As the U.S. deliberates on imposing stricter trade restrictions to prevent advanced chip technology from reaching China, American chip manufacturer Nvidia is reportedly developing a new version of its artificial intelligence chips to adhere to these regulations.
Berkshire Hathaway holds $234.6 billion in U.S. Treasury bills, surpassing the total held by the Federal Reserve.
According to unnamed sources cited by Reuters, Nvidia is working on a version of its latest Blackwell AI chips aimed at the Chinese market. The company plans to collaborate with local distributor Inspur to introduce and market the chip, provisionally named the “B20,” in China.
This B20 chip is anticipated to begin shipping in the second quarter of 2025, as per sources. Nvidia has not provided any comments on the report.
The company has three chips designed specifically to meet U.S. export regulations, including the H20, which has had its prices reduced due to declining sales as it competes with homegrown rival Huawei. However, H20 sales are reportedly increasing, with projections indicating that Nvidia could sell over one million of its H20 chips in China this year, amounting to approximately $12 billion, despite existing U.S. trade restrictions. This expected sales figure nearly doubles Huawei’s anticipated sales for its Ascend 910B chip, based on data from SemiAnalysis.
Additionally, Jeffries analysts have indicated that Nvidia’s H20 chips could face challenges under potential future U.S. trade regulations. With the U.S. set to conduct its annual review of semiconductor export controls in October, it is “highly likely” that the H20 will be prohibited from sale to China. This ban could be implemented in various forms, including a “product-specific ban,” a reduction in computing power limits, or restrictions on memory capacity.
Furthermore, analysts expressed that the U.S. may also broaden its export controls on chips sold to other countries in the region, like Malaysia, Indonesia, and Thailand, or target overseas Chinese firms, though such implementations would be more complex.