Center for Strategic and International Studies, April 24, 2024, blog post by Paolo Triolo
Washington and Brussels are bewildering the impact that China's industrial policies could have on the traditional chip market, what some industry insiders call "mature node semiconductors," similar to what is likely to happen in the photovoltaic industry (PVs) over the past decade and electric vehicles (EVs) in the next decade. That said, there are growing fears of "overcapacity" in mature types of semiconductors, partly due to the desire to aggressively address potential areas of overcapacity in the future by seeing the impact of China's subsidies on steel and PV on global markets.
To be sure, the Chinese government has provided subsidies for the entire semiconductor supply chain, including subsidies for mature node manufacturers. The core issue is China's current plans to build mature node fabs and the conclusion that this will lead to overcapacity. However, the reality is more complicated, as it turns out that mature node semiconductors are very different from photovoltaics or electric vehicles. In fact, this may not be the overcapacity problem that anyone is looking for.
Global and Chinese markets
Let's look at the market structure of mature semiconductor production around the world, and then look specifically at China to illustrate why the concern may be a mistake.
When discussing this topic, the industry prefers to use the term "mature node" rather than "traditional chips". Mature nodes are generally thought of as products that are manufactured at 28 nanometers (nm) or higher, a definition used by the Department of Commerce (as part of the wording of the CHIPS Act) and the Bureau of Industry and Security (BIS) when asking U.S. companies to provide advice on their use of China-origin semiconductors. Semiconductors produced using these more mature processes are typically produced in factories where equipment costs have been fully amortized, but their margins are much lower than those of the most advanced semiconductors, including central processing units (CPUs), graphics processing units (GPUs), and other application-specific integrated circuits (ASICs) used in smartphones, AI model training, or other applications that require high performance and low power consumption. Typically, the term "mature" does not include memory chips, as most memory is produced and sold on more advanced nodes, where capacity, speed, and density are all important.
There are four key fundamentals in the mature node semiconductor market, all of which are related to the overcapacity debate:
First, the term "mature semiconductor" encompasses a range of different types of chips, each with its own supply and demand dynamics. This includes both specific types of semiconductors, including logic, power, RF, analog, and digital hybrid semiconductors, as well as mature semiconductors for specific types of end-uses, such as automotive, robotics, drones, industrial automation, aerospace, and other industries. As a result, "traditional semiconductors" do not have a single market like products such as electric vehicles and photovoltaics. This makes "overcapacity" an inappropriate perspective. For example, aggregating fab capacity in a particular country by process node does not reflect the diversity of applications and requirements actually covered by the semiconductors produced at a particular node.
Second, most of the world's mature semiconductor node capacity is in the hands of so-called integrated equipment manufacturers (IDMs), while in China, the capacity of mature nodes is mainly occupied by companies engaged in foundry services, including highly specialized foundry services. Foundries manufacture semiconductors based on customer-provided designs. These are contract production levels that depend on market demand determined by the foundry's customers rather than the foundry itself. Companies design semiconductors that are closely related to the needs of their specific industries, and they seek to carefully balance supply and demand.
In fact, their business model is designed to avoid "overcapacity", which is more common to international IC manufacturers. Fabs tend to be highly specialized, and each fab facility a company operates is often designed to meet the needs of a customer's specific product. Therefore, aggregating the capacity figures for specific nodes does not account for the supply and demand functions of specific types of semiconductors produced by those nodes. In China, according to the authors' discussions with industry officials, manufacturing capacity expansion at mature nodes will be led by foundries by 2030. In fact, in 2023, the foundry/IDM ratio is around 60/40, and it is expected to be 64/36 in 2030.
Third, because foundries have very thin margins in the mature node semiconductor space, it is difficult to quickly switch production lines, so foundries and customers are more willing to enter into long-term contractual arrangements to lock in the supply of specific types of semiconductors. This is especially true in industries with long product life cycles, high safety requirements, and strict qualification of product quality and reliability, such as medical devices and automotive applications.
Fourth, and perhaps most importantly, is the concept of "economic overcapacity" that is often overlooked or misunderstood within the industry. Because, in fact, the global industry believes that a certain amount of excess supply is desirable, and it is essential to smooth out the expected and common supply and demand disturbances. These risks include unpredictable tool downtime, natural disasters such as the Fukushima earthquake that affected front-end manufacturing and material suppliers, Texas winter freezes that affected Samsung, and other accidents such as critical facility fires that have occurred in the past few years. Some industry insiders believe that the optimal level of overproduction is about 15%-20%! Even if there is a major chip shortage during the epidemic, there is still a continuous shortage of chips in some mature nodes, and the healthy development of the entire system requires a certain degree of economic overcapacity.
Finally, focusing on issues such as fab utilization may not be a good proxy for overcapacity, which can translate into dumping, which is defined as exporting products at a price lower than the domestic market. For example, a plant involves a large number of complex tools that run at high speeds and requires a lot of maintenance. Utilization varies from 70% to 90%, depending on the node, and crucially, on demand. During the pandemic chip shortage, some factories were operating as high as 99%. Obviously, their tools are running too hard, too fast, and too long, which is neither normal nor desirable. However, the circumstances at the time required this. Of course, this situation did not lead to overcapacity or dumping. Only when a plant is operating at 99%-100% utilization for two or three years, without a broader global supply chain crisis, will government officials worry that high utilization rates will lead to potential dumping.
China and concerns about overcapacity
To address the question of whether there may be overcapacity in mature semiconductor production, it is important to understand what is driving the expansion of foundry capacity in China. A number of factors highlight the difficulty for Chinese companies, whether foundries or IDMs, to "flood the market" and push prices down and disrupt the market.
At present, Chinese companies account for an increasing share of the world's mature semiconductor front-end manufacturing. This market share varies by process node, accounting for about 27% in 28-65nm process node production and dropping to around 20% in 90-180nm process node production. If the currently announced fab capacity expansion can be achieved in the next 2-5 years, then by 2030, Chinese companies may occupy a larger share of global capacity, however, it is difficult to determine. This is because mature and advanced node semiconductor production capacity in the United States, Japan, Taiwan, South Korea, and Europe will also expand significantly.
There are three reasons why the problem of overcapacity has been exaggerated or misunderstood as coming from China alone:
First of all, the expansion targets are different. The goal of expanding production capacity of Chinese companies such as SMIC, Huahong, Grace, Holley, and Sien is mainly to supply domestic demand. Since the vast majority of China's domestic demand for semiconductors is still dependent on imports, there is a commercial driver for expanding domestic production capacity. U.S. export controls have also accelerated this trend, preventing companies like SMIC from focusing on advanced node production, and prompting the Chinese government to encourage domestic companies to seek domestic suppliers to replace foreign suppliers for hardware and software in a range of government agencies and industrial sectors. For example, SMIC has gone from 60% of its production capacity serving foreign customers five years ago to nearly 80% of its production capacity serving domestic customers now. Huahong also has nearly 80% of its production capacity for domestic customers. This is very different from companies such as photovoltaics, which have targeted the export market from the beginning.
Second, demand is key. Most of the comments on the potential for overcapacity in China and mature semiconductors have focused on the supply side, i.e., extrapolating capacity expansion after the completion of all mature node fabs under construction in the next 3-5 years. However, the key problem with semiconductors is demand. China's domestic demand remains high and is set to grow significantly by 2030. Matching demand forecasts based on industry-specific applications such as servers, PCs, mobile devices, automotive, industrial sectors, etc., with capacity at specific process nodes such as 28, 40, 65, 90, 180 nm, provides a way to assess if and when signs of overcapacity will become a reality. For example, a non-public industry research report I saw showed that by 2030, domestic capacity would be able to meet about 90% of domestic demand, including Chinese OEMs and foreign OEMs building factories in China, assuming that all announced Chinese fabs are up and running by 2030. In 2020, the figure was around 37%.
Third, the benefits of government support have not yet begun to be felt. Concerns about China and mature node semiconductors have focused on how the potential benefits of Chinese government subsidies can accumulate to Chinese foundries and how this could lead to overcapacity. When assessing this issue, it is important to note that the margins, capex, and depreciation of leading Chinese foundries such as SMIC and Hua Hong are all better than the industry average. For example, as the global semiconductor industry has just emerged from a severe recession, all foundries are underutilized, are cutting prices to get orders, and are building fabs with high depreciation rates and low profit margins, with the exception of advanced nodes that TSMC produces for products such as Nvidia's GPUs. Similarly, SMIC's and Hua Hong's financial performance levels are broadly in line with industry averages, so it's hard to say that SMIC's price cuts are due to government subsidies, or that their higher profit margins are due to government subsidies.
In addition, even though Chinese foundries can offer below-market prices for mature node semiconductors in some cases, the cost of these semiconductors is already relatively low, so there is little incentive for domestic and foreign companies to increase orders from Chinese foundries. This is also because semiconductors are intermediate products, sourced by original equipment manufacturers (OEMs), unlike electric vehicles or photovoltaics. In addition, domestic OEMs are unlikely to react to SMIC's slightly lower prices or the Chinese government's advice to source semiconductors domestically. Most, but not all, Chinese OEMs are market-oriented, globally competitive, and they also focus on other factors such as reliability and quality. Foreign OEMs are likely to react to lower prices and government-driven local sourcing requirements as they have to compete with Chinese semiconductor companies and local OEMs. Here, "in China for China" is a trend for foreign businesses.
In addition, the quality of mature node semiconductors for specific end-use applications produced by Chinese foundries varies widely: the products produced by these foundries are becoming increasingly competitive in certain consumer electronics applications, but are still significantly less reliable in end-use applications such as automotive. In this regard, foreign companies such as NXP, Infineon, Renesas and Texas Instruments will continue to dominate the Chinese market, while local Chinese players are becoming more competitive in certain automotive semiconductor segments (e.g., powertrains) but lagging significantly behind in other areas (e.g., ADAS platforms). Finally, while Chinese foundries did lower prices, putting pressure on non-Chinese foundries. For example, PSMC competes directly with Chinese foundries in areas such as dedicated memory and display integrated circuits. In addition, some fabless companies and OEMs have embraced the low prices and diversification of manufacturing sources that come with healthy competition.
Looking Ahead: U.S. and EU Concerns About China
Officials in both the U.S. and the European Union are taking a hard look at possible overcapacity in mature node production and trying to develop policies to deal with it. To determine whether there is a real problem, it is particularly important to understand the position of Chinese companies in the global and domestic value chains. The U.S. Department of Commerce surveyed about 100 U.S. companies in the first quarter of 2024 to determine their dependence on Chinese companies for mature semiconductors. The dependency issue is closely related to the overcapacity issue, as U.S. officials fear that overcapacity could lead to greater dependency, leaving companies vulnerable to supply chain disruptions at mature nodes.
The situation is also compounded by the fact that foreign companies, including U.S. companies, are assembling large quantities of finished products in China with proven semiconductors. U.S. companies rarely import individual semiconductors, but rather products that contain one or more established semiconductors, which may come from China, including domestic and foreign foundries operating in China. The share of mature Chinese semiconductors in the IT products produced by foreign companies in China is likely to rise due to U.S. export controls and China's increasing pressure on the supply chains of certain governments and state-owned enterprises to reduce the amount of semiconductors purchased from foreign countries.
For example, the Ministry of Finance and the Ministry of Industry and Information Technology released new draft guidelines for government procurement of PCs, laptops and servers in August last year, and issued implementation documents at the end of December. At the same time, the China Information Technology Security Assessment Center has also released a whitelist of "safe and controllable" processors recognized in China. The center's whitelist includes processors from Huawei, Phytium, Loogsong and Zhaoxing, with the first three companies on the U.S. Department of Commerce's Entity List.
Procurement efforts and these types of guidelines, which are usually managed by internal documents, appear to be a rare disclosure of these guidelines this time. This work is known internally as "Xinchuang", which means innovation in the application of information technology. A local government official said the new guidelines are the first national, detailed and clear guidelines for the promotion of "new creations". In addition, the SAAC has instructed state-owned enterprises to work towards transitioning all technology needs to domestic suppliers by 2027, which presumably means hardware and software for products such as computers, laptops and servers, and more and more mature semiconductors are likely to be whitelisted as domestic production increases and quality levels improve.
Given that this process can also be seen as protectionist and non-market behavior, and, apparently, has also attracted a lot of attention from the IT industry, the domestic demand curve for key applications of mature node semiconductors will only rise, and this process also undermines the view that there will soon be some degree of overcapacity in domestic semiconductor production. Given past experience with PV and current concerns about electric vehicles, this will not stop governments from worrying about potential overcapacity. However, semiconductors are really different. Nonetheless, the scale of China's subsidies to semiconductor manufacturers and the scale of domestic mature node capacity growth (see Figure 1) are significant and have raised legitimate concerns in the semiconductor industry and in Western capitals.
When assessing the impact of China's massive expansion, it's important to remember that companies in mature node sectors outside of China are not standing still. For instance, Global Foundries, along with TSMC and PSMC, are adding mature node capacity in the United States, Singapore, Japan, Germany, and India, providing a diversified mature node footprint to balance some of China's organic growth. For Chinese companies to "win" in the mature node semiconductor space by 2030, both geopolitical tensions between China and the United States need to be eased, and Chinese OEMs need to be able to achieve cost advantages. Under this vision, Chinese OEMs using mature node semiconductors to manufacture electric vehicles and other related IT products in the automotive industry will gain a significant global market share and demonstrate significant cost advantages in using mature semiconductor node products from Chinese companies, and drive the increasing adoption of these products by incumbent OEMs.
This means that mature node semiconductors originating in China are used in products sold in both the Chinese and international markets. In contrast, the status quo remains largely the same: Chinese OEMs continue to gain domestic market share, but fail to gain traction outside of China as geopolitical tensions overwhelm cost advantages. In this case, China-based mature node semiconductor companies can only gain a certain share of Chinese OEMs, while global OEMs continue to favor products from existing mature node semiconductor producers. As a result, a bifurcation between the Chinese market and the rest of the world seems more likely than China creating "overcapacity" that would flood the global market. Current estimates of the capacity and demand balance in 2030 suggest that meeting the large domestic demand will remain the focus of Chinese foundries even if domestic capacity grows substantially.
Any policy response related to overcapacity in mature node semiconductors must take into account all of the above factors. In determining whether there is reasonable and sufficient reason to believe that there will be similar overcapacity by 2030, the government must be very cautious and fully understand the complexity of the sector on both the supply and demand sides, which is very different from electric vehicles and photovoltaics. Policymakers in the U.S. and the U.S. and the U.S. should draw on industry expertise to understand the market direction at each technology node and seek industry input on areas where reliable data is particularly difficult to obtain, such as demand forecasts through 2030. Without a thorough understanding of the industry, supply and demand, business considerations, regional and geopolitically driven business preferences, and technology roadmaps, it is impossible to determine whether there is a problem in fact or what should be done to address it.