AI computing chain, Japan bottlenecks from start to finish (old yin b)

AI computing power isn’t just about GPUs; the real bottlenecks might be a pile of the most inconspicuous materials and equipment. NVIDIA can design the strongest chips, TSMC can do advanced packaging, but if upstream physical materials and equipment like electronic cloth, ABF film, glass substrates, and thinning machines can’t keep up, AI servers still can’t be built. More critically, many of these links have long been monopolized by Japanese companies.

1. Electronic cloth: It is the skeleton material for copper-clad laminates and PCBs. The higher the grade of an AI server motherboard, the higher the requirements for thinner, lower-loss, and more stable electronic cloth. The core equipment for high-end electronic cloth—Toyota air-jet looms—is in severe short supply, with orders backlogged to 2028, creating a real capacity bottleneck.

2. ABF film: It is a key insulating material in high-end chip packaging substrates. Ajinomoto holds over 95% of the global market share. GPUs, CPUs, and AI chips all depend on it. Even if glass substrates ramp up in the future, they won’t directly replace ABF; instead, higher layer counts may further increase its usage.

3. Japan’s eight choke points: Toyota Industries looms, Nittobo electronic cloth, Ajinomoto ABF film, Admatechs spherical silica powder, DISCO thinning machines, Screen exposure equipment, Sumitomo/Fujikura polarization-maintaining fiber, and SABIC/MGC high-frequency resin collectively form the “invisible valve” of the AI computing chain.

4. Three domestic substitution lines: International Composite Materials targets low-dielectric electronic cloth, Honghe Technology targets ultra-thin electronic cloth, and Woge Optoelectronics targets TGV glass substrates. These are not just vague talks about domestic substitution; they are catching up with Japanese and global leaders in specific materials, specific processes, and specific production capacities.