ABF Substrates: Shaping AI Data Center Supply
This guide explains how ABF (Ajinomoto Build-up Film) substrates, a critical component in advanced semiconductor packaging, are reshaping the supply chain for AI data centers. You’ll learn the key factors driving demand, how supply is responding, and what it means for the industry. No prior knowledge of semiconductor manufacturing is required, but a basic understanding of computer hardware is helpful.
What You’ll Need
- A basic understanding of what a CPU/GPU is and why they’re used in AI.
- Familiarity with the concept of a supply chain (raw materials → manufacturing → end product).
- An interest in the business/technology of data centers.
Step-by-Step
- Understand what ABF substrates are. ABF (Ajinomoto Build-up Film) is a dielectric film used to build the insulating layers of a substrate—the package that connects a chip to a printed circuit board (PCB). In AI accelerators (like GPUs), the substrate must handle massive numbers of connections and high-speed signals. ABF substrates are the high-end, multi-layer boards that make this possible.
- Recognize the AI demand driver. AI workloads, especially training large language models, require enormous parallel processing power. This is delivered by specialized chips (GPUs, TPUs, custom ASICs) that are physically large and have very high pin counts. Each of these chips requires a large, complex ABF substrate—often much bigger than those used in consumer PCs. So, as AI data centers grow, they directly consume a disproportionate share of the world’s ABF substrate production.
- Identify the supply chain bottleneck. ABF substrate manufacturing is not like making standard circuit boards. It requires extreme precision, cleanroom environments, and advanced equipment. Only a few companies in the world (primarily in Taiwan, Japan, and South Korea) have the capability to produce the high-layer-count substrates needed for AI chips. This concentration creates a bottleneck: even if chip production increases, the number of AI accelerators that can be packaged is capped by substrate supply.
- Observe the capacity expansion race. In response to AI demand, the major substrate makers have announced massive capital expenditure programs to build new plants and expand existing ones. This is a multi-year process, as these factories are complex to build and qualify. Watch for announcements from companies like Ibiden, Shinko Electric, Unimicron, and AT&S regarding new capacity.
- Monitor material supply constraints. ABF film itself is produced primarily by Ajinomoto, and its production is also being expanded. However, other materials like copper foil, glass fiber, and specialty chemicals are also needed. Any disruption in these upstream materials will directly affect substrate output.
- Track the shift to advanced packaging. AI chips often use advanced packaging technologies like CoWoS (Chip-on-Wafer-on-Substrate) from TSMC. This process places multiple chips (logic, memory) on a silicon interposer, which then sits on a large ABF substrate. This increases the size and layer count of the substrate required, further straining supply. Understanding this integration is key to seeing why substrate demand is growing faster than just chip volume.
- Analyze the impact on data center timelines. The lead time for ABF substrates is now a critical factor in how quickly AI data centers can be built. If substrate supply is tight, it can delay server shipments and thus the deployment of new AI capacity. This is why cloud providers and AI companies are increasingly signing long-term supply agreements directly with substrate makers.
- Watch for pricing signals. When demand exceeds supply, prices rise. ABF substrate prices have been on an upward trend, and this is passed on to the cost of AI accelerators and servers. This is a clear indicator of the supply-demand imbalance.
- Consider the geographic risk. Most ABF substrate production is concentrated in East Asia. Any geopolitical disruption, natural disaster, or pandemic-related lockdown in that region can have an outsized impact on global AI data center supply. This is a major reason why some governments are trying to encourage domestic substrate manufacturing.
- Look at the long-term outlook. The substrate industry is investing heavily, and new capacity is expected to come online in the next 2-3 years. However, AI demand is also growing exponentially. The supply-demand gap is likely to persist for the foreseeable future, meaning ABF substrates will remain a strategic bottleneck for AI infrastructure.
Common Mistakes
- Confusing ABF substrates with PCBs. A substrate is the chip’s package, not the motherboard. They are different products with different manufacturing processes.
- Assuming all substrates are equal. The substrates used in AI chips are far more complex (more layers, larger size) than those in phones or laptops. The supply constraint is specifically for these high-end substrates.
- Overlooking the lead time. Building a new substrate plant takes years, not months. The supply response is inherently slow.
- Focusing only on chip makers. While TSMC, Intel, and Samsung get the headlines, the substrate makers are equally critical to the AI supply chain.
Tips
- Follow industry analysts. Firms like TrendForce and IC Insights regularly publish reports on substrate supply and demand. These are valuable for staying current.
- Watch quarterly earnings calls. Substrate makers and chip foundries often discuss capacity and demand outlooks in their earnings calls. These are public and can be found online.
- Understand the technology roadmap. As AI chips get larger and faster, they will require even more advanced substrates. This will keep the pressure on supply.
- Consider the ecosystem. ABF substrates are just one part of the AI hardware chain. Memory (HBM) and power delivery are also bottlenecks, so it’s important to see the whole picture.
Troubleshooting
- If you can’t find recent data on substrate supply: Check the financial reports of the major substrate manufacturers. They are publicly traded and must disclose capacity and investment plans.
- If you’re confused about the difference between substrate and interposer: Think of the interposer as a silicon bridge that connects chips, and the substrate as the larger organic base that connects the interposer to the PCB. Both are needed for AI accelerators.
- If you’re unsure why ABF is used instead of cheaper alternatives: ABF allows for very fine lines and spaces (high wiring density) and good electrical performance at high frequencies. This is essential for AI chips that need to move huge amounts of data quickly.
- If you’re wondering if the supply issue is temporary: It’s not likely to be resolved quickly. The capacity expansion is happening, but the demand curve is steep. Expect the tightness to continue for several years.
Frequently Asked Questions
What does ABF stand for?
ABF stands for Ajinomoto Build-up Film, named after the Japanese company that invented the material.
Why are ABF substrates so important for AI?
AI chips are extremely complex and have very high pin counts. ABF substrates provide the high-density, high-speed wiring needed to connect these chips to the rest of the system, making them a critical component in AI accelerators.
Who are the main ABF substrate manufacturers?
The leading manufacturers are Ibiden (Japan), Shinko Electric (Japan), Unimicron (Taiwan), and AT&S (Austria). These companies control the vast majority of the market for high-end substrates.
How long does it take to build a new ABF substrate plant?
It typically takes 2-3 years to build and qualify a new plant. This is why the supply response to AI demand is slow.
Will the ABF substrate shortage ever end?
It’s likely that supply will eventually catch up with demand, but the timeline is uncertain. As AI technology evolves, the demand for even more advanced substrates will continue to grow, so the industry will need to keep investing.
How does ABF substrate supply affect the price of AI servers?
When substrate supply is tight, prices rise, and this cost is passed on to server manufacturers and ultimately to data center operators. This can influence the overall cost of AI infrastructure.
Related
Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.