Nvidia为中国市场准备新旗舰人工智能芯片版本

Nvidia plans to collaborate with Inspur, a major distributor partner in China, for the launch and distribution of the chip, tentatively named the "B20," according to two sources. Shipments of the "B20" are expected to commence in the second quarter of 2025.

The AI chip giant introduced its "Blackwell" chip series in March, which is scheduled for mass production later this year. These new processors feature two silicon squares, each the size of the company's previous models. Within this series, the B200 chip is touted to be 30 times faster than its predecessor in certain tasks, such as responding to chatbot queries.

Nvidia CEO Jensen Huang indicated in March that AI chips based on the Blackwell architecture are expected to begin deliveries in the second quarter, with scaled deliveries in the third quarter. By the fourth quarter, data centers will be able to run on these new products, which are anticipated to generate substantial revenue for Nvidia this year. Huang also highlighted a continued strong demand for AI computing power, with growing needs for the previous Hopper architecture products.

In 2023, Washington tightened its controls on the export of advanced semiconductors to China, aiming to prevent technological advancements in supercomputing that could benefit China's military. In response, Nvidia has developed three chips specifically tailored for the Chinese market.

The stricter U.S. export controls have enabled Chinese tech giant Huawei and startups like Tencent-backed Enflame to make significant progress in the domestic market for advanced AI processors. A version of Nvidia's Blackwell series chip for the Chinese market would strengthen the U.S. firm's position against these competitors.

Due to U.S. sanctions, China accounted for approximately 17% of Nvidia's revenue in the fiscal year ending in January, down from 26% two years earlier. Nvidia's most advanced chip for the Chinese market, the H20, initially experienced a slow start when deliveries began this year. Reuters reported in May that Nvidia priced it below a competing chip from Huawei. However, sales of the H20 have recently seen rapid growth, according to two of the sources.

Nvidia is projected to sell over one million H20 chips in China this year for more than $12 billion, based on estimates from research group SemiAnalysis.

Earlier reports noted that due to higher memory capacity, the H20 chip is more costly to manufacture than the H100, resulting in a price that is about half of the H100, with Nvidia accepting a compromise on profit margins. Renowned investment bank Morgan Stanley has also highlighted the strong demand in China, emphasizing that Nvidia cannot afford to forgo the revenue from H20 chip sales.

Huang has assured that Nvidia’s special version of AI chips will fully comply with U.S. export control requirements while aiming to optimize Nvidia’s presence and business in China.

Expectations remain high that the U.S. will continue to enforce stringent semiconductor-related export controls. The U.S. is reportedly urging the Netherlands and Japan to further limit the export of chipmaking equipment to China. Additionally, the Biden administration has preliminary plans to implement regulations around the most advanced AI models, which form the core software of AI systems like ChatGPT.

Last week, chip stocks globally declined after Bloomberg News reported that the Biden administration was considering a measure called the foreign direct product rule. This rule would enable the U.S. to halt the sale of a product if it was manufactured using American technology.