ABF Substrates: Shortages to Last Until 2028

The AI chip boom is reshaping the semiconductor supply chain. Behind high-performance computing chips such as GPUs, ASICs, and CPUs, a key material previously unfamiliar to the general public—ABF (Ajinomoto Build-up Film) substrate—is becoming an "invisible bottleneck" in the expansion of AI infrastructure.

As demand for generative AI, large model training, and AI inference explodes, NVIDIA, AMD, and global hyperscalers continue to expand their purchases of AI servers. The rapid rise in demand for advanced chip packaging is increasingly straining the capacity of IC substrates, a core component of advanced packaging.

According to Digitimes, citing industry insiders, even though NVIDIA, AMD, and major cloud service providers have locked in ABF substrate capacity until 2028 through Long Term Agreements (LTAs), the market still struggles to fully alleviate anxiety over material shortages.

In fact, for AI chips, ABF substrates have become an irreplaceable key link.

Traditional consumer electronics chips typically use smaller packages with relatively limited performance requirements for substrates. However, high-performance chips like AI GPUs and AI accelerators feature continuously expanding die areas, increasing transistor counts, and the need to connect more high-speed I/O interfaces. Consequently, package sizes are growing, and the number of packaging layers is steadily increasing.

ABF substrates serve as the crucial bridge connecting chip dies to motherboards.

Simply put, while the chip itself handles computation, high-speed transmission of tens of thousands or even hundreds of thousands of signal and power channels relies on advanced packaging substrates for electrical connections. Compared to traditional BT substrates, ABF substrates offer finer lines, higher layer counts, and superior electrical performance, making them the mainstream choice for packaging high-end chips such as CPUs, GPUs, AI accelerators, and network processors.

This is especially true in the AI era, where single GPU packages are becoming increasingly complex.

Taking NVIDIA's latest generation of AI GPUs as an example, the package not only includes the GPU die but also integrates HBM (High Bandwidth Memory), achieving high-speed interconnects through advanced packaging technologies. Such packaging places far higher demands on substrate area, line density, and signal integrity than previous server CPUs, further driving up demand for ABF substrates.

For this reason, some customers have begun requesting IC substrate manufacturers to plan capacity expansions for 2029 to 2030 in advance. Meanwhile, to mitigate the risks of massive capital expenditures, the "co-investment model," which was prevalent from 2021 to 2022, is re-emerging.

The so-called co-investment model involves chip manufacturers or end customers directly participating in supply chain investments—for example, by paying equipment deposits in advance, entrusting key production equipment, or even sharing factory construction costs—in exchange for priority supply guarantees in the future.

This model appeared frequently during the previous semiconductor supply shortage cycle and is now being adopted again as the AI industry enters a phase of rapid expansion.

Currently, the ABF and BT substrate industries have fully returned to a seller's market. To avoid future supply constraints, major customers have been actively signing long-term orders since 2026, locking in capacity for the coming years.

However, even with order visibility extending to 2028, not all customer demands can be met. This has even forced the industry to restart joint venture expansion models, reflecting high market concern about future supply gaps.

AI Drives ABF Demand into a New Cycle

The IC substrate industry has experienced significant cyclical fluctuations in the past.

The previous global tightness in IC substrate supply and demand mainly occurred from 2021 to 2022. At that time, the pandemic-driven stay-at-home economy spurred rapid growth in demand for consumer electronics like PCs and game consoles. Simultaneously, the 5G upgrade cycle, High-Performance Computing (HPC), and early AI application demands rose in tandem, causing widespread shortages across the semiconductor supply chain.

In this context, chip manufacturers rushed to lock in IC substrate supplies through advance payments, long-term orders, and even co-investments.

Many manufacturers at the time believed demand growth would continue, leading to large-scale capacity expansions. However, starting in the second half of 2022, the consumer electronics market cooled rapidly, customer inventory pressures increased, and the semiconductor industry entered a multi-year inventory adjustment cycle.

A bigger issue was the long lead time for IC substrate capacity expansion.

From factory construction and equipment installation to process validation and customer certification, the process typically takes 18 months or longer. Therefore, new capacity planned by substrate manufacturers in 2021 and 2022 was largely released in 2022 and 2023, ultimately leading to oversupply and causing IC substrate manufacturers in Taiwan to experience nearly three years of downturn.

Today, AI-driven demand growth is pushing the industry back into an expansion cycle.

However, unlike the previous cycle, supply chain companies believe that while AI infrastructure investment may slow down in the future, it will not see a sudden collapse in demand similar to that seen in consumer electronics.

Furthermore, this round of expansion is largely based on customer orders, advance payments, and revenue guarantee mechanisms, reducing the risk of blind capacity expansion for substrate manufacturers. Therefore, the supply chain generally believes that a repeat of the previous overcapacity scenario is unlikely.

Taiwanese IC Substrate Manufacturers Expand Capacity Across the Board

Facing structural demand growth driven by AI, major global IC substrate manufacturers have launched a new round of capital investment.

Unimicron, Taiwan's largest IC substrate manufacturer, has increased its 2026 capital expenditure to NT$53.7 billion, focusing on the construction of its Guangfu Plant 2 and Yangmei Plants 2 and 3 to strengthen high-end ABF substrate capacity.

Kinsus has also added NT$19.6 billion in capital expenditure, initiating the construction of Yangmei K6C and K6D plants, and beginning to search for land for new K7 and K8 plants to prepare for future AI server demand.

Nanya PCB's capital expenditure is also rebounding. In addition to investing in advanced capacity at its Shulin plant and upgrading processes at its Jinxing plant, it plans to lease southern facilities from its parent group to expand high-end substrate supply capabilities.

In addition, Avaryn-owned Lidin is actively expanding production in Shenzhen, expecting to increase the number of factories to more than seven by 2030 to meet future high-end packaging demands.

Notably, AMD CEO Lisa Su recently visited Taiwan and announced plans to expand investment in the region by over $10 billion. Through procurement commitments and strategic cooperation, she aims to promote the establishment of an "EFB" (Elevated Fanout Bridge) 2.5D packaging ecosystem in Taiwan.

EFB is regarded by the industry as an important alternative route to TSMC's CoWoS, potentially becoming a second supply path for advanced AI chip packaging in the future.

Unimicron, Kinsus, and Nanya PCB are all included in the list of partners, indicating that this substrate expansion is not merely a bet on the market but is driven by demand from AI chip customers.

T-glass Shortage Further Limits Expansion

In addition to rapid demand growth, supply-side constraints are exacerbating the tightness in ABF substrates.

Industry insiders point out that one of the biggest current challenges comes from upstream key materials—T-glass fiberglass cloth.

T-glass is an ultra-low thermal expansion, high-performance glass fiber material and is one of the important materials for manufacturing high-end IC substrates. As AI chip package sizes continue to expand, requirements for material stability are increasing, driving rapid growth in T-glass demand.

However, the speed of T-glass supply expansion is far below market demand, meaning that even if IC substrate manufacturers have the equipment, they may face material shortages.

Meanwhile, high-end ABF substrate production involves multiple complex steps such as exposure, circuit fabrication, plating, and lamination, requiring long-term accumulation of process experience. Simply adding equipment does not quickly increase output.

Therefore, the speed of supply expansion in the IC substrate industry is naturally limited.

ABF and BT Substrates Enter a Price Hike Cycle

Starting in the second half of 2025, continued growth in demand for high-end chips such as GPUs, CPUs, and ASICs first pushed high-end ABF substrates into a state of supply falling short of demand.

Due to tight ABF capacity, resources are prioritized for AI chip packaging, further squeezing the supply space for traditional BT substrates and causing BT substrates to simultaneously enter a tight cycle.

As the supply-demand imbalance continues to widen, prices for both ABF and BT substrates have entered a rising cycle.

The market expects that price increases in the second half of 2026 will be higher than in the first half, with average quarterly quotes potentially rising by about 8%.

Previously, the market believed that since ABF is mainly used for high-end AI chips while BT substrates have broader applications, BT substrates might face greater pricing pressure.

However, as AI server demand continues to concentrate on ABF substrates, widening the supply-demand gap, the rate of price increases for ABF substrates may actually exceed that of BT substrates in the future.

In a sense, ABF substrates are becoming a new "strategic resource" in the AI era.

In recent years, competition in the AI industry has mainly revolved around GPUs, HBM, and advanced packaging. However, as the scale of AI servers continues to expand, supply chain bottlenecks are gradually shifting to underlying materials and components.

For AI infrastructure, GPUs determine computing power, HBM determines data throughput, and ABF substrates determine whether these advanced chips can truly achieve scaled delivery.

Therefore, against the backdrop of the continuing AI wave, the tight supply and demand for ABF substrates may persist until 2028 or even longer.

Risk Warning and Disclaimer

The market carries risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment decisions made based on this content are the sole responsibility of the investor.