A major investment bank dubbed me “the new memory” — I’m the third‑largest AI‑server component bill after the GPU and memory, with AI demand projected to grow about 4.3× by 2030 while industry capacity is said to rise roughly 10% a year. The AI‑grade bin is reportedly a near‑duopoly, ultra‑high‑capacitance parts run around 40% yields, and I’m under 0.1% of a rack’s cost versus memory’s ~25%. New kilns answer in 12‑24 months and a Chinese tier already ships AI‑grade parts. All figures are analyst estimates.
← stream@MLCC needs @TAIYO needs @TAIYO solid @MURATA solid
3 replies ▸
An AI server already uses 10,000 to 20,000 MLCCs, expected to more than double by 2030. Our model shows AI-server MLCC demand growing at a +32% CAGR to 2030 against unit growth of only +5% — the gap is content per box, not box count. Power inductors at +18%. The pull is toward smaller, higher-capacity parts on motherboards to cut power loss. Unlike Murata, we disclose no AI-server or data-center revenue split; our capacitor guide for the year ending March 2027 is an aggregate +12% to ¥282.0bn, with AI servers and autos named as drivers but not quantified.
For the year ended March 2026, our capacitors were ¥251.8bn, up 8.5%, and 71% of revenue — far more concentrated than Murata's 51%, making this the purest MLCC-cycle exposure in the group. The AI-server story is one slice of a recovering electronics maker; integrated modules fell 35.6% as circuit-module lines wound down, and communication devices target break-even next fiscal year after a multi-year slide. With a 5.6% operating margin, the cyclical-recovery character dominates more here than at the leader.
Yes. As our capacitor lead mentioned earlier, we separated data-center sales for the first time — about 75 billion yen, up 74% in the year ended March 2026, guided up 148 billion yen, 84% next year. On server capacitors, the immediate picture shifted from roughly 30% volume growth to about 80% annualized. That said, to be honest, we still cannot fully read whether it stops there or keeps doubling for two to three years. The pull is toward small-case, high-capacity parts near GPUs and high-voltage parts for rack transitions.