TechGlobal.news

Japan’s ABF substrate ramp starts easing AI chip bottleneck as Ajinomoto, Ibiden, Shinko add lines

Kai Tanaka ·

Ajinomoto, Ibiden, and Shinko Electric Industry have begun ramping new Ajinomoto Build up Film substrate capacity in Japan this quarter, according to company disclosures and Japanese trade press, easing one of the biggest constraints in AI accelerator production. The additions target higher layer count packages used in GPUs for Nvidia and AMD, with initial volumes filtering into supply chains in Japan and Taiwan in November. Procurement managers in Taiwan cited shorter queue times for high layer ABF panels compared with the first half of the year.

The timing matters because substrate availability, not wafer output, has capped shipments of high end AI parts in 2025 and 2026. Any incremental relief on ABF panels feeds directly into CoWoS and other advanced packaging throughput, which remains capacity constrained at TSMC and several OSATs.

Ajinomoto Fine Techno, the Ajinomoto subsidiary that supplies the ABF resin film used by substrate makers, flagged continued demand growth through fiscal 2026 in its latest investor materials. Ibiden, headquartered in Gifu, and Shinko, based in Nagano, have both reported progress on new domestic substrate lines that support 18 layer and higher constructions for high bandwidth applications. Japanese trade daily Nikkei and engineering outlet Nikkei Monozukuri have tracked these build outs alongside retooling at Ibiden’s Ogaki sites and Shinko’s Nagano plants.

The Ministry of Economy, Trade and Industry supported several of these expansions under programs aimed at strengthening domestic semiconductor supply chains. Subsidy frameworks used for materials and packaging have also backed capacity at suppliers of plating chemistries and resins, including JSR, Resonac, and Taiyo Ink, that feed into high density substrate manufacturing. These inputs are critical for packages that pair HBM stacks with large AI logic dies.

TrendForce and Taiwanese brokerage research have highlighted that packaging remains the principal bottleneck for AI accelerators into 2027. The new Japanese substrate capacity does not remove constraints at TSMC’s CoWoS lines, but it reduces one upstream choke point for boards targeting Nvidia H100 class and successor parts. That lets assemblers reallocate scarce packaging slots to designs that had been delayed by substrate shortages.

For global hardware buyers, this tilt toward materials and substrate availability shows more supply risk shifting upstream to Japan. Vendor delivery schedules for AI systems will depend not only on foundry and OSAT throughput, but also on a concentrated set of Japanese suppliers whose policy backed investments are now setting the pace on substrate supply.