In today’s era of explosive growth in AI computing power, a single voltage drop can disrupt training tasks, resulting in significant economic losses. A seemingly ordinary electronic component is becoming a hidden champion in ensuring stable power supply for AI servers.
Under high loads, the power transient of a single GPU in an AI server can surge to 1.4 kilowatts, generating a 10-kilowatt surge current for the entire system, placing extreme demands on the power supply system. The economic losses from restarting training tasks after a voltage drop far exceed the initial investment in the power supply.
Traditional UPS systems, with their slow response and short cycle life, are no longer adequate for the millisecond-level protection required for AI computing loads.
01 The Shift from “Optional” to “Standard” for BBUs
NVIDIA has officially upgraded the backup power supply for its GB300 servers to “standard.” The cost increase per rack due to the addition of supercapacitors and batteries exceeds ten thousand yuan, reflecting the rigid demand for “zero-interruption” power supply.
AI data center power supply systems face three core challenges: instantaneous power surges of up to 150% during GPU load changes, compact chassis space, and the high-temperature environment of data centers exceeding 35°C.
Traditional UPS solutions are bulky, slow to respond, and have short lifespans, becoming a bottleneck restricting data center availability.
02. YMIN LIC Supercapacitor Technological Breakthrough
Addressing industry pain points, YMIN Electronics launched a new generation BBU solution based on lithium-ion supercapacitor technology. This solution uses an all-tab winding process to compress internal resistance to <1mΩ, achieving millisecond-level response capabilities.
Through graphene-modified electrodes and a compact square design, this product achieves a 4500F ultra-large capacity within a limited volume, supporting wide temperature range operation from -30°C to 70°C and a 1 million cycle life.
The YMIN LIC solution reduces volume by 50%–70% and weight by 50%–60% compared to traditional UPS systems, significantly freeing up rack space and supporting high-density, ultra-large-scale AI cluster deployments.
03 Significant Practical Application Results
In NVIDIA GB300 server applications, a single rack requires up to 252 supercapacitor units. YMIN’s LIC modules, with their high capacity density, ultra-fast response, and superior reliability, have become the preferred choice for domestic customers seeking high-end domestic alternatives.
This solution keeps voltage fluctuations within ±1%, fundamentally preventing AI training tasks from being interrupted by voltage dips. Its cycle life exceeds 1 million cycles, with a lifespan of over 6 years, charging speed increased by 5 times, and a significantly reduced total cost of ownership (TCO) over its entire lifecycle.
YMIN’s supercapacitor BBU solution constructs a hybrid energy storage architecture of “supercapacitor + BBU,” perfectly matching the high power density requirements of the GB300.
As the demand for AI computing power continues to grow, the reliability of the power supply system will become an indispensable cornerstone of data centers. YMIN Electronics, with its lithium-ion supercapacitor technology, has not only redefined the performance standards of AI server BBUs but has also achieved comprehensive breakthroughs in size, weight, and cycle life.
In the future, as AI clusters continue to expand, this efficient and compact power supply solution will become a key element in ensuring stable computing power, providing reliable guarantees for AI applications with “millisecond response and ten years of protection”.
Post time: Nov-28-2025