On July 10, 2024, NATO and its Indo-Pacific partners announced four new joint projects, one of which is dedicated to artificial intelligence (AI). This collaboration marks a stronger effort to counter China’s AI advancements and underscores growing concerns about perceived threats posed by Chinese AI. In Washington, developing advanced AI algorithms to confront those threats has emerged as a strategic priority.
AI competition with China hinges on state-of-the-art, massive data centers that can house immense computational resources – i.e., advanced semiconductors and processors – to support advanced AI model development. The increasing scarcity of these resources is now defining China-U.S. AI competition more than algorithmic superiority.
Unfortunately, U.S. attempts to exclude China from the most advanced chips are proving costly and ineffective. The United States must either prepare for an endless and costly battle for AI resources with China or shift toward cooperation rather than pure competition.
Scarce Resources and Expensive Limitations
Beyond the space, water, and electrical needs for data centers to support these models, processors using state-of-the-art semiconductors are necessary for advanced AI. Graphical Processing Units used in these data centers that leverage advanced chips are scarce due to increasing demand from the private sector, supply chain disruptions in critical minerals, and trade restrictions. Only a handful of companies control the highly globalized semiconductor supply chain, making each piece of the servers in data centers scarce in their own unique way.
Preventing China from accessing semiconductors has therefore become a strategic necessity for the United States to “win” in AI competition against China. The idea was that the U.S. is dominant or has allies who are dominant in the design of chips and semiconductor manufacturing equipment, which would be major chokepoints in Chinese semiconductor manufacturing.
However, China produces an estimated 60 percent of all rare earth minerals and commands 90 percent of the refining and processing of those minerals. China has incontrovertible dominance in the mining and processing of many minerals vital to semiconductor manufacturing, and is a significant player in the manufacturing of materials for semiconductors.
The scarcity of rare earth minerals and both highly globalized and monopolized semiconductor supply chains are a consistent hamper on the “chip war” being undertaken by the Biden administration. This effort has already proven expensive, costing U.S. taxpayers $574 billion as of September 2023. Worse yet, it has not been successful in prohibiting Chinese companies’ access to advanced chips or adequately controlling semiconductor supply chains, despite a plethora of new rules and regulations enacted by the Biden administration.
Chip War Regulation Keeps Finding New Targets
In 2022, the Biden administration implemented a series of export bans on advanced chips and equipment using them, then expanded the scope of these bans in 2023. The U.S. government also applied pressure on the Dutch government and ASML, one of the few manufacturers of extreme ultraviolet lithography systems needed to manufacture highest-tech semiconductors, to stop exports to China. As of July 2024, half of ASML’s second quarter earnings came from Chinese sales.
The smuggling of banned chips into China has increased with no indication of slowing down. Some experts have estimated that tens of thousands of advanced chips could be smuggled annually into China via shell companies and other illicit means. Methods to track and minimize chip smuggling would require massive, long-term mobilization of U.S. government resources, likely to little advantage because of cloud computing.
In January 2024, U.S. Secretary of Commerce Gina Raimondo announced a new regulation that requires cloud computing companies to verify if foreign companies are using U.S. data centers to train AI models in an attempt to close some of this loophole. On July 17, 2024, The Information published a credible report stating that Google, Microsoft, and other non-U.S. cloud computing companies have been providing Chinese firms access to servers equipped with advanced AI chips, among other recent examples of Chinese businesses trying to circumvent bans.
Cooperation Is the Logical Path Forward
Thus far, U.S. efforts to cut China off from advanced chips appear to have fallen short of meeting the intended objectives. Beijing has doubled down on its semiconductor self-sufficiency drive and seems to be successfully manufacturing advanced chips domestically. U.S. policy has also served to further stimulate the chip smuggling market and put global cloud computing services in their crosshairs. The scarcity, concentration, and globalization of the semiconductor supply chain make computation a consistent and long-term defining feature of a chip war with no end in sight.
Both countries are damaging their own AI advancement by restricting the movement of experts between countries and limiting access to already constrained computation resources. Chinese and American AI experts collaborate more than any other countries, benefitting from a continuous flow of AI experts between the two in academia and the private sector.
As the uncontested global leaders in AI with a history of mutually beneficial cooperation, there are two paths that the United States and China can take: continue exclusive, zero-sum competition at great consequence, or pursue a more productive competition that accompanies mutually beneficial cooperation.