US Imposes New Curbs on Nvidia AI Chips to China
The U.S. government has imposed new export curbs on Nvidia's H20 AI chips destined for China, a move that could have a significant financial impact. The restrictions are projected to put up to $55 billion in sales at risk. This action escalates trade tensions and is expected to create ripple effects across global technology supply chains.
- This is part of a recurring pattern where Nvidia has created specific, less powerful chips for the Chinese market to comply with evolving U.S. regulations; previous examples include the A800 and H800 chips, which were designed as alternatives to the more powerful A100 and H100. - The restricted H20 chip, while a significant downgrade in overall computing power compared to the top-tier H100, was engineered with more HBM3 memory (96GB vs. 80GB) and higher memory bandwidth (4.0 Tb/s vs. 3.4 Tb/s), making it paradoxically more than 20% faster for certain large language model (LLM) inference tasks. - The U.S. first established a comprehensive set of export controls on semiconductor technology to China in October 2022, aiming to restrict the country's access to high-performance chips that could be used for military modernization. - In response to the ongoing restrictions, China is accelerating its push for self-sufficiency in semiconductors, with companies like Huawei, whose Ascend 910B chip competes with Nvidia's A100, and SMIC, a leading domestic manufacturer, receiving significant government support. - Nvidia CEO Jensen Huang has stated that the U.S. export controls have backfired by providing Chinese companies with the "spirit, the energy and the government support to accelerate their development." - The Commerce Department's licensing policy has been dynamic, at times reversing bans on certain chips and moving to a "case-by-case review" for exports, reflecting a tense balance between national security concerns and economic interests. - The semiconductor supply chain is deeply globalized, with
