Nvidia, a leading chip designer, announced on Tuesday that new U.S. export restrictions will inhibit the sales of two high-end artificial intelligence (AI) chips tailored for the Chinese market.
Additionally, one of Nvidia’s premier gaming chips will be impacted by the news regulations. The move aims to curb the transfer of cutting-edge U.S. technologies to China, particularly those potentially enhancing the country’s military capabilities.
Nvidia detailed the consequences of these restrictions, revealing that two modified advanced AI chips, the A800 and H800, specifically designed for compliance with prior export regulations in China, will now be barred for sale under the new rules. Furthermore, the restrictions will extend to a high-end gaming chip and the L40S chip, introduced in August.
Industry analysts predict that the regulations will not only affect Nvidia but will also impact competing chips produced by Advanced Micro Devices (AMD) and Intel, both are based in the US.
Intel’s Gaudi 2 chip, which entered the Chinese market in July, is expected to face a ban. AMD’s current AI chip, the MI250, and the upcoming MI300 chip are also likely to be subjected to the tightened controls, Reuters news report said.
Nvidia, Intel, and AMD have yet to provide detailed public comments beyond acknowledging the impact of the new restrictions. Nvidia, in its filing, expressed concern that these restrictions may compel the company to relocate certain business operations away from countries affected by U.S. export curbs. The chip manufacturer emphasized that these regulations could hinder the timely development of specific products, support existing customers, and supply customers outside the impacted regions.
How US Strengthens AI Chip Export Restrictions
The United States has announced a comprehensive update to export restrictions concerning advanced artificial intelligence (AI) technology, targeting key components essential for the development of AI applications, including large language models such as ChatGPT.
The primary objective is to prevent China and other countries from acquiring cutting-edge chips vital for AI technologies, amid concerns about potential military applications, Reuters news report said.
A closer examination of the updated regulations reveals several critical adjustments:
Expanded Scope of Controlled Chips: The original rules implemented last year assessed chips based on both computing power and communication speed. The recent update, however, primarily focuses on computing power alone, broadening the range of chips subject to restrictions, including recent creations by top AI chip supplier Nvidia (NVDA.O) for the Chinese market.
Notification Requirements: The new regulations mandate that chip companies notify U.S. authorities when selling chips that slightly fall below the restriction limits. This provision enables officials to track whether these chips are being utilized in significant quantities for AI-related work.
Addressing ‘Chiplet’ Loopholes: To counter potential workarounds, the regulations address emerging “chiplet” technology, wherein smaller sections of a chip are combined to form a complete chip. The new rules impose a “performance density” limitation to prevent such loopholes.
Blacklisting China’s Nvidia Challengers: In anticipation of a potential ban on U.S. AI chips in China, U.S. officials have added leading Chinese chip startups, Biren and Moore Threads, to a trade blacklist. Both companies were considered primary candidates to produce domestic alternatives to Nvidia’s chips. This move restricts their access to chip manufacturers that utilize U.S. equipment.
Identification of ‘Red Flags’ for Chip Factories: The U.S. identified specific characteristics in chips, such as containing 50 billion or more transistors and utilizing high bandwidth memory, as “red flags” for exporters to consider when evaluating the need for export licenses to ship goods to China. These indicators aim to help chip factories spot potential efforts to circumvent the rules.
Expanded License Requirements: The updated rules extend license requirements for exporting advanced chips from China and Macau to all 22 countries under U.S. arms control embargoes. Additionally, controls are now extended to any company with an ultimate parent company headquartered in these countries to prevent offshore subsidiaries from acquiring banned chips.
These stringent measures underscore the U.S. government’s commitment to enforcing a robust regulatory framework to control the export of advanced AI technology and ensure it does not compromise national security interests.