
Dongfang Securities: Metal soft magnetic powder cores are a must-have option for ASIC power saving, and the future usage of inductive materials is expected to further increase

Dongfang Securities released a research report stating that as overseas AI giants accelerate the deployment of self-developed ASIC chips, the application of chip inductors in AI servers will significantly increase. Metal soft magnetic powder cores are considered a necessary option for energy-saving ASICs, as they can effectively address PDN voltage drop and loss issues. In the future, with the upgrade of the DDR6 standard, the usage of inductor materials is expected to further grow. ASIC chips have significant advantages over GPUs in terms of power consumption and performance, becoming an important direction for the development of AI computing power
According to the Zhitong Finance APP, Dongfang Securities released a research report stating that due to the inevitable PDN voltage drop and losses in the traditional horizontal power supply model, as well as the synchronous increase in PDN losses brought about by the rise in AI computing power consumption, the core pain point for end users deploying ASICs is energy saving. Metal soft magnetic powder cores are a necessary option for ASIC energy saving. In the future, as the DDR6 standard continues to upgrade memory performance, energy efficiency, and power supply solutions compared to previous generations, the usage of inductive materials is expected to further increase. With overseas AI giants accelerating the large-scale deployment of self-developed ASIC chips next year, the application proportion of chip inductors in AI servers may significantly increase.
The main points of Dongfang Securities are as follows:
As overseas AI giants turn to self-developed ASIC chips, energy saving has become the core pain point for end users
With the development of AI computing power, AI computing chips represented by NVIDIA's GB300 have reached new heights in parameters and performance, while power consumption has also increased significantly. Due to the characteristics of ASIC (Application-Specific Integrated Circuit) being customized hardware architecture for a single task, which aligns closely with AI inference requirements, and having significant advantages over GPUs in terms of area, energy consumption, integration, and price, as well as being superior to GPUs in core indicators of computing power to power consumption ratio, overseas AI giants are turning to self-developed ASIC chips. Among them, Meta, in collaboration with Broadcom, is set to launch its first AI ASIC chip MTIAT-V1 in the fourth quarter of this year. Due to the inevitable PDN voltage drop and losses in the traditional horizontal power supply model, as well as the synchronous increase in PDN losses brought about by the rise in AI computing power consumption, the core pain point for end users deploying ASICs is energy saving.
Metal soft magnetic powder cores are a necessary option for ASIC energy saving
The demand for energy saving in ASICs places higher requirements on the construction and materials of power modules and inductors. In terms of structure, power modules designed with vertical stacking break the size limitations of planar PCB layouts, achieving a PDN path shortened to millimeter levels by closely coupling the power module with the processor, significantly reducing PDN losses; in terms of materials, due to the reduced structure of power modules and the high power and large current application scenarios of AI servers, higher requirements are also placed on the large current carrying capacity of inductors. Inductors made from metal soft magnetic powder cores have a higher saturation magnetic flux density compared to ferrite, allowing them to withstand larger currents within a unit volume and exhibit better high-temperature stability, as well as lower core losses and simpler molding processes, making them a new type of inductor material that is essential for ASIC energy saving.
The 26-year ASIC expansion and mid-term DDR6 memory power management bring broad growth space for metal soft magnetic powder
Currently, overseas AI giants are competing to advance their self-developed ASIC chip layouts, with Meta having signed a long-term capacity agreement with TSMC, planning to launch multiple MTIA series chips in phases between the end of 2025 and 2026, and Microsoft planning to massively ship self-developed ASIC chips in 2027, leading to a significant increase in inductor demand; NVIDIA's AI GPU shipments are expected to maintain growth, and the AI computing chip market may enter a dual-drive era of GPU + AISC by 2026, with the total demand for inductors expected to exceed 400 million pieces In terms of DDR memory, the high capacity and high speed of DDR5 chips pose higher requirements for power integrity. DDR5 is the first to use a PMIC power supply scheme, and due to changes in the power supply architecture, the inductor form has evolved from large-sized, high-current assembled inductors to small-sized, thin inductors, which also places higher demands on power modules and inductor materials. In the future, as the DDR6 standard continues to upgrade memory performance, energy efficiency, and power supply schemes compared to previous generations, the usage of inductor materials is expected to further increase.
Leading company in the metal soft magnetic powder core industry, POCO, has technological advantages, and mid-term production capacity is expected to grow rapidly
POCO is the pioneer of industrial gas atomization powder production. The gas atomization metal powder preparation technology mastered by the company has advantages over water atomization powder, such as higher granularity and lower oxygen content, making it suitable for pressing into powder cores. The conventional and ultra-fine metal alloy soft magnetic powders produced by the company have finer particle sizes and lower oxygen content compared to industry peers, thus possessing stronger product competitive advantages. In terms of production capacity, the company's fundraising projects are expected to be fully completed by the end of 2025, with production capacity ramping up in 2026, and total production capacity is expected to reach 300 million pieces. With the improvement of production line efficiency, the company's inductor material capacity may have further growth potential in the mid-term.
Risk Warning
Lower-than-expected demand for inductors in downstream industries, intensified industry competition leading to decreased profitability, slower-than-expected capacity deployment, risks of changes in U.S. tariff policies, and changes in assumptions affecting calculation results

