Shortages reach MLCCs
AI infrastructure demand is spreading supply pressure beyond headline chips and memory into multi‑layer ceramic capacitors (MLCCs) — small passive components used across servers and power systems — which can delay system deliveries despite healthy demand. Because these parts are easy to overlook in forecasts, sellers now face fulfilment risk driven by obscure components that can change deployment sequencing or force substitutions. (digitimes.com)
A data center can have every expensive chip it needs and still miss its ship date because it is waiting on parts that cost a tiny fraction as much. The part now causing trouble is the multi-layer ceramic capacitor, a small ceramic stack that smooths voltage the way a shock absorber smooths a pothole. (samsungsem.com) Artificial intelligence servers use far more of these ceramic capacitors than ordinary servers because their graphics processors swing from low to high current in an instant. Samsung Electro-Mechanics says artificial intelligence servers use more than 10 to 15 times as many multi-layer ceramic capacitors as general-purpose servers. (samsungsem.com) Those capacitors sit close to the graphics processor and act like tiny local batteries for split seconds. Samsung Electro-Mechanics says the processors in these systems run below 1 volt while drawing thousands of amperes, which is why the capacitor count keeps rising even as board space gets tighter. (samsungsem.com) The power system around the server is changing too, and that pulls in a different class of capacitor. Vendors are moving from 12-volt designs toward 48-volt systems and even 800-volt distribution inside high-power racks, which increases demand for higher-voltage ceramic capacitors as well as the small high-capacitance ones near the chips. (samsungsem.com 1) (samsungsem.com 2) That is why the squeeze is spreading beyond the famous bottlenecks like graphics processors and high-bandwidth memory. DigiTimes reported on April 10, 2026 that tightening availability is now showing up in multi-layer ceramic capacitors from suppliers including Murata and Samsung Electro-Mechanics. (digitimes.com) The supply base here is narrower than it looks because the hardest parts to make are not generic commodity pieces. TrendForce said on February 5, 2026 that high-end multi-layer ceramic capacitor lines at Murata, Samsung Electro-Mechanics, and Taiyo Yuden were all running above 80% utilization as demand climbed around Nvidia GB200 and GB300 server platforms and custom cloud chips from Amazon Web Services and Google. (trendforce.com) When a manufacturer is already near full utilization, buyers stop treating these parts like background items and start locking in supply. In January 2026, Chosun Biz reported that customers were shifting from quarterly purchases to annual contracts for Samsung Electro-Mechanics capacitors to secure volumes for artificial intelligence servers. (biz.chosun.com) Prices are starting to reflect that change. Bloomberg reported on February 17, 2026 that Murata had begun internal discussions about raising prices on advanced ceramic capacitors used to control large power loads in artificial intelligence servers, and TrendForce later reported expected double-digit increases on some lines. (bloomberg.com) (trendforce.com) The awkward part for server makers is that a missing capacitor does not always kill demand; it scrambles the build plan. If one qualified part is short, an original design manufacturer may delay a board, swap to a different approved part, or ship racks in a different sequence, which turns a shortage of obscure passives into a delivery problem for an entire artificial intelligence cluster. (digitimes.com) (samsungsem.com) That is how an artificial intelligence boom spreads through the supply chain. First the headlines go to graphics processors and memory, then the pressure moves into the tiny ceramic parts that keep the power clean enough for those processors to run at all. (digitimes.com) (samsungsem.com)
