01

Company Updates（July.28）

AMD Partners with South Korean Government to Establish AI Research Center, Building Open Ecosystem for Heterogeneous AI Computing

On July 28, according to Yonhap News Agency, South Korea’s Ministry of Science and ICT signed a memorandum of understanding with AMD on cooperation in the AI chip ecosystem. The two sides will establish an AI research center in South Korea, known as the “AI Center of Excellence,” with the aim of building an open AI infrastructure ecosystem. The two sides plan to develop heterogeneous AI computing infrastructure by connecting AMD’s CPUs and GPUs with AI chips, or NPUs, from South Korean companies that have strengths in inference. The cooperation will cover areas including memory, networking, servers, and software. The center will provide support for South Korean companies, universities, and research institutions in areas such as technical cooperation, AI chip software development, computing technology validation, and joint research. It will also promote collaboration with local South Korean AI chip companies.

Commentary: AMD’s establishment of an AI research center in South Korea is an important move in its global AI ecosystem strategy. South Korea is home to world-leading memory chip manufacturers such as Samsung and SK Hynix, as well as a group of startups with accumulated expertise in AI inference chips. By integrating AMD’s CPU/GPU compute capabilities with South Korea’s local NPU, memory, and networking resources, the two sides are seeking to build a full-stack AI solution spanning chips to systems. This collaborative model of “global compute leader + local chip ecosystem” not only broadens AMD’s commercial entry point into South Korea’s AI market, but also provides South Korean AI chip companies with a technical pathway to connect with an internationally leading compute platform.

From an industry trend perspective, as AI compute demand extends from training to inference, the importance of heterogeneous computing architectures is becoming increasingly prominent. It is difficult for any single vendor to cover the entire technology stack from CPUs and GPUs to NPUs, making open ecosystems and cross-platform collaboration an industry consensus. AMD’s decision to cooperate with the South Korean government rather than pursue purely commercial sales reflects the evolution of global AI competition from “single-point technological breakthroughs” toward “ecosystem construction.” For China’s domestic AI chip industry, AMD’s deeper integration with South Korea’s ecosystem highlights the importance of building an open and diversified AI compute ecosystem. Beyond compute chips themselves, coordinated innovation in interconnect standards, software stacks, and system integration will also determine a country’s long-term competitiveness in the AI industry.

02

Market Updates(July.28)

Magnitude 7.1 Earthquake in Kumamoto Hits Japan’s “Silicon Island” Semiconductor Supply Chain, Forcing Multiple Fabs and Equipment Makers to Halt Operations

On the afternoon of July 28, a magnitude 7.1 earthquake struck Kumamoto Prefecture, Japan, with the hypocenter at a depth of approximately 10 kilometers and the maximum seismic intensity reaching 7. Kumamoto Prefecture is a key hub of Japan’s semiconductor industry, home to numerous supply chain companies including wafer fabs, equipment makers, materials suppliers, and component manufacturers.

TSMC’s Kumamoto fab, JASM, experienced seismic intensity above 5. The company immediately evacuated employees and gradually resumed production after confirming their safety. According to its inspections, the plant’s building structure remained safe, and water supply, power, and industrial safety systems were operating normally. However, due to the strong tremors, equipment inspection and calibration will require some time. The second fab currently under construction was not affected by the earthquake, and some construction work resumed on July 29 after safety was confirmed. Tokyo Electron (TEL), a major Japanese semiconductor equipment maker, halted operations at its Koshi and Ozu sites in Kumamoto Prefecture on July 29 for safety inspections. The two sites are mainly engaged in the R&D and production of coater/developer equipment, cleaning equipment, and 3D packaging equipment. Renesas Electronics has suspended production at both its Kawashiri and Nishiki plants in Kumamoto. Damage such as fallen ceiling panels and water leakage has been identified, and the company is assessing the calibration time required before production can restart. In addition, two Mitsubishi Electric power semiconductor plants have halted operations. Sony’s Kumamoto Technology Center, which produces industrial equipment and automotive image sensors, has also suspended production, with no restart date yet determined. The earthquake also disrupted the local automotive industry, with Toyota, Nissan, and Honda again suspending operations at their plants in the Kyushu region until the end of the month.

Commentary:The impact of the magnitude 7.1 earthquake in Kumamoto on Japan’s “Silicon Island” semiconductor supply chain once again highlights the vulnerability of global semiconductor manufacturing to natural-disaster risks caused by geographic concentration. As an important advanced-process capacity site in Japan, TSMC’s Kumamoto fab demonstrated mature emergency response capabilities through its rapid recovery. However, the physical damage at Renesas’ plants and Tokyo Electron’s production halt for safety inspections still pose short-term disruptions to the supply of automotive MCUs and semiconductor equipment. Against the backdrop of already tight supply and demand for automotive chips and semiconductor equipment, any production interruption could be amplified into broader supply chain volatility. This earthquake also serves as a warning to the global semiconductor industry: beyond geopolitical risks, natural disasters are also a core variable that must be considered in concentrated capacity deployment. For the global supply chain, which remains highly dependent on Japanese semiconductor materials and equipment, such events further reinforce the urgency of supply chain diversification.

03

Supply-Demand Updates(July.29)

Samsung Electro-Mechanics Raises MLCC Prices by 30%, as AI Server Demand Continues to Tighten Passive Component Supply-Demand Balance

According to industry sources on July 29, Samsung Electro-Mechanics recently notified its sales partners that it will raise prices for MLCC, or multilayer ceramic capacitor, products by 30% starting August 1. In the notice, Samsung Electro-Mechanics said that global electronics demand has seen a structural surge over the past few months. Although the company has continued to invest in improving production efficiency and expanding capacity, supply chain pressure has exceeded a reasonable level. Taiyo Yuden, the world’s third-largest MLCC manufacturer, also issued a notice to customers on July 23, requesting price adjustments for products shipped from September 1.

According to TrendForce data, Samsung Electro-Mechanics shipped approximately 98bn MLCCs in June, while Murata shipped 140bn units and Taiyo Yuden shipped 40bn units. The combined monthly shipments of the three major suppliers reached a five-year high. However, demand has far exceeded shipments. As of the end of June, Samsung Electro-Mechanics’ book-to-bill ratio reached 1.31, while Murata’s stood at 1.30 and Taiyo Yuden’s at 1.25, all marking their highest levels since the pandemic. Industry inventory has fallen to fewer than 30 days. According to estimates, Nvidia’s GB200 AI server requires approximately 6,500 MLCCs, while the next-generation Vera Rubin platform is expected to require around 12,000 units. Samsung Electro-Mechanics has recently signed long-term supply agreements with two global technology companies, worth approximately KRW 454bn and KRW 295bn, respectively.

Commentary: Samsung Electro-Mechanics’ 30% MLCC price hike is another landmark signal of tightening supply and demand in the passive component industry, following Yageo’s price increases across its full capacitor portfolio. Consistent with the logic behind Yageo’s price hikes, the core driver is the exponential growth in MLCC usage by AI servers. From around 6,500 MLCCs in GB200 to approximately 12,000 units in Vera Rubin, MLCC usage per AI server is nearly doubling, while global MLCC capacity expansion is far from keeping pace with demand growth. With the book-to-bill ratios of the three major suppliers all reaching around 1.3 and industry inventory falling below 30 days, the sector has entered a deep supply-shortage phase that is unlikely to ease in the short term.

From a competitive landscape perspective, the supply-demand imbalance in the MLCC market is reshaping pricing power and bargaining power across the supply chain. Samsung Electro-Mechanics, Murata, and Taiyo Yuden, the three major Japanese and South Korean leaders, control the majority of global high-end MLCC capacity, leaving pricing initiative firmly in the hands of suppliers. For China’s domestic MLCC manufacturers, price hikes and longer lead times among overseas leaders are creating a rare qualification and adoption window for domestic substitution. Under the dual pressure of rising costs and tight supply, end customers now have stronger incentives to validate and adopt products from local suppliers. Whether domestic MLCC makers can seize this window and achieve substantive breakthroughs in technology and capacity for high-end categories such as high-capacitance, high-voltage, and high-reliability MLCCs will determine their final position in the restructuring of the global passive component supply chain.

04

Company Updates(July.29)

UMC Raises 2026 CapEx to USD 2bn, Launching New Tainan Fab and Singapore Silicon Photonics Expansion

On July 29, foundry company UMC said that in response to continued growth in customer demand, its board of directors had approved an increase in 2026 capital expenditure to USD 2bn. The company has also formulated a phased investment plan, under which it will expand cleanroom capacity at its Singapore site while simultaneously starting construction of a new wafer fab at the Tainan Science Park in China’s Taiwan region. In Singapore, UMC will invest in cleanroom construction and equipment procurement at its Phase 4 site to expand silicon photonics capacity. The new fab building in Tainan will serve as the future base for the company’s Phase 7 (P7) and Phase 8 (P8) facilities.

Commentary: UMC’s decision to raise its capital expenditure to USD 2bn and expand along two fronts represents one of the company’s more aggressive capacity expansion moves in recent years. The Singapore site will focus on expanding silicon photonics capacity, extending UMC’s previously launched 12-inch silicon photonics chip production and targeting strong structural demand for high-speed interconnects from AI data centers. The new Tainan fab, meanwhile, lays the groundwork for the future P7 and P8 sites, with the aim of consolidating UMC’s market share in mature processes and specialty technologies. This dual-track strategy reflects UMC’s optimistic view on the sustainability of AI-related demand: the company is betting on the emerging silicon photonics track while also maintaining its traditional foundry business foundation.

From a regional footprint perspective, UMC’s decision to place silicon photonics expansion in Singapore and build its new fab in Tainan creates a dual-core structure of “overseas advanced technology platform + local mature-capacity base.” The Singapore site benefits from local research capabilities and talent reserves in photonics, while Tainan relies on UMC’s existing manufacturing infrastructure and supporting supply chain.

05

Market Updates(July.30)

SEMI Report: Global Silicon Wafer Shipments Rise 7.4% YoY in Q2 2026, as AI Demand Continues to Drive Market Expansion

On July 30, the quarterly silicon wafer industry analysis report released by SEMI’s Silicon Manufacturers Group showed that global silicon wafer shipments rose 7.4% YoY in the second quarter of 2026 to 3,573 million square inches (MSI), up 9.1% QoQ from the first quarter. Ginji Yada, Chairman of SEMI SMG and General Manager of the Sales and Marketing Division at SUMCO Corporation, said silicon wafer shipments maintained solid growth in the second quarter. Strong and growing AI-related demand has expanded from advanced logic and memory into power devices, optoelectronics, and other markets. In addition, demand from the industrial and automotive sectors is recovering, while memory price pressure has constrained demand from PCs and smartphones. Device manufacturers are making major investments in capacity expansion, and growth in silicon wafer demand is expected to continue.

Commentary: The 7.4% YoY and 9.1% QoQ growth in global silicon wafer shipments confirms the continued upward momentum in the semiconductor industry cycle. What is more noteworthy is the broadening of the demand mix. AI-driven demand for silicon wafers has expanded from advanced logic and memory into a broader range of categories such as power devices and optoelectronics, indicating that AI infrastructure construction is penetrating from “compute chips” into the entire semiconductor supply chain. The recovery signals from the industrial and automotive markets provide a second growth curve for semiconductor demand, partially offsetting the pressure caused by weak consumer electronics.

From the supply side, continued growth in silicon wafer shipments means that fab capacity utilization remains at a high level. This also cross-validates earlier data points such as record ASML lithography system shipments and YoY growth in global semiconductor equipment sales. For China’s domestic silicon wafer industry, the expansion of global demand provides a rare market window for local large-diameter wafer manufacturers. However, domestic high-end 12-inch silicon wafers still lag international leaders in quality consistency and capacity scale. Whether Chinese wafer makers can complete technology catch-up and customer qualification during this industry upcycle will determine their final position in the restructuring of the global silicon wafer supply chain.

06

Company Updates(July.29)

Jiuzhou Yigui Subsidiary Plans RMB 630mn Investment in Wafer Laser Stealth Dicing Project, Targeting Silicon Photonics Chip Cutting and Processing

On July 28, Jiuzhou Yigui announced that its wholly owned subsidiary, Suzhou Jingxi Semiconductor Technology Co., Ltd., plans to invest in the construction of an advanced semiconductor wafer laser stealth dicing industrialization project. The project is expected to involve total investment of no more than RMB 630mn, including no more than RMB 600mn for equipment purchases. According to the plan, the project will adopt a leased-factory model. The leased facility will undergo adaptive renovation, including cleanroom construction and supporting production facilities, to meet production requirements. The project is expected to complete production line construction in the first quarter of 2027 and gradually enter production. It will mainly provide cutting and processing services for 8-inch and 12-inch silicon photonics chips and silicon dummy bars. Downstream customers will include optical module manufacturers and chipmakers. The project will offer services including high-precision semiconductor wafer stealth dicing, grooving, and sorting. The investment is still subject to approval by the company’s shareholders’ meeting.

07

Company Updates(July.31)

China’s First Domestic OLED TDDI Chip Enters Mass Production, as Chipone Achieves Commercial Breakthrough in High-End Display Driver Chips

Recently, ICNA3611, China’s first domestically developed OLED touch and display driver integration (TDDI) chip independently developed by Chipone, entered large-scale mass production in mainstream models of a well-known domestic smartphone brand. The chip supports high-performance specifications including 1.5K ultra-high-definition resolution, a 165Hz high refresh rate, and a dual-finger 360Hz touch sampling rate, delivering a comprehensive upgrade in display precision and touch responsiveness. Through display and touch integration technology, ICNA3611 uses a single-chip architecture to simultaneously optimize display quality and touch response. It addresses electromagnetic interference challenges under high refresh rates, simplifies component layout to create more space for thinner and lighter terminal designs, and integrates packaging and testing processes to significantly reduce reliance on key overseas materials. The overall solution delivers a significant cost reduction compared with traditional discrete solutions, accelerating the adoption of OLED technology in mid- to high-end smartphone models. The chip was named one of the “Top Ten Innovative Products” at the World Display Industry Innovation and Development Conference held at the end of 2025.

08

Company Updates(July.31)

Zhongji Innolight Refutes Rumors of 1.6T Optical Module Price War, Says 2027 Orders Are Fully Booked and AI Investment Visibility Extends to 2028

Recently, market rumors suggested that some industry players had lowered quotations for 1.6T optical modules to USD 600, sparking concerns over an early outbreak of a price war and a potential cooling of CSP capital expenditure. In response, China’s leading optical communications company Zhongji Innolight urgently held an investor conference call to clarify the matter. The company said that optical module prices do undergo normal annual adjustments based on supply and demand conditions. However, current market demand remains strong, supply of key components is tight, and suppliers capable of large-scale mass production of 1.6T optical modules remain very limited. Therefore, there is no malicious price-cutting competition. Zhongji Innolight emphasized that the company has already secured firm orders for 2027, while some customers have even provided demand plans for new products in 2028. Nearly all major customers have placed formal orders for 2027, clearly specifying monthly delivery volumes. The company believes that if major cloud service providers expected future investment to slow, they would not be planning product demand for 2028 in advance. At present, the company sees no signs of cooling in AI infrastructure investment. In response to doubts over the technical barriers of optical modules, Zhongji Innolight stressed that the optical communications industry has fairly high technological barriers. As AI clusters continue to expand, the number of optical modules required per GPU has at least doubled compared with the past.