CHINA WATCHThe Chinese Military’s Access to AI Chips

Published 8 July 2022

The Chinese military has made rapid progress in artificial intelligence. This progress largely depends on continued access to high-end semiconductors designed by American companies and produced in Taiwan and South Korea. The aggressive moves by the Trump and Biden administrations to limit technology exports to the Chinese military notwithstanding, China continues to order large quantities of American-designed advanced semiconductors from manufacturers in Taiwan and South Korea.

The Chinese military’s progress in artificial intelligence largely depends on continued access to high-end semiconductors. By analyzing thousands of purchasing records, a new report from the Center for Security and Emerging Technology (CSET) at Georgetown University offers a detailed look at how China’s military comes to access these devices.

The authors find that most computer chips ordered by Chinese military units are designed by American companies, and outline steps that the U.S. government could take to curtail their access.

Here is the report’s Executive Summary:

Over the last five years, the Chinese People’s Liberation Army (PLA) has made significant progress adopting artificial intelligence for combat and support functions. Chinese leaders broadly expect AI to usher in the “intelligentization” (智能化) of military affairs, characterized by ubiquitous sensor networks, more frequent machineon- machine engagements, and a faster tempo of operations.

But the PLA’s progress in AI and related technologies largely depends on continued access to a special class of semiconductors—AI chips—which are used to train advanced machine learning systems. By analyzing 24 public contracts awarded by PLA units and state-owned defense enterprises in 2020, this policy brief offers a limited but detailed look at how the Chinese military comes to access these devices.

Despite more aggressive efforts by the Trump and Biden administrations to limit technology exports to the Chinese military, the PLA is placing orders for AI chips designed by U.S. companies and manufactured in Taiwan and South Korea. The authors of this policy brief consider alternative approaches that the United States could take to curtail Chinese military access to AI chips—such as a limited crackdown on Chinese military suppliers, or an embargo on AI chips exported to China. However, both of these options have serious limitations, and could prove counterproductive to U.S. national and economic security interests. Instead, the authors propose that the United States expand its collection of open-source intelligence and adopt new export control measures based on high-end chip features. Key findings include:

1. The PLA is placing orders for AI chips designed by U.S. companies and manufactured in Taiwan and South Korea.

·  Of the 97 individual AI chips we could identify in public PLA purchase records, nearly all of them were designed by Nvidia, Xilinx (now AMD), Intel, or Microsemi.

·  By comparison, we could not find any public records of PLA units or stateowned defense enterprises placing orders for high-end AI chips designed by Chinese companies, such as HiSilicon (Huawei), Sugon, Sunway, Hygon, or Phytium.

2. The U.S. government is limited in its ability to meaningfully constrain the sale of AI chips to specific end-users in China. A strategy based on end-user export controls is likely insufficient to limit Chinese military access to AI chips or broader progress in AI-related technologies.

·  Difficulties associated with tracking AI chips and the variety of potential vendors would make it challenging for U.S. regulators to wage a targeted crackdown on the PLA’s intermediary chip suppliers.

·  The PLA often buys commercial off-the-shelf AI systems from Chinese academic institutions and private companies, which also buy U.S.-designed chips, and are not easily captured by U.S. restrictions on military end-users.

3. Effectively managing the PLA’s access to AI chips will require a deeper understanding of China’s defense industry and new forms of export control for technologies related to AI. For example, such controls could be based on the physical and technical characteristics of chips exported to China, rather than on their intended applications or end-users—but this regime would require buy-in from U.S. allies and partners to be effective.

·  Abandoning end-user controls in favor of a more extreme policy, such as an embargo on chip exports to China, would alienate regional partners and jeopardize the long-term viability of the U.S. semiconductor industry.

·  The United States could gain a better understanding of China’s AI defense industry by expanding its collection and analysis of open-source information.

4. Ultimately, how the United States should best manage PLA chip access—or whether it should even attempt to do so—is a contentious issue that touches competing interests across U.S. industry and government. These questions involve too many unknown variables to answer confidently. But by carefully studying what limited information the Chinese military chooses to publish about its chip purchases, this report illuminates some of the barriers the United States faces in regulating this practice—and opportunities for reform.