Artificial Intelligence Servers and 5G: A Qualitative Shift in the Demand Structure

[Opinion] Artificial intelligence servers and 5G base stations have not only increased the demand for multilayer ceramic capacitors (MLCCs) but have also driven market trends toward higher capacitance values, higher frequency bands, and lower parasitic effects. This shift has given rise to new specialized product categories—such as three-terminal and four-terminal devices, as well as reverse‑structure components—thereby reshaping the overall product portfolio landscape.

I. A Significant Increase in the Value of AI Servers. A typical general-purpose server motherboard requires approximately one thousand MLCCs. In contrast, AI accelerator cards equipped with GPUs see an exponential increase in the number of power rails and the associated transient currents, necessitating a large quantity of low-ESR/ESL energy‑storage capacitors and high‑frequency decoupling capacitors to ensure stable voltage regulation across all rails—resulting in a significantly higher per‑board cost compared to conventional designs. As the switching frequency of power modules rises, the requirements for capacitor performance—including ripple current handling, self‑heating, and effective capacitance under DC bias—become increasingly stringent. Moreover, traditional high‑capacitance X7R ceramic capacitors exhibit pronounced capacitance degradation under bias, further exacerbating these challenges.

II. High-Frequency Requirements of 5G Base Stations. Massive MIMO and AAU RF front-ends urgently require high‑Q, low‑loss RF MLCCs. As operating frequencies expand into sub‑6 GHz bands and even millimeter wave, dielectric loss (DF) and temperature stability have become critical performance metrics. Conventional X7R ceramics exhibit a marked increase in dielectric loss and a reduction in effective capacitance at high frequencies, necessitating the use of C0G/NP0‑type or dedicated RF dielectric materials. Furthermore, matching networks in RF chains are highly sensitive to capacitor temperature drift and bias‑voltage stability.

III. Miniaturization and high capacity advance in tandem. Arranging more power and signal traces within a limited PCB footprint, coupled with the concurrent advancement of ultra‑small packages such as 01005 and 0201, as well as high‑capacitance multilayer ceramic capacitors in the 100 µF range, poses significant challenges to thin‑film printing, interlayer alignment, and the reliability of terminal electrodes, while also driving up the overall cost of monolithic capacitors.

IV. Structural Shortages and New Product Categories. Due to the rapid growth in AI‑related computing power, high‑capacitance, low‑ESL capacitors—including three‑terminal and four‑terminal multilayer ceramic capacitors as well as LGA‑packaged devices—are experiencing structural shortages; their low‑ESL characteristics enable more effective suppression of power‑supply noise. When selecting components, it is essential not only to consider the nominal capacitance value but also to comprehensively evaluate ESL, self‑resonant frequency (SRF), and capacitance retention under bias conditions.

◆ Perspectives and Analysis: The ongoing advancement of AI computing power has turned high‑capacitance, low‑ESL capacitors into a category that remains in persistent short supply; meanwhile, the pace of 5G network deployment will drive fluctuations in demand for RF MLCCs. When conducting procurement and sales assessments, priority should be given to metrics such as high‑frequency loss, ESL, and capacitance retention under bias voltage, rather than focusing solely on nominal capacitance values and unit prices.