I'm LongbridgeAI, I can summarize articles.With the rise of AI agents, a comprehensive restructuring of computing power has begun, and this trend has not only affected GPU supply but has also spread to the server CPU market, igniting a century-long shortage crisis for high-end Ajinomoto Build-up Film Substrate (ABF). As the trend of increasing chip size and packaging layers continues, the pricing power of the entire industry has clearly shifted to a seller-dominated market. Those who can master process barriers and complete strategic alliances will have the opportunity to earn double long-term profits in this bullish market.
Industry insiders point out that the nature of the current ABF substrate market is fundamentally different from the traditional economic cycles driven by consumer electronics in the past.
In a framework where multiple AI agents operate collaboratively, CPUs are playing a core role in tool scheduling and task control. Research institutions estimate that the share of PCs in the overall ABF consumption structure will plummet from 61% in 2020 to only 10% by 2028; conversely, the combined share of AI chips and server CPUs is expected to rise to 85%.
It is estimated that between 2026 and 2028, the supply-demand gap for ABF will sequentially widen to 8%, 27%, and 35%, indicating that this bullish pattern can last at least another 4 to 5 years.
Supply-side warnings emerge, raw material bottlenecks extend to 2027
From the supply side, warnings of limited production capacity are continuously emerging. Global ABF film leader Ajinomoto has indicated that it will adopt a cost-pass-through strategy to cope with the surge in demand; Japanese substrate giant Ibiden has also announced a significant increase in capital expenditure to 210 billion yen for the 2026 fiscal year, stating that the demand area for high-end ABF in 2028 will soar 10.7 times compared to 2024.
As the number of substrate layers continues to rise to 16 to 24 layers, the supply bottleneck of key raw material T-glass fiberglass cloth is expected to last at least until 2027.
Regarding the glass substrate technology that the market is concerned about, due to the difficulty in breaking through the yield rates of key copper and lamination in the short term, the industry generally believes that large-scale mass production will be difficult to achieve before 2030, resulting in relatively limited actual impact on existing ABF demand.
In the context of a foreseeable long-term supply-demand gap, the bargaining power and cost-pass-through ability of substrate manufacturers have significantly increased.
Currently, the delivery time for high-end orders has been extended to 1.5 to 2 years, and customers are even willing to provide long-term financial assistance to manufacturers to secure production capacity, greatly releasing the operational leverage of related Taiwanese supply chain manufacturers and officially opening up long-term profit space.
Essential packaging materials in the AI era
In the era of AI and high-performance computing (HPC), chip packaging technology has become a key factor determining system performance. With the explosive growth in the number of I/O ports on computing chips and data transmission speeds reaching levels of 112Gbps and even 224Gbps, traditional PCBs and BT substrates can no longer meet the demand, which is the reason for the rise of ABF substrates ABF is not only an important foundation for advanced packaging but also one of the bottlenecks in the global semiconductor supply chain, with strategic importance that is comparable to the most advanced chip manufacturing processes themselves.
ABF substrates are located between the chip and the system motherboard, acting as a bridge for high-speed signal transmission and power distribution, commonly found in FC-BGA packaging.
Structurally, the chip is soldered to the ABF substrate via micro bumps, and then connected to the PCB motherboard through BGA solder balls.
Its core functions can be summarized into four points:
- Signal fan-out and amplification: Converts the extremely dense signal pins (with a spacing of only 40 to 55 micrometers) inside the chip into a spacing acceptable by the PCB of 0.4 to 1.0 millimeters.
- High-speed signal transmission: Supports SerDes high-speed transmission requirements of 112Gbps or even higher.
- Power integrity management: Provides a stable and low-noise power network for chips with power consumption often reaching hundreds of watts.
- Mechanical protection: Mitigates the differences in thermal expansion and contraction between materials, preventing chip solder joints from breaking due to stress.
The industry often describes ABF's role as a "highway interchange." The chip is the "high-speed data center," the PCB is the "city road," and ABF is the key interface that allows data to flow smoothly in a high-speed environment.
Why can't chips be directly soldered to PCBs?
On the surface, both ABF substrates and PCBs are "a piece of board," but their structures and processes are actually quite different.
Taking NVIDIA's (NVDA-US) B100 GPU as an example, its chip I/O count exceeds 100,000, and PCBs simply cannot achieve such dense wiring and micron-level solder joints, so it must first go through ABF for "fan-out" processing.
Additionally, ABF materials possess low dielectric constant (Dk) and low loss factor (Df) characteristics, ensuring high-speed signal integrity and avoiding signal delay and distortion.
Without this ABF link, the chip would struggle to operate stably, with signals severely attenuated during transmission, making high-speed computing impossible.
Currently, AI and HPC chips almost entirely rely on ABF substrates. Although glass substrates are seen as a potential alternative for the future, they still require years of research and validation before they can be commercialized.
Ajinomoto's exclusive monopoly as the world's only supplier
ABF stands for Ajinomoto Build-up Film, which was developed by Japan's Ajinomoto Company in the 1990s. This film material has three major advantages:
- Low dielectric constant (Dk around 3.0), beneficial for high-speed signal transmission;
- Low loss (Df around 0.004), which can reduce signal attenuation;
- The thermal expansion coefficient is close to that of silicon chips, preventing solder joints from failing due to fatigue It is worth noting that Ajinomoto is the only manufacturer in the world producing ABF film, with a market share close to 100%. In other words, all ABF substrate manufacturers globally must rely on the materials provided by Ajinomoto.
Actual Application Scenarios: From AI GPUs to Optical Communications
In the packaging of AI GPUs, taking NVIDIA's H100 as an example, the silicon interposer is responsible for the ultra-high-density connection between the GPU and HBM memory, while the ABF substrate is responsible for amplifying and transferring the signals of the entire module and providing stable power supply.
In the field of Co-Packaged Optics (CPO), 400G/800G optical communication chips also require ABF to play the role of a bridge between high-speed electrical signals and optical modules to ensure extremely low latency transmission performance.
The ABF industry currently faces two major challenges. The first is the capacity bottleneck. The AI boom has driven a surge in demand from companies like NVIDIA and Advanced Micro Devices (AMD-US), but the expansion cycle for ABF production takes 2 to 3 years, and the supply-demand imbalance has made it a critical position at the forefront of the global AI supply chain.
The second challenge is the approaching technical limits. The line width has been reduced to the 2-micron level, which imposes extremely stringent requirements on lithography and etching yield. Any slight fluctuation in production line yield could lead to significant supply shortages.
As for the potential alternative technology of glass substrates, while they offer better dimensional stability and lower thermal expansion coefficients, theoretically, they have the opportunity to replace some ABF applications, but they are still in the verification stage, and the industry estimates that it will take at least 3 to 5 years to reach mass production scale