“Great results can be achieved with small forces,” Sun Tzu wrote in “The Art of War” some 2,500 years ago.
That quote is so old it’s now an adage. But it appears the U.S. isn’t content to wager that small actions can achieve the wide-ranging impacts necessary to gain an edge over China in the development of AI and machine learning technologies.
After implementing sweeping restrictions on the export of semiconductors to China last October, the U.S.’ recent deal with Japan and the Netherlands to restrict the export of vital semiconductor parts and chipmaking technologies to China is throwing the $600 billion global semiconductor industry into turmoil.
The implications of these restrictions are broad, given that China accounts for approximately 80% of the world’s electronics production and is a large consumer of semiconductors. To make things even more complicated, nearly every major chipmaker has Chinese customers.
But Washington doesn’t seem to be concerned with the worries of global chipmakers or near-term supply chain volatility. It’s looking far to the future: It wants to choke out China’s ability to develop and access AI technology while diversifying its sources of the increasingly important semiconductor.
The United States’ aggressive moves are about “AI dominance, which underpins what many call the fifth industrial revolution, and ultimately, about global economic leadership in the next few decades,” according to Josep Bori, research director at GlobalData.
The recent deal with Japan and the Netherlands, which includes “preventing legacy deep ultraviolet (DUV) machine exports and outright advanced AI chips,” targets China’s semiconductor business and its ability to develop its AI technology well beyond just hardware, Bori said.
You can’t make pancakes without a pan
You see, while China makes a ton of different semiconductors, it doesn’t have some of the advanced equipment that’s needed to make the fastest processors, chips and memory storage devices.
Manufacturers in the country import a lot of the chips and equipment from companies across the world, including Taiwan’s TSMC; the U.S.’ Intel, Nvidia and AMD; South Korea’s SK Hynix and Samsung; the Netherlands’ ASML Holdings; and Japan’s Nikon and Tokyo Electron.
This, to an extent, means that Chinese manufacturers like Semiconductor Manufacturing International Corporation (SMIC) rely heavily on the global semiconductor industry for the machines to make high-end chips.
According to Bori, a number of the high-end logic and memory chips are made using extreme ultraviolet (EUV) and deep ultraviolet (DUV) lithography machines.
“Initially, the [U.S.’ export] bans to China only affected EUV machines, used for the most advanced process nodes, such as 3 nm, 5 nm and 7 nm,” Bori said.
This played into the U.S.’ strategy to slow Chinese companies’ advances in AI, machine learning and other cutting-edge tech. Basically, the smaller the distance between each transistor, the faster and more power-efficient a chip becomes. The smallest process nodes, such as 3 nm, 5 nm and 7 nm, are used to develop artificial intelligence systems, smartphones, cloud data centers and self-driving cars and are used in military applications.
But the January agreement targets older DUV machines that could let Chinese manufacturers make 14 nm chips, as well as 18 nm DRAM chips and NAND flash chips with more than 128 layers, Bori added. DUV machines let you make chips at the 14 nanometer, 28 nanometer and larger process nodes; such chips are commonly used in automobiles, industrial equipment and home appliances.
It doesn’t help that global manufacturers such as Nvidia, AMD and Intel have been basically restricted from selling their high-end manufactured products to Chinese companies.
“The bans on export of products such as high-end GPUs and AI chips largely impact [Chinese] cloud vendors, self-driving car companies, the military and any AI lab, which now will have no choice but to use domestically manufactured chips. These will likely be worse than what’s currently available as they’ll have to be manufactured using older node processes,” Bori said.
Such wide-ranging restrictions have meant that global manufacturers are scrambling to diversify their customer base and set up new foundries to distance themselves from China as they’re forced to pick sides.
Carrots and sticks
Besides the oft-repeated national security concerns, another motivator for the U.S. is to diversify the production of semiconductors, which is currently concentrated in Asia, and bring more electronics production to the West.
In addition to the restrictions, last year’s CHIPS Act, which promises $52 billion in subsidies for semiconductor manufacturing in the U.S., is primarily aimed at attracting manufacturing stateside. The bill also prohibits recipients of the funding from increasing their semiconductor manufacturing footprint in China.
The U.S. also said recently that it would use funds from the CHIPS Act to set up a minimum of two large-scale logic fabrication plants to make semiconductors, as well as multiple high-volume advanced packaging facilities, by 2030.
The various incentives and restrictions seem to be working. Several chipmakers, including Intel, Global Foundries and Micron Technology, have committed to building chip manufacturing plants and foundries in the States and will invest around $80 billion through 2025.
These moves could also be in reaction to China’s own push to bring more semiconductor manufacturing within its borders. Beijing in 2014 set up a $150 billion fund to acquire chipmakers and is pouring money into its domestic semiconductor manufacturing ecosystem to meet its goal of reaching chip self-sufficiency of 70% by 2025.
Between a rock and a hard place
The U.S.’ export curbs are not merely throttling the supply and demand side of the semiconductor industry: The ban could drive deglobalization of the global chip supply chain, which is “a reverse of the trend of the last couple of decades,” and impede artificial intelligence development, Bori told TechCrunch.
While the new measures will bring more production to the West, it risks splitting the global semiconductor supply chain across geopolitical spheres, which will make it challenging for firms in the industry to do business with both the U.S. and China, he added.
The world’s semiconductor makers have read the writing on the wall and are taking steps to get ahead of the fallout.
Taiwan
The world’s largest contract chip producer, Taiwan Semiconductor Company (TSMC), makes about 65% of the world’s chips and nearly 90% of all advanced chips — and is finding itself walking a fine line.
TSMC, which counts Apple, AMD, Qualcomm and Nvidia among its customers, plans to increase its investment in the U.S. from $12 billion to $40 billion by 2024 to cooperate with the U.S. push and boost domestic supplies of chips.
The company plans to produce more advanced chips, including 3 nm process nodes, at its second semiconductor plant in Arizona, which is slated to begin operations in 2026. TSMC’s first plant in Phoenix is planned to start operations in 2024 and will make chips at the 4 nm node.
TSMC is also considering expanding its global manufacturing footprint to Japan and Europe, per its latest earnings call. The company said it is considering setting up a second plant in Japan and its first specialty fab, focusing on “automotive-specific technologies,” in Europe.
The company is still operating in China, though. TSMC said in its third-quarter 2022 earnings call that it had secured a one-year authorization from the U.S. government to operate its Nanjing facility in Jiangsu province, where it builds chips at the 28 nm and 16 nm process nodes.
The approval allows the Taiwanese firm to continue ordering U.S. chipmaking equipment to expand its Chinese operations as well. According to its earnings call in January, TSMC will expand its 28 nm chip manufacturing operations in Nanjing.
South Korea
South Korea, which has a close relationship and trade agreements with China for semiconductor manufacturing, is in a challenging position.
South Korea’s biggest memory chipmakers, Samsung Electronics and SK Hynix, both have huge facilities in China that account for a massive share of their contract chip production. Samsung has a NAND flash memory chip plant in Xi’an, which reportedly makes 40% of its NAND chips, as well as a chip-packaging facility in Suzhou.
SK Hynix has a facility in Wuxi, in Jiangsu province, which produces more than 40% of its DRAM chips; a chip-packaging operation in Chongqing; and a NAND flash plant in Dalian.
“South Korea will want to balance its exposure to the Chinese market, including its local manufacturing capability, with its exposure to the U.S. market and its significant reliance on U.S. semiconductor manufacturing equipment and design software,” Bori said in a GlobalData statement last July.
“However, given that China is aggressively working towards developing semiconductor self-sufficiency, the opportunity for South Korea to gain or even retain semiconductor market share in China is likely to diminish over time. As such, GlobalData expects South Korea to navigate the situation by keeping an equidistant position in the short term but become more aligned with the U.S. in the long term,” the statement said.
In October, Samsung and SK Hynix received a one-year exemption from U.S. restrictions on China’s semiconductor industry, and they might appeal to extend the exemption for another year.
However, the companies are quickly making plans to diversify their production. SK Hynix said in its third-quarter earnings call in October that it would consider selling its memory chip production facilities in China as a contingency plan. It also intends to invest about $15 billion in an advanced chip-packaging business and a new R&D center in the U.S.
Samsung, which operates a foundry plant in Austin, Texas, has unveiled plans to invest $17 billion for a manufacturing facility in Taylor, Texas. The South Korean company is also considering investing an additional $200 billion to set up 11 chip facilities in Texas, Bloomberg reported.
“China’s chip self-sufficiency plan, if and when it actually materializes, will be bad news for South Korea,” GlobalData analyst Daniel Clarke said in a statement. “Therefore, joining the U.S.-led alliance is a savvy geopolitical move by South Korea because it threatens to delay China’s self-sufficiency plans and sends a strong signal of South Korea’s own power in the semiconductor value chain.”
South Korean trade officials did not respond to a request for comment on ongoing discussions with the U.S.
Japan
Japan is reportedly mapping out plans to impose an export ban on cutting-edge semiconductor-making tools to China this spring, about a week after it agreed to join the U.S. in limiting exports of advanced chip technology to China.
Japan is reviving its semiconductor production to offset risks to global trade amid rising tensions between the U.S. and China and supply chain disruptions.
Last August, the Japanese government backed the launch of Rapidus, an advanced logic chip foundry, which is also supported by several major companies: Denso, Kioxia, MUFG Bank, NEC, NTT, SoftBank and Toyota. The fab will research, develop, design, manufacture and sell advanced logic semiconductors and has signed a deal with IBM to produce 2 nm chips by 2027.
Last November, Japan also unveiled plans to spend about $2.38 billion to build an R&D research hub for making advanced semiconductors with U.S. and European companies. Tokyo also has approved subsidies for TSMC, Kioxia and Micron Technology to build plants in Japan to make semiconductors that will be used for data centers, AI and other cutting-edge technologies, per a Nikkei report.
What’s in store for China?
According to Bori, semiconductor manufacturing is expected to move gradually to countries like India and Vietnam from China in the midterm, especially with companies like Apple eyeing those countries to make their products in the future.
Bori pointed out though this move was likely, it would take a while to actually happen. “It is a complex process involving not only the technology and capital equipment but also building a critical mass of human skills in this very specialized sector,” he said.
It is less clear what the far future holds, but “if China succeeds in developing its own semiconductor manufacturing technology for 7 nm and below, both in terms of tools and fabrication capabilities, then it might take back a larger market share than it enjoys today,” Bori said.
“The truth is that China’s manufacturing capabilities today are focused on the back-end process (assembly, packaging and testing), which are more labor-intensive.”
From what we can tell, it seems that the center of gravity is shifting from East Asia to a more global footprint, where most major competitors have strong capacities in the U.S. and EU and limited exposure to China. How successful the United States will be in limiting China’s near-term access to high-end chips and the equipment required to build them is not yet clear, though early signs point to material progress.
Comment