s CUDA, making Ascend chips less user-friendly.
U.S. restrictions on access to advanced manufacturing equipment and high-bandwidth memory (HBM) pose another challenge. Huawei has been accused of obtaining chips from TSMC through intermediaries, prompting an investigation by the U.S. Department of Commerce. However, Huawei has denied these allegations, and TSMC confirmed it has complied with regulations and ceased supplying Huawei since 2020.
Impact on Nvidia
Export restrictions have significantly affected Nvidia, which has lost a substantial share of the Chinese market, estimated at 13-33% of its revenue. JPMorgan analysts estimate that the H20 chip ban could cost Nvidia up to $16 billion in fiscal year 2025. In response, Nvidia is developing new chips, such as the B20, to comply with export regulations, but their approval for China remains uncertain.
On the other hand, the bans have accelerated China’s push for self-sufficiency, with Huawei emerging as the primary beneficiary. Analysts believe that if Huawei can scale production and improve chip performance, Nvidia could quickly lose market share in China.
Conclusion
The rivalry between Nvidia and Huawei in the AI chip market reflects geopolitical tensions and the race for technological supremacy. Huawei has made remarkable progress, launching chips like the Ascend 910C, 910D, and 920, which provide viable alternatives to Nvidia’s products in the Chinese market. However, challenges related to production, software ecosystem maturity, and export restrictions limit Huawei’s ability to overtake Nvidia in the short term.
Meanwhile, Nvidia faces financial losses and growing competition from Huawei, which is capitalizing on the void created by U.S. bans. While Nvidia remains the global leader, Huawei’s rise signals a shift in the AI market dynamics, with profound implications for the future of the semiconductor industry.
