Against the backdrop of the acceleration in digital transformation (DX), the rapid spread of generative AI, and the expansion in the cloud service market, servers for data centers are being required to deliver computing performance and power supply capabilities far greater than conventional IT infrastructure. In particular, high-density servers equipped with numerous GPUs and AI accelerators are becoming the mainstream for AI workloads, including large language models (LLMs) and image and speech recognition applications. As a result, power consumption per server is increasing. This has meant that accommodating steep power supply loads involving very large currents flowing instantaneously has become a critical issue.
In the latest CPUs, GPUs, and AI accelerators, operating voltages have been decreasing as semiconductor process technologies continue to scale down. On the other hand, higher computational density has led to current consumption reaching the level of several hundreds of amperes.
Therefore, the following characteristics are essential in server power supplies.
Moreover, the interior of servers features high-density mounting. This raises concerns about the heat generated building up inside servers. Consequently, there is demand for low-profile capacitors that can be mounted beneath heat sinks.
As mentioned earlier, there has been a trend recently for higher performance driven by advances in generative AI and DX in servers.
Accordingly, components must possess the functions of low ESR, high capacity, low profile, and long life at high temperatures.