The recent boom in building artificial intelligence (AI) data centers is spreading beyond graphics processing units (GPUs) to a supply crunch in power semiconductors, creating bottlenecks across the global supply chain. Demand for power management parts for AI servers has surged explosively, while supply has failed to keep up.
On the 7th, according to the semiconductor industry and market research firm TrendForce, lead times for some power semiconductors from major suppliers such as Infineon Technologies AG, Texas Instruments Incorporated (TI), and onsemi have recently exceeded 35 to 40 weeks. Citing delayed parts deliveries, TrendForce lowered its forecast for global server shipment growth this year to 13% from 20%. Power management integrated circuits (PMICs), which handle server voltage conversion and power control, and metal-oxide-semiconductor field-effect transistors (MOSFETs) have effectively emerged as strategic goods that "sell out because there isn't enough to sell." A MOSFET is a semiconductor that uses voltage to control the flow of current as precisely as a faucet, serving as a key switch to turn power on and off.
The fundamental cause of the supply crunch lies in a surge in power density stemming from changes in AI server architecture. AI servers consume far more power than general-purpose servers, and as next-generation GPUs' power consumption has soared, demand for analog semiconductors that handle power conversion and control has grown exponentially. In particular, as multi-stage power delivery and chip-level precision control designs have become standard, the number of related components installed has structurally increased. On top of this, heightened tensions in the Middle East and the resulting Strait of Hormuz risk have fueled supply instability from Qatar, the core of global helium supply, prompting concerns that semiconductor production will be disrupted.
The impact on industry is also becoming visible. As major suppliers allocate volumes to AI servers first, demand from manufacturers that failed to secure parts is spreading quickly to second- and third-tier suppliers such as Panjit in Taiwan and Taiwan Semiconductor (TSC). Leveraging Taiwan's robust analog semiconductor supply chain, these companies are filling gaps left by leading firms and directly benefiting. Some say the Taiwanese supply chain has emerged as a key pillar in building global AI infrastructure.
Memory chipmakers such as Samsung Electronics and SK hynix are continuing their earnings momentum on the back of rising demand for high bandwidth memory (HBM), but they are watching closely for the possibility that bottlenecks in power semiconductors could delay overall server shipments and dampen memory demand.
The government is also moving quickly. The Ministry of Trade, Industry and Energy and the Ministry of Science and ICT plan to invest 8.6 trillion won this year in national strategic technology research and development (R&D), up 30% from a year earlier, to support self-reliance in core technologies including power semiconductors.
The next-generation power semiconductor task force, launched as a public-private partnership, recently unveiled a roadmap to foster the next-generation power semiconductor industry, announcing plans to build the "southern power semiconductor innovation belt" connecting Busan, Naju, and Pohang, and presenting a blueprint to more than double technological self-reliance by 2032.
An industry official said, "Power semiconductors have now become a variable that determines the 'energy security' of AI infrastructure," and added, "Korea also needs to secure next-generation material technologies such as silicon carbide (SiC) and gallium nitride (GaN) to overcome the structural limits of the supply chain."