On September 14, channel and A-share market data confirmed in tandem that TLVR coupled inductors built for AI server power delivery have entered a fresh price-upcycle: overall prices are up 45%-70% this year, while spot prices on critically scarce part numbers have jumped 150%. The driver is an architectural shift. As NVIDIA's Rubin platform rolls out 48V vertical power delivery (VPD) at scale, the power loop's appetite for inductors has leapt — per-GPU inductor counts have risen from 22 to 48-64, and per-rack usage has climbed to 9,000-11,000 units, an increase of more than 200%.
Supply is hitting its ceiling faster than demand. TLVR is an entirely new product category: production lines need 12-18 months from commissioning to yield ramp, orders at major Taiwan-based makers are booked through 2028, and factory utilization already runs at full load. In parallel, Murata notified customers on September 10 that it will discontinue selected consumer-grade and automotive-grade part numbers across nine series including GRM, GRJ, GCM and GCJ, with last orders due by end-March 2028, freeing capacity for high-capacitance AI-server products — where lead times have stretched from 8-10 weeks to 20-26 weeks. Guohai Securities expects the high-capacitance MLCC supply gap to widen next year, with tightness persisting through 2028.
For the domestic components supply chain, passive-component pricing is switching from a cyclical to a structural logic: tightness is concentrated in AI power-delivery and automotive high-capacitance parts, generic consumer-grade parts remain over-supplied, and the two curves will keep diverging. For buyers, annual quota and price negotiation windows on molded power inductors, high-capacitance MLCCs and LTCC filters are moving earlier; alternative-part qualification and automotive-grade validation need to start ahead of demand, supported by part-number-level supply tracking, dynamic stocking alerts and cross-brand comparison to absorb this architecture-driven cost shift.