Summary

AI is advancing fast, shifting the paradigms of data centers and server infrastructure at an unprecedented pace. The recent surge in demand for AI computation has driven the explosive growth in the AI server market based on high-performance GPUs and CPUs, highlighting the importance of power systems and computing/network boards. ​

Power

The rapid advancement of AI not only requires more sophisticated GPU technologies, but also power systems to support them in a stable and efficient manner. Adoption of 48V power lines is accelerating, as their higher voltage compared to 12V systems reduces power loss and boosts board design efficiency. At the same time, resonant circuits, such as resonant converters (LLC), are being used more and more for high-efficiency power conversion. Furthermore, the introduction of new 800V power systems marks the next step in AI server power architecture. Thus, AI power systems are becoming increasingly complex and sophisticated. These advancements enable higher power density, improved efficiency and scalable infrastructures that can meet the growing computational demands of modern AI workloads.

Below, we introduce you to our key high-voltage, high-capacitance MLCC products designed to support these sophisticated power architectures and deliver reliable performance in challenging AI environments.

Computing/Network

GPUs and CPUs run at voltages below 1V and their current can immediately change by tens to hundreds of amperes. Depending on clock frequency and computational load, high-capacitance MLCCs located near the GPU must act as current buffers for a stable system operation.
NIC and PCIe used in AI servers operate in frequency bands of tens of GHz. High-capacitance MLCCs absorb low-frequency noise and, when combined with small sizes, suppress power noise across a wide bandwidth. Thus, maximizing AI server computing performance depends not only on processor performance but also on securing and optimally designing high-capacitance MLCCs—making MLCC NPI products key components in AI computing and network board design.

Please refer to the following for related product articles on small/large ultra–high-capacitance MLCCs.​