Weekly News | Supply chain trends in Semiconductor industry #217

217nd of weekly news

01

On June 16, TSMC, the world’s leading foundry, and Amkor Technology, the world’s second-largest outsourced semiconductor assembly and test (OSAT) provider, announced that they had entered into a ten-year strategic partnership agreement. Under the agreement, the two companies will set up a framework for close collaboration, with TSMC procuring advanced packaging and testing services from Amkor as they jointly expand advanced semiconductor packaging capacity in Arizona. Amkor had previously chosen Peoria, Arizona, as the site for its new facility, with proximity to TSMC’s Arizona campus as a key consideration.

TSMC plans to establish CoWoS and 3D-IC capacity in Arizona by 2029 and build its first advanced packaging facility at its local campus. Apple and Nvidia have already begun sourcing chips from TSMC’s Arizona fab, but many of those chips still need to be shipped back to Asia for packaging, leaving a weak link in supply chain resilience.

Driven by the news, Amkor Technology’s U.S.-listed shares, traded under the ticker AMKR, rose more than 13% intraday to USD 96.66, setting a new intraday high. Amkor said that as demand for high-performance computing (HPC), AI and advanced electronics grows rapidly, advanced packaging has become a key technology for improving system-level performance and integration. The partnership will help end customers bring products to market faster while building a more integrated and resilient semiconductor supply chain.

U.S.-based customers such as Apple and Nvidia contribute more than 70% of TSMC’s revenue, making localized production a hard requirement. However, building packaging capacity from scratch in the United States would be time-consuming and costly for TSMC. By joining forces with Amkor, which has already established a local presence, TSMC can significantly shorten the capacity ramp-up cycle and reduce trial-and-error costs. As an independent OSAT provider, Amkor does not compete directly with TSMC, making the partnership naturally complementary.

Advanced packaging is moving from an ancillary part of the supply chain to a strategic battleground. According to Yole, the global advanced packaging market is expected to surpass the traditional packaging market for the first time in 2025 and continue growing at a CAGR of 10.6% to reach USD 78.6bn by 2028. The ten-year agreement between TSMC and Amkor marks a shift in advanced packaging from standalone efforts to ecosystem alliances. In the future, deep integration between foundries and OSAT providers will become the norm. Whoever can build an integrated closed loop spanning manufacturing, packaging and testing will be better positioned to seize the initiative in the race for AI computing power. For Amkor, aligning itself with TSMC means securing access to a pipeline of orders for the world’s most advanced packaging technologies. For TSMC, Amkor’s capacity flexibility gives it greater room to respond to the explosive growth in AI demand. In essence, this partnership is a win-win arrangement in which the arrangement aligns closely with the strategic needs of both sides

02

In June 2026, Nvidia returned to the bond market for the first time in five years, issuing USD 25bn in bonds and attracting as much as USD 85bn in orders, underscoring strong market demand. This marked Nvidia’s largest financing since its USD 5bn bond issuance during the pandemic in 2021, with the size of the deal expanding fivefold.

Although Nvidia held approximately USD 13.24bn in cash and cash equivalents at the end of the first quarter of fiscal 2027, and analysts expect its full-year free cash flow to exceed USD 200bn, the company still chose to raise funds through debt issuance. The proceeds will be used in three main areas: first, to support the production of advanced AI chips and the expansion of its supply chain, as the company currently has USD 500bn in undelivered orders scheduled through 2026; second, to step up ecosystem investments, including a USD 5bn equity investment in Intel, a USD 10bn investment in Anthropic and a USD 30bn capital injection into OpenAI; and third, to optimize its capital structure, lower its average financing costs and maintain its AA credit rating.

Notably, AI-related bond issuance has become an industry-wide trend. Morgan Stanley expects global AI-related bond issuance to exceed USD 570bn in 2026, representing a sharp increase from previous levels. The spread on Nvidia’s ten-year bonds tightened from the initial price talk of 75 basis points to 50 basis points, reflecting the market’s strong confidence in the company’s credit quality.

Debt issuance gives Nvidia the financial firepower it needs to transform from a chip seller into a provider of AI factories. The company is accelerating its push into a more diversified product lineup, including Rubin-architecture chips, Vera CPUs and RTX Spark PC chips. At the same time, it is using equity investments to deepen ties with core customers such as OpenAI and Anthropic, building a “hardware + ecosystem” moat. The USD 25bn bond issuance is well aligned with the funding needs and pace of Nvidia’s ecosystem investments.

The funding needs of AI infrastructure buildout are shifting on a large scale from the equity market to the bond market. Capital expenditure by hyperscale cloud service providers is expected to exceed USD 1tn in 2027, making credit markets a core source of funding for this AI infrastructure cycle. Nvidia’s debt-funded expansion as the “picks-and-shovels” provider of the AI ecosystem reflects confidence in its own growth outlook, and also suggests that demand for AI computing power is far from peaking. When even the most profitable companies are borrowing to expand, the industry upcycle may only be entering its middle stage.

03

According to Bloomberg, Apple is planning to debut its A22 Pro chip in high-end iPhones in 2028, with the process node expected to advance from 2nm to 1.4nm. TSMC’s A14 process will serve as the primary manufacturing platform, while Apple is also considering bringing in Intel as a second supplier.

Based on Apple’s chip roadmap, the current iPhone 17 series uses TSMC’s N3P process, its third-generation 3nm technology. The iPhone 18 Pro and the foldable iPhone, expected to launch in September 2026, are likely to be the first to enter the 2nm era. Apple is expected to continue using 2nm in 2027, before some high-end chips officially move to the 1.4nm node in 2028.

TSMC’s A14 process adopts second-generation GAAFET nanosheet transistors and NanoFlex Pro technology. Compared with the 2nm N2 process, it delivers around 15% higher performance at the same power, around 30% lower power consumption at the same performance, and more than 20% higher logic density. The process is scheduled to enter production in 2028, with development progressing smoothly and yield already reaching target levels ahead of schedule.

To reduce its excessive reliance on TSMC as a single foundry source, Apple is seeking to diversify supply chain risk. Intel is pushing ahead with its 1.4nm-class 14A process, targeting risk production in 2028 and mass production in 2029. The cooperation between the two companies may begin with mid- to lower-end chips for the iPad and Mac, before potentially expanding to non-Pro iPhone chips in the future.

The 1.4nm process marks a critical milestone as the semiconductor industry moves from the nanometer era into the angstrom era. A 30% reduction in power consumption would be highly significant for mobile device battery life and the deployment of AI computing capabilities. However, the difficulty of mass-producing advanced nodes, wafer costs that are expected to exceed USD 30,000 per wafer, and tight capacity remain core challenges. With demand for AI servers surging, giants such as Nvidia are competing aggressively for TSMC’s advanced-node capacity, leaving consumer electronics with an increasingly limited share of available resources.

Apple’s move to bring Intel into its foundry supply chain carries dual strategic significance. For Apple, it would break TSMC’s exclusive foundry position, strengthening its bargaining power and supply chain resilience. For Intel, it would mark a symbolic breakthrough as its foundry business shifts from internal use to external customers. If Intel can successfully secure Apple orders, it would greatly boost market confidence in the company’s advanced process technology. However, Intel’s 14A process is scheduled for mass production in 2029, later than TSMC’s A14 in 2028, while its yield performance and capacity ramp-up still need to be proven. In the near term, TSMC’s dominant position will be difficult to shake.

04

Company Trend（June.17）

Driven by surging demand for AI computing power, TSMC is building new fabs at an unprecedented pace. Its capital expenditure in 2026 is set to reach USD 52-56bn, a record high.

In 2026, as many as ten fabs across Taiwan’s science parks could be under construction or preparing to enter production at the same time. The P3 and P4 phases of Fab 20 in Baoshan, Hsinchu, will serve as bases for sub-2nm processes. The five-phase expansion of Fab 22 in Kaohsiung is now fully underway, with all phases expected to enter production in Q4 2027. Fab 25 in Taichung is being developed into a 1.4nm process hub, with four fabs planned and mass production scheduled for the second half of 2028. In Area A of the Tainan Special Industrial Park, construction of a new 2nm fab will begin immediately after environmental approval.

TSMC’s total investment in the United States has surged from the initial USD 12bn to USD 165bn. Its plan has expanded from six fabs to as many as twelve facilities, including eight wafer fabs and four advanced packaging plants, making it the largest single foreign direct investment in U.S. history. The first fab has already entered mass production of 4nm chips, while construction of the second fab has been completed, with tool installation planned for Q3 2026 and 2nm mass production targeted for the end of 2027.

TSMC’s 2nm process entered mass production in Hsinchu and Kaohsiung at the end of 2025. In 2026, five fabs are expected to ramp up production, with monthly capacity projected to exceed 60,000 wafers and potentially approach 100,000 wafers by the end of the year. A16, or 1.6nm, is planned for mass production in the second half of 2026, while A14, or 1.4nm, is scheduled to enter production in 2028. TSMC has stressed that keeping its most advanced processes rooted in Taiwan remains a core principle.

TSMC’s aggressive capacity expansion is by no means blind growth; it is being driven by demand for AI computing power. Core customers such as Nvidia, Apple and AMD have secured capacity through long-term orders, while the price of a 2nm wafer has already exceeded USD 30,000 and continues to rise. TSMC’s advanced-node capacity is booked out through 2028. This “customer orders first, fabs built later” model effectively hedges against the cyclical risks of the semiconductor industry.

The dual-track footprint of ten fabs in Taiwan and twelve facilities in the United States both reinforces TSMC’s domestic technology base and meets customers’ demand for supply chain security through U.S.-based manufacturing. U.S. customers contribute more than 70% of TSMC’s revenue, making localized production inevitable. However, fab construction costs in the United States are two to three times higher than in Taiwan. In Q3 2025, TSMC’s Arizona fab reportedly saw quarterly profit plunge by 99% after a power outage at a gas supplier, highlighting the clear gap in operating efficiency.

TSMC’s 2026 capital expenditure accounts for more than a quarter of total global semiconductor investment, far exceeding that of Samsung and Intel. While its competitors are still struggling with yield improvement and capacity ramp-up, TSMC is using a scale-driven strategy to turn its technology leadership into a capacity moat.

05

On June 15, 2026, the ChiNext IPO application of CanSemi Technology Inc. was approved by the Shenzhen Stock Exchange’s Listing Committee, putting the company on track to become the first wafer manufacturer listed on ChiNext. The company plans to raise RMB 7.5bn through the IPO, of which RMB 3.5bn will be used for a 12-inch integrated-circuit analog specialty process production line, or Phase III project; RMB 2.5bn for the development of specialty process technology platforms; and RMB 1.5bn to replenish working capital.

Founded in 2017, CanSemi began mass production on its Phase I production line in 2019, ending Guangdong’s history of having no local 12-inch wafer fab in mass production and earning the company the title of “Guangzhou’s First Chip Company.” The company focuses on three major areas: integrated circuits, power devices and optoelectronic integration. Its process nodes range from 180nm to 55nm, and it is the only company in the Chinese Mainland with large-scale mass production capability for 12-inch silicon photonics wafers.

In terms of financial performance, CanSemi’s revenue reached RMB 1.044bn, RMB 1.681bn and RMB 2.582bn in 2023, 2024 and 2025, respectively, representing a CAGR of 57.30%. However, due to the asset-heavy nature of wafer manufacturing, its net loss attributable to shareholders of the parent company stood at RMB 1.917bn, RMB 2.253bn and RMB 2.346bn over the same period, with cumulative losses exceeding RMB 6.5bn. The company is expected to turn profitable no earlier than 2029. As of the end of May 2026, CanSemi had orders on hand for 321,200 wafers, corresponding to RMB 1.533bn in value. In the first quarter, its wafer foundry shipments reached 180,500 wafers, up 62.20% YoY.

06

On June 15, 2026, China National Nuclear Corporation (CNNC) announced that its Institute of Physical and Chemical Engineering of Nuclear Industry had achieved a key breakthrough in stable isotope enrichment and high-purity silicon preparation. For the first time, the institute has successfully achieved domestic mass production of silicon-28 isotope material with an abundance of more than 99.99%, with key product indicators reaching internationally advanced levels.

Silicon-28 is a stable isotope of silicon. Because its nuclear spin is zero, it can greatly reduce environmental noise interference in quantum computing. Known as “the purest silicon in the world,” it is an indispensable core material for silicon-based quantum chips. In natural silicon, silicon-28 accounts for only 92.2%, with the remainder made up of silicon-29 and silicon-30. Among them, silicon-29 can cause serious interference in quantum computing. The purification process is not a chemical conversion reaction; rather, it physically separates the three isotopes, much like “sorting beans,” enriching silicon-28 on one side and silicon-29 and silicon-30 on the other. The total amount remains unchanged, but the abundance of each component changes.

The team at the Institute of Physical and Chemical Engineering of Nuclear Industry that delivered this breakthrough had previously achieved production of twenty-six stable isotopes across twelve elements, including molybdenum, tellurium and nickel. This breakthrough comes in the final year of the Three-Year Action Plan for the High-Quality Development of the Nuclear Technology Application Industry (2024–2026). It marks a substantive step by China toward building an independently controllable, coordinated and efficient stable isotope industrial system, helping the nuclear technology application industry move from isolated breakthroughs to chain-based development.

07

On May 22, 2026, Unigroup Guoxin Microelectronics Co., Ltd. (002049.SZ) disclosed a draft major asset restructuring plan, proposing to acquire a 100% equity interest in WeEn Semiconductors for RMB 1.9bn. The consideration includes RMB 1.52bn to be paid in shares and RMB 380mn in cash, while the company also plans to raise no more than RMB 396mn in supporting funds. On June 8, the plan was approved at the company’s shareholders’ meeting; on June 16, the Shenzhen Stock Exchange formally accepted the relevant application documents.

WeEn Semiconductors is a rare IDM power semiconductor company in China. It owns two wafer fabs in Jilin and Beijing as well as a module plant in Shanghai, with integrated capabilities spanning chip design, wafer manufacturing, packaging and testing. Its core thyristor products hold a leading global market share, while its silicon carbide diode products are among the domestic leaders. In 2024 and 2025, the company recorded revenue of RMB 784mn and RMB 871mn, respectively, with net profit of RMB 19.19mn and RMB 47.15mn. Its six-inch automotive-grade wafer fab in Beijing is expected to enter production in the second half of 2026, further unlocking capacity.

Unigroup Guoxin Microelectronics had previously focused on special-purpose ICs and smart security chips, while its power semiconductor business operated under a fabless model, leaving a clear gap in manufacturing. This acquisition will quickly fill that gap in power semiconductor manufacturing and help the company build a diversified business portfolio covering “special-purpose chips + power devices + quartz devices.” In Q1 2026, Unigroup Guoxin Microelectronics generated revenue of RMB 1.499bn, up 46.11% YoY, while net profit attributable to shareholders of the parent company reached RMB 334mn, up sharply by 180.27% YoY and marking its highest quarterly level since 2022.