It has now been over a month since the U.S. Commerce Department issued new rules that clamped down on the export of certain advanced chips—which have military or AI applications—to Chinese customers.
China has yet to respond—but Beijing has multiple options in its arsenal. It’s unlikely, experts say, that the U.S. actions will be the last fighting word in an industry that is becoming more geopolitically sensitive by the day.
This is not the first time that the U.S. government has constrained the flow of chips to its perceived adversaries. Previously, the United States has blocked chip sales to individual Chinese customers. In response to the Russian invasion of Ukraine earlier this year, the United States (along with several other countries, including South Korea and Taiwan) placed Russia under a chip embargo.
But none of these prior U.S. chip bans were as broad as the new rules, issued on 7 October. “This announcement is perhaps the most expansive export control in decades,” says Sujai Shivakumar, an analyst at the Center for International and Strategic Studies, in Washington.
The rules prohibit the sale, to Chinese customers, of advanced chips with both high performance (at least 300 trillion operations per second, or 300 teraops) and fast interconnect speed (generally, at least 600 gigabytes per second). Nvidia’s A100, for comparison, is capable of over 600 teraops and matches the 600 Gb/s interconnect speed. Nvidia’s more-impressive H100 can reach nearly 4,000 trillion operations and 900 Gb/s. Both chips, intended for data centers and AI trainers, cannot be sold to Chinese customers under the new rules.
Additionally, the rules restrict the sale of fabrication equipment if it will knowingly be used to make certain classes of advanced logic or memory chips. This includes logic chips produced at nodes of 16 nanometers or less (which the likes of Intel, Samsung, and TSMC have done since the early 2010s); NAND long-term memory integrated circuits with at least 128 layers (the state of the art today); or DRAM short-term memory integrated circuits produced at 18 nanometers or less (which Samsung began making in 2016).
Chinese chipmakers have barely scratched the surface of those numbers. SMIC switched on 14-nm mass production this year, despite facing existing U.S. sanctions. YMTC started shipping 128-layer NAND chips last year.
The rules restrict not just U.S. companies, but citizens and permanent residents as well. U.S. employees at Chinese semiconductor firms have had to pack up. ASML, a Dutch maker of fabrication equipment, has told U.S. employees to stop servicing Chinese customers.
Speaking of Chinese customers, most—including offices, gamers, designers of smaller chips—probably won’t feel the controls. “Most chip trade and chip production in China is unimpacted,” says Christopher Miller, a historian who studies the semiconductor trade at Tufts University.
The controlled sorts of chips instead go into supercomputers and large data centers, and they’re desirable for training and running large machine-learning models. Most of all, the United States hopes to stop Beijing from using chips to enhance its military—and potentially preempt an invasion of Taiwan, where the vast majority of the world’s semiconductors and microprocessors are produced.
In order to seal off one potential bypass, the controls also apply to non-U.S. firms that rely on U.S.-made equipment or software. For instance, Taiwanese or South Korean chipmakers can’t sell Chinese customers advanced chips that are fabricated with U.S.-made technology.
It’s possible to apply to the U.S. government for an exemption from at least some of the restrictions. Taiwanese fab juggernaut TSMC and South Korean chipmaker SK Hynix, for instance, have already acquired temporary exemptions—for a year. “What happens after that is difficult to say,” says Patrick Schröder, a researcher at Chatham House in London. And the Commerce Department has already stated that such licenses will be the exception, not the rule (although Commerce Department undersecretary Alan Estevez suggested that around two-thirds of licenses get approved).
More export controls may be en route. Estevez indicated that the government is considering placing restrictions on technologies in other sensitive fields—specifically mentioning quantum information science and biotechnology, both of which have seen China-based researchers forge major progress in the past decade.
The Chinese government has so far retorted with harsh words and little action. “We don’t know whether their response will be an immediate reaction or whether they have a longer-term approach to dealing with this,” says Shivakumar. “It’s speculation at this point.”
Beijing could work with foreign companies whose revenue in the lucrative Chinese market is now under threat. “I’m really not aware of a particular company that thinks it’s coming out a winner in this,” says Shivakumar. This week, in the eastern city of Hefei, the Chinese government hosted a chipmakers’ conference whose attendees included U.S. firms AMD, Intel, and Qualcomm.
Nvidia has already responded by introducing a China-specific chip, the A800, which appears to be a modified A100 cut down to meet the requirements. Analysts say that Nvidia’s approach could be a model for other companies to keep up Chinese sales.
There may be other tools the Chinese government can exploit. While China may be dependent on foreign semiconductors, foreign electronics manufacturers are in turn dependent on China for rare-earth metals—and China supplies the supermajority of the world’s rare earths.
There is precedent for China curtailing its rare-earth supply for geopolitical leverage. In 2010, a Chinese fishing boat collided with two Japanese Coast Guard vessels, triggering an international incident when Japanese authorities arrested the boat’s captain. In response, the Chinese government cut off rare-earth exports to Japan for several months.
Certainly, much of the conversation has focused on the U.S. action and the Chinese reaction. But for third parties, the entire dispute delivers constant reminders of just how tense and volatile the chip supply can be. In the European Union, home to less than 10 percent of the world’s microchips market, the debate has bolstered interest in the prospective European Chips Act, a plan to heavily invest in fabrication in Europe. “For Europe in particular, it’s important not to get caught up in this U.S.-China trade issue,” Schröder says.
“The way in which the semiconductor industry has evolved over the past few decades has predicated on a relatively stable geopolitical order,” says Shivakumar. “Obviously, the ground realities have shifted.”
- The First Million-Transistor Chip: the Engineers’ Story ›
- U.S. Suspicions of China's Huawei Based Partly on NSA's Own Spy Tricks ›
- China's Tech Revolution - IEEE Spectrum - IEEE Spectrum ›
- The First Million-Transistor Chip: the Engineers’ Story - IEEE Spectrum ›
- How the EU Chips Act Could Build “Innovation Capacity” in Europe - IEEE Spectrum ›
- The Long and Mostly Short of China’s Newest GPT - IEEE Spectrum ›