The transaction price of NVIDIA B300 AI servers in the Chinese mainland market has soared to about 7 million yuan per unit, or about 1 million U.S. dollars, nearly double the price of about 4 million yuan at the end of last year.As a reference, the same server equipped with eight B300 GPUs is priced at approximately US$550,000 in the US market, which is only a slight increase from US$500,000 at the end of last year. The price difference in the Chinese market is nearly double.
Soaring prices have prompted some cash-constrained AI companies to shift from purchasing to leasing. It is reported that,The price of a one-year lease contract for the B300 server has risen to 190,000 yuan per month, and there is still a queue to wait for resource release.
Supporting the market pricing are the hardware specifications of the B300. The B300 is equipped with 288GB HBM3e high-bandwidth memory and can provide 14 petaFLOPS computing power at FP4 precision. It is currently one of Nvidia's strongest chips that can be used for AI inference tasks.
Uncertainty over H200 exports to China is also pushing up B300 prices. Although the U.S. government has approved the export license of H200 to China, the two sides still have differences on the specific terms and have not been able to actually ship it so far. This impasse has made the B300 even more scarce on the market.
China's local AI chip manufacturers are gradually increasing their competitiveness. Huawei has formulated a clear product iteration roadmap and plans to launch four Ascend series chips in phases from 2026 to 2028, including the Ascend 950PR in the first quarter of 2026.
Currently, manufacturers such as Haiguang, Cambrian, and Alibaba Pingtouge are also simultaneously promoting the iteration of AI training and inference chips.
