Japan's Murata and other component makers are raising prices on the small capacitors, power semiconductors, and circuit board materials inside AI server racks — and the moves are confirmed, not just rumored.
The artificial intelligence supply squeeze has a new address. After two years of headlines about scarce GPUs and high-bandwidth memory, the price pressure inside AI hardware is now drifting up the bill of materials, into the small capacitors, power semiconductors, and base materials that most buyers never have to think about.
That is the takeaway from a confirmed round of supplier-side price moves — now backed by company statements, earnings commentary, and independent third-party tracker confirmation, not just analyst reads. Japan's Murata, the dominant maker of multilayer ceramic capacitors (MLCCs), the tiny, ubiquitous components that smooth and store current on virtually every circuit board, confirmed in March 2026 that it is implementing price increases of 15 to 35 percent on AI server and high-end automotive grade MLCCs, effective April 1 — a move first reported by Bloomberg in February, when Murata president Norio Nakajima described demand for AI-related components as "extremely strong" (Bloomberg). TrendForce, a semiconductor industry price tracker, confirmed the April 1 price hike on March 17 (TrendForce). Power-semiconductor maker Infineon separately notified customers on February 5, 2026, that continued tight supply of power switches and related ICs — driven by rapid AI data center buildout — along with rising raw material and infrastructure costs, will prompt price adjustments starting April 1, 2026, with all new orders and existing backlog scheduled to ship on or after that date reflecting the revised pricing (TrendForce).
The shift matters because it changes where the leverage sits in the AI hardware stack. For most of 2024 and 2025, the binding constraint was at the top of the rack: Nvidia's GPUs and the HBM memory stacked next to them. The new round of confirmed price actions suggests that constraint is now migrating one tier down, into the unglamorous components that turn a GPU into a working server.
"Capacitors and power semiconductors are now both moving into a price-increase cycle, with adjustments spreading from selected products to wider categories," Liu Gaochang, an analyst at Chinese brokerage Sinolink Securities, told the South China Morning Post (SCMP). One sentence, but it captures the structural story: the inflation is broadening, not narrowing.
Why the cost pressure is migrating upstream is mostly mechanical. A modern AI server pulls far more passive components than a traditional one. Industry estimates cited in the SCMP analysis put the ratio at three to ten times as many capacitors per AI server, an order-of-magnitude jump that turns every additional GPU shipment into a much larger order book for parts makers like Murata (SCMP). Power semiconductor order books, in the same reporting, are described as "fully loaded," meaning suppliers have less room to absorb cost without passing it on.
Behind those parts sits a second layer of price pressure: copper-clad laminates (the reinforced base material for printed circuit boards), the glass fabric that reinforces them, industrial gases used in chip fabs, valves, and ceramic parts inside the tools that make the chips themselves. The same SCMP analysis flags rising costs in aluminium foil, chemical materials, and electricity, the kind of input inflation that does not show up in a GPU price quote but does show up in a finished server's margin profile.
For data center operators, hyperscalers, and the original device manufacturers that buy the parts, the practical effect is that the AI buildout is starting to look less like a pure GPU story and more like a full-stack one. When the bottleneck was at the top of the rack, the workaround was to buy more, wait longer, or pay Nvidia a premium. When the bottleneck moves into hundreds of small capacitors per server, the workaround is harder to design around: the suppliers are more fragmented, the products more commoditized, and the cost of switching higher.
The sourcing picture has materially tightened since the original reporting. Murata's April 1 price increase — 15 to 35 percent on AI-server-grade and automotive-grade MLCCs — is confirmed by TrendForce and corroborated by Murata president Norio Nakajima's direct comments to Bloomberg in February describing AI component demand as "extremely strong." Infineon's April 1 price hike is confirmed by the company's own customer notice, reported via TrendForce, citing AI data center buildout as the explicit driver. TrendForce, an independent third-party price tracker, has confirmed both moves. The 3 to 10 times capacitor ratio remains an industry estimate rather than a measured average, sourced via Sinolink Securities through SCMP. And no third-party price tracker such as TrendForce, IDC, or Omdia has yet published a comparable index specifically tracking the broad-based spread thesis for capacitors, power semiconductors, and circuit-board base materials — so while the company-confirmed price moves are solid, the full breadth of the spread across all upstream categories still rests on the analyst read and named supplier list.
What to watch next is whether the price moves hold through the second half of 2026. Murata has already announced an additional ¥80 billion (~$500 million) in MLCC capital expenditure on top of an already-elevated spending plan, with management telling Japan's Nikkan Kogyo Shimbun that capacity is "still very much insufficient." A second round of MLCC price increases — reported in the 10 to 40 percent range for July 2026 on AI-server and automotive high-end models — is in the market, with spot market data confirming the trajectory, according to industry sources cited by Digitimes and passive-components.eu. Whether those July increases fully materialize, and whether Texas Instruments and Yangjie Technology join the confirmed column, will be the next signal to watch for.