ABF is being snapped up!

Recently, NVIDIA announced partnerships with supply chain partners including Ibiden, Samsung Electro-Mechanics, Unimicron, Amkor, KYEC, and SPIL to expand related manufacturing capacity. Among them, product substrate capacity is planned to increase to 2.5 times the current level over the next three years; Amkor plans to increase interposer wafer capacity by more than 300% by the end of 2028; and SPIL plans to increase capacity for optical engines used in co-packaged optics (CPO) to 15 times the current level by mid-2027. With the release of this news, the continued expansion of AI computing infrastructure is igniting supply chain competition that is spreading from core chips such as GPUs further into key areas including advanced packaging, testing, and high-speed optical interconnect. Among these, ABF has become a fiercely contested battleground for major chipmakers. ABF lead times continue to lengthen In fact, as everyone can see, supply pressure on ABF substrates is intensifying further. According to AI infrastructure key component lead time data compiled by TrendForce, ABF substrate lead times have lengthened to 48 to 56 weeks, equivalent to 4 to 4.7 times the 12-week baseline under supply-demand balance. In other words, a new order may take up to about 13 months from placement to delivery. By comparison, HDD lead times are 50 weeks, DRAM 20 weeks, and MLCC 30 weeks, illustrating the tightness of the AI infrastructure supply chain. ABF substrates are important connection carriers between high-performance chips and motherboards, widely used in products such as CPUs, GPUs, and AI accelerators. As AI chips move toward higher performance, larger package sizes, and more complex interconnects, substrates need to support higher-density circuitry and more layers of routing, increasing manufacturing difficulty. At the same time, AI servers are driving synchronized growth in demand for components such as memory, networking, and power management, so supply chain pressure is no longer confined to any single segment. TrendForce data also shows that among the six categories of AI infrastructure key components it tracks, all except GPUs are in shortage. Among them, ABF substrates, HDDs, and DRAM are listed as severely short, while MLCCs and enterprise SSDs also face insufficient supply. Even if GPU lead times are within the supply-demand balance range of 20 to 30 weeks, shortages of other key components may still slow the production and delivery of complete AI servers. The problem is that new capacity cannot immediately fill the gap. Expanding high-end substrate lines involves plant construction, equipment procurement, process tuning, and yield improvement; capital investment does not mean capacity can be released quickly. Current AI demand is growing rapidly, while suppliers' earlier expansion plans may not keep pace in time, so supply-demand mismatch persists. More noteworthy is that constraints on supply also extend to semiconductor equipment components. Reports indicate that lead times for some key equipment components made in Japan reach as long as 40 months; some deposition equipment parts that could originally be delivered in about 4 months now require about 10 months of waiting. For some packaging equipment components, even after shifting to domestic supply in South Korea, lead times are still about 50% longer than in the past. This forms a worrying cycle: growing AI chip demand drives expansion of substrates, memory, and other components; but expansion itself requires semiconductor manufacturing equipment, and equipment makers are likewise affected by shortages of key parts. The stronger downstream demand is, the more the supply chain needs to expand capacity, yet the more it may be constrained by upstream equipment delivery cycles. In addition, reports note that factors such as increased capital expenditure, slower gains in production efficiency, and HBM production consuming more wafer input are also putting pressure on supply expansion in the memory industry. The current tight supply cannot be solved immediately by simply increasing investment; it involves capacity, equipment, and manufacturing efficiency across multiple segments. Lengthening lead times are also changing the way the supply chain transacts. Reports indicate that prices for high-spec FC-BGA substrates face upward pressure, and the importance of long-term supply agreements is rising. Customers hope to reduce future shortage risk by locking in volumes in advance and signing long-term contracts; suppliers with stable capacity and on-time delivery have opportunities to gain stronger bargaining power. Take Samsung Electro-Mechanics as an example. Reports indicate that utilization rates at its FC-BGA substrate and MLCC production facilities are already near full load. When capacity is difficult to expand quickly, manufacturers with stable delivery capability will find it easier to secure long-term orders and increase the revenue share of high-value-added products. For the ABF industry, what truly deserves attention is no longer just how many weeks lead times have reached, but when new capacity can come online and whether it can match the continuously upgrading specification requirements of AI chips. If demand continues to grow while equipment, materials, and manufacturing capabilities cannot improve in tandem, tight supply and demand for ABF substrates may persist for longer. However, lead time data reflects the current state of the supply chain and does not mean lead times are exactly the same for all specifications and all customers. Under such demand conditions, how to ensure ABF supply has become the top priority for major chip companies. This has also triggered a capacity scramble. Three chip giants lock in capacity According to Economic Daily News, the three major chipmakers Intel, NVIDIA, and AMD are reported to have paid deposits totaling more than NT$100 billion to book high-end IC substrate capacity in advance. At the same time, Taiwan's top three substrate makers Unimicron, Kinsus, and Nan Ya PCB are accelerating expansion, with announced plant investment totaling NT$161.2 billion this year, a record high. Behind this capacity contest over ABF substrates is the reality of continued growth in AI chip demand and the difficulty of rapidly expanding high-end substrate capacity in the short term. For chipmakers, GPU and AI accelerator performance is certainly important, but whether products can be successfully packaged, tested, and ultimately delivered also depends on whether the upstream supply chain can provide enough capacity. When high-end substrate supply tightens, locking in capacity in advance becomes an important means of ensuring product delivery. According to industry estimates cited in reports, Intel's long-term prepurchases of substrate capacity amount to about US$3 billion. AMD previously announced a US$10 billion investment in Taiwan, and the market speculates that about US$1 billion of that may involve the substrate supply chain. NVIDIA is also speculated by the market to have invested about US$1 billion in Taiwan substrate makers, with further increases possible in the future. Based on these estimates, the three companies' funding related to the substrate industry totals about US$5 billion. However, these figures are industry rumors or market estimates, not officially confirmed order and prepayment amounts by the companies. Customers paying prepayments in advance to lock in capacity also gives substrate makers momentum to expand. Building high-end ABF substrate lines requires substantial capital input, and equipment procurement, plant construction, and subsequent process tuning all take time. For chipmakers, laying plans early helps reduce uncertainty over future supply; for substrate makers, growing customer demand provides the basis for expansion. Taiwan's three major substrate makers have already begun increasing capital investment. Unimicron added NT$19.7 billion in capital expenditure, bringing total capex this year to NT$53.7 billion, and also added NT$17.6 billion in long-lead-time equipment purchase orders. The company also acquired Bridgestone's Hukou plant office for about NT$10.08 billion to prepare for medium- and long-term operations and expansion of high-end substrate capacity. Kinsus, meanwhile, continues to transition toward high-end substrates and is gradually phasing out traditional PCB production. Through NT$23.5 billion in capital expenditure, the company will meet ABF substrate equipment expansion needs over the next three years, with plans to spend NT$8 billion in 2026, about NT$6 billion of which will be for ABF substrate investment. In addition, the company added a NT$19.6 billion special capital expenditure budget. Nan Ya PCB also plans to invest NT$46.8 billion to build a smart factory producing large-size, high-layer-count high-end substrates. The expansion directions of the three companies reflect how AI chip demand is pushing the substrate industry into a new investment cycle. However, expansion does not mean tight supply can be eased immediately. Plant construction, equipment delivery, process tuning, and yield improvement all take time, and whether new capacity can meet demand in time also depends on customer product specifications and mass production schedules. From chip giants locking in capacity in advance to substrate makers sharply raising capital expenditure, the impact of AI infrastructure buildout is gradually transmitting upstream through the supply chain. As high-end chips place ever higher demands on packaging capability, ABF substrates are no longer just supporting materials in chip packaging; their supply capability is also becoming an important factor affecting AI chip production and delivery. Samsung Electro-Mechanics is being courted As AI chip giants race to lock in ABF substrate capacity, Samsung Electro-Mechanics is emerging as an important player in supply chain competition. On October 21, Samsung Electro-Mechanics announced a strategic partnership with NVIDIA under which the two sides will jointly invest in expanding dedicated production facilities in Sejong and Busan, South Korea, to supply ABF packaging substrates for NVIDIA's next-generation AI processors and high-speed network switches for data centers. Samsung Electro-Mechanics plans to begin supplying in 2026, with supply volume expected to grow at a double-digit rate annually for many years; by 2028, capacity allocated to NVIDIA will account for a considerable portion of the company's total annual substrate capacity. This means the cooperation between NVIDIA and Samsung Electro-Mechanics is not an ordinary supplier purchase, but a combination of capacity expansion with long-term supply demand. For NVIDIA, AI infrastructure buildout requires continuous expansion of chip and data center networking equipment supply, and locking in high-end substrate capacity in advance helps reduce the impact of supply chain bottlenecks on product delivery. Debora Shoquist, NVIDIA's executive vice president of operations, also said that the pace of AI infrastructure buildout requires the company to expand manufacturing capacity globally, and cooperation with Samsung Electro-Mechanics will further expand its substrate supplier network. For Samsung Electro-Mechanics, this represents an important opportunity to enter the high-value-added AI semiconductor substrate market. The company will set up dedicated production lines at its main production bases in Sejong and Busan, South Korea, supplying not only packaging substrates required for NVIDIA computing chips but also substrates for network switches that enable high-speed chip interconnect within data centers. As AI computing and high-speed networking develop in parallel, substrate makers need to simultaneously meet the high-density interconnect requirements of computing chips and networking equipment. The real challenge lies in technology. Samsung Electro-Mechanics noted in its announcement that as packaging substrate sizes expand and embedded devices increase, manufacturing complexity also rises. To reduce signal loss, substrates need ultra-fine circuitry; to ensure mass production yield and long-term reliability, advanced warpage control and flatness management capabilities are also required. In addition, the high-speed operation of next-generation AI chips places higher demands on power supply stability. Samsung Electro-Mechanics plans to leverage its technical accumulation in multilayer ceramic capacitors (MLCCs) and silicon capacitors to embed passive components into substrate structures, thereby suppressing electrical noise generated by high-speed computing and improving power management efficiency. This approach combining packaging substrate manufacturing capability with capacitor technology is also one of Samsung Electro-Mechanics' competitive advantages that distinguishes it from pure substrate suppliers. This cooperation shows that competition in the AI chip industry is extending into packaging and interconnect segments. GPU performance is important, but whether chips can be stably mass-produced also depends on whether substrates can meet requirements for larger sizes, higher-density circuitry, and stricter reliability. NVIDIA locks in capacity in advance, while Samsung Electro-Mechanics uses long-term supply cooperation to expand its high-end substrate business, with each side getting what it needs. As AI infrastructure continues to expand, ABF substrates have become a link in the chip supply chain that cannot be ignored. Whether Samsung Electro-Mechanics can turn this cooperation into a long-term competitive advantage will still depend on the construction progress of dedicated production lines, mass production yield, and whether it can continuously meet NVIDIA's rising product requirements. Final thoughts If we shift our view away from ABF substrates themselves, we can see that this capacity scramble reflects a change in the logic of competition in the AI industry: in the past, the industry focused more on who could design stronger chips and who could advance advanced processes first; now, as the scale and complexity of AI chips continue to rise, the factors determining competitive outcomes are expanding from single-chip performance to the coordination capability of the entire manufacturing system. This also means the supply chain can no longer simply understand expansion as adding plants and purchasing equipment. For key segments such as ABF substrates, real competitiveness comes from long-term among materials, equipment, processes, yield, and customer demand. Even if capital investment continues to increase, if key equipment delivery, technology accumulation, and talent reserves cannot keep pace, new capacity may not effective supply in time. For chipmakers, locking in capacity in advance can ease immediate supply pressure, but it cannot replace the supply chain's own capability building; for substrate makers, the high-growth opportunity brought by AI is worth seizing, but they also need to be wary of equating short-term demand directly with long-term growth. AI infrastructure investment will eventually go through cyclical changes, and how to strike a balance between expansion speed and investment returns will test every supplier's judgment. From a longer-term perspective, what the ABF industry truly needs is not just more capacity, but a supply system that is more resilient and better able to adapt to technological iteration. As chip design, advanced packaging, and substrate manufacturing become increasingly closely linked, suppliers also need to move from passively taking orders to participating earlier in customers' technology planning and product development. In the AI era, what is scarce has never been just a certain material or a certain technology, but the ability to truly connect technology, manufacturing, and scaled delivery. The frenzy over ABF is just one facet of this industrial competition that has surfaced. This article is reprinted from the "Semiconductor Industry Observation" WeChat public account. GMTEight editor: Feng Qiuyi.