Nvidia, the world’s leading semiconductor company, is reportedly developing a new artificial intelligence (AI) chip tailored for the Chinese market, based on its flagship B200 product. The new chip, tentatively named B20, is set to enter mass production later this year, with shipments anticipated to begin in the second quarter of next year, according to sources cited by Reuters.
The American tech giant plans to collaborate with Inspur, one of its main distributors in China. However, Inspur has refuted claims of any ongoing business or cooperation related to the B20, labeling the Reuters report as inaccurate.
The design of the B20 aims to navigate around the United States’ stringent export controls. However, uncertainties linger about the chip’s ultimate release, given potential tightening of export restrictions by Washington.
A research note from Jefferies analysts, referenced by QZ.com, indicates a high likelihood that the H20 chip, another Nvidia product, may face a sales ban to China during the US Commerce Department’s annual review of export controls in October. The potential ban could be enforced through various measures, such as product-specific prohibitions, reduced computing power limits, or capped memory capacities.
The US might also extend its export controls to other Asian countries, including Malaysia, Indonesia, and Thailand, or to overseas Chinese firms, although these measures could be more challenging to implement.
This report follows a Wall Street Journal investigation on July 2 that uncovered an underground network smuggling A100 chips, banned for sale to China since October 2022, into the country. Middlemen have reportedly established research centers in Southeast Asian nations like Singapore to procure A100 chips from the US, then utilize Chinese students returning home to transport them covertly. In one instance, a 26-year-old student successfully smuggled six A100 chips in his luggage, earning $100 per chip.
The A100’s official price is around $10,000, but it fetches about $22,500 on the Chinese gray market, where buyers forego Nvidia’s warranty and maintenance services.
A History of Export Restrictions
In October 2022, the US Commerce Department’s Bureau of Industry and Security (BIS) prohibited exports of the A100 and H100 chips to China. A year later, the ban extended to the A800, H200, and other AI chips. In response, Nvidia introduced the H20, L20, and L2 chips for the Chinese market.
The H20, often described as a “castrated version” of the H100, offers about 15% of the H100’s performance. Despite this, it remains in high demand from major Chinese tech giants like Alibaba, Tencent, Baidu, and ByteDance. The Financial Times projects that Nvidia could sell up to one million H20 units in China this year, equating to approximately $12 billion in revenue.
Shanghai-based financial news site Cailianpress.com reported that the first shipments of H20 chips arrived in China in May and June, with leading Chinese tech firms currently testing them. Nvidia’s Compute Unified Device Architecture (CUDA) platform, which enhances GPU acceleration capabilities, continues to attract many Chinese users.
Li Yali, a columnist for Guancha.cn, recently questioned the future availability of Nvidia’s H20 chips in China amid ongoing geopolitical tensions. She emphasized the H20’s superior AI training and reasoning capabilities, attributed to its higher high-bandwidth memory (HBM) capacity compared to both the H100 and domestically produced AI chips.
As Nvidia navigates these challenges, the tech industry watches closely, anticipating how US export controls will shape the global AI chip market landscape.
Related topics:
Intelligent Building Management System Software
Intelligent Data Extraction: Revolutionizing Data Processing in the AI Era
Condition Monitoring and Control for Intelligent Manufacturing