The OECD's 2025 policy paper, "The Chip Landscape: Geographical Distribution of Wafer Fabrication Capacity," provides a comprehensive analysis of global semiconductor front-end manufacturing, emphasizing the need for enhanced data to bolster value chain resilience. Drawing from the newly developed OECD Semiconductor Production Database, which aggregates commercial sources like SEMI and TechInsights, the report maps wafer capacity by economy, process node density, chip type, ownership, and business model. This effort stems from the Semiconductor Informal Exchange Network (SIEN), aiming to identify bottlenecks, assess substitutability, and inform policies amid growing geopolitical tensions and supply disruptions.
Key findings reveal high geographic concentration: China, Chinese Taipei (Taiwan), Korea, Japan, and the United States account for nearly 90% of global in-production capacity, measured in wafer starts per month (WSPM) in 8-inch equivalents. Within these, production is dominated by a few firms—the top ten manufacturers hold about 50% globally. For instance, in Japan, 73 companies operate fabs, but the largest five (Kioxia, Sony, Toshiba, Micron, Renesas) control 58%. Upcoming investments, including planned and under-construction fabs, largely reinforce this pattern, with the U.S. seeing an 81% capacity boost from players like Micron and TSMC.
Capacity distribution varies significantly by process node and chip type. Korea's output is heavily skewed toward advanced nodes (<22nm) for memory chips, while the U.S. spans a broader range. Breaking down by chip categories—power/discrete, analog, mature logic (≥20nm), advanced logic (<20nm), commodity memory (DRAM/NAND), and specialty memory—highlights disparities. China leads in power (6.28M WSPM), analog (3.64M), and mature logic (4.23M), but trails in advanced logic, where Taiwan dominates (1.55M). Korea excels in commodity memory (4.58M). Fabs often have mixed capabilities; analog and mature logic are produced in versatile plants, whereas advanced logic and commodity memory require specialized, larger fabs averaging 72k and 111k WSPM, respectively.
Ownership analysis shows domestic firms control most capacity in top economies—e.g., 93% in China—but foreign investments are rising, as seen with TSMC and Samsung expanding in the U.S. and Japan. Business models also differ: foundries (pure-play and IDM-foundries) comprise over 50% in China and Taiwan, serving fabless designers, while IDMs prevail elsewhere, like in Korea's memory sector.
The report underscores limited substitutability: fabs optimized for one chip type cannot easily switch, amplifying disruption risks. Data limitations— incomplete process tech details, uneven coverage (especially China), and uncertainty in projections—hinder precise assessments. Future work includes integrating demand data to monitor imbalances and extending scope to back-end manufacturing.
Bottom line: the paper calls for international cooperation to develop robust datasets, promote diversification, and mitigate systemic vulnerabilities. As semiconductors underpin digital economies, addressing these concentrations is vital for resilience against shocks like pandemics or trade restrictions.
