Murata will mass-produce iPaS, a power substrate built for AI chips, next year, and is targeting ¥50 billion in annual sales within 18 months of launch. The strategic signal matters more than the number: the world's dominant maker of multilayer ceramic capacitors is climbing up the value chain into substrates it does not yet own.
The logic is power. As accelerator packages draw ever more current at lower voltages, the bottleneck is shifting from raw transistor density to how cleanly power reaches the die. Voltage droop, transient response, and thermal headroom are now first-order constraints on AI system performance. Whoever controls the substrate layer that sits between the board and the package captures a structurally growing slice of the bill of materials, and Murata is betting that its ceramics expertise translates into an advantage there. The risk is equally clear: substrates are a different manufacturing discipline, with entrenched incumbents and packaging houses that will not cede ground quietly. A ¥50 billion target is modest against Murata's total revenue, which tells you this is an option on a much larger platform, not a near-term earnings driver.
For the global supply chain, the move is a reminder that AI's hardware value is migrating into unglamorous layers. The market obsesses over GPUs and HBM, but power delivery, interposers, and advanced substrates are where quiet oligopolies are forming. Expect more Japanese and Korean passive-component specialists to make similar upstream moves, and expect hyperscalers and accelerator designers to lock in substrate supply the way they already scramble for memory.
For Japan, this is a rare offensive story rather than a defensive one. The country's component makers, Murata, TDK, Kyocera, Ibiden, have spent a decade watching value concentrate in foundries and design houses abroad. iPaS is an attempt to convert deep materials know-how into a defensible position adjacent to the AI boom, and it strengthens Japan's argument that its edge lies in the physical layer of computing rather than in models or cloud.
Japanese enterprises and the SIers serving them should read this as a shift in where domestic hardware leverage sits. Data-center and edge-AI system designs that assume commodity power components may need revisiting as differentiated substrates change thermal and density envelopes. The practical near-term takeaway for local engineering teams is narrower: qualification cycles for these parts are long, so any roadmap dependent on next-generation AI hardware should track substrate availability as carefully as it tracks GPU allocation.
