AI-Driven Power Revolution Sparks MLCC Shortage, Delivery Times Extend Beyond 20 Weeks, Supply Tightness Expected to Last Until 2027

The chairman of Hesintang, Tang Jinrong, stated on May 25th at the investor conference that the current wave of AI application is triggering a "power revolution," with the demand for multilayer ceramic capacitors having risen to the highest level in over 20 years. The company's production capacity remains fully utilized at present, and they announced a two-stage expansion plan: new equipment is to be installed in batches in the third quarter and the first quarter of 2027, with the expectation of a 20% to 30% increase in production capacity by the end of 2026 after commissioning, and a further 30% to 40% increase in 2027.

This surge in demand has a clear technological trajectory. Hesintang traces this AI-driven increase in demand back to the release of NVIDIA's A100 chip in 2020, the further strengthening of the trend with the H100 in 2022, and the acceleration of demand due to the large-scale shipment of the Blackwell chip in 2026. The Vera Rubin is scheduled to begin mass production in the fourth quarter, with the Rubin Ultra to be introduced after 2027, and the demand curve continues to extend upward.

The supply chain pressures have been passed upstream. Tang Jinrong pointed out that low-loss NPO and medium to high-voltage X7R specification products are particularly scarce, with the delivery time for upstream equipment extended to 1 to 1.5 years, and the rush to purchase equipment is comparable to the expansion cycle of 2018. He emphasized that the fundamental difference from 2018 is that AI demand is driving a real "increase in both quantity and price," and some material numbers have already run out of stock.

The AI server manufacturer, WeiYing, corroborated this assessment on the same day at the shareholder meeting, directly naming memory, high-end PCBs, and MLCCs as the biggest bottlenecks in the AI server supply chain. Hesintang stated that the company has entered the supply chains of four major American cloud service providers and Dell, and the current strategy is to enhance product specifications rather than raising prices, and plans to reduce shipments of mid to low-end products, focusing resources on high-end product lines.

Looking forward to the next three to five years, Hesintang believes that emerging markets such as AI data centers, backup battery modules, humanoid robots, and eVTOL will continue to increase the penetration rate of high-end MLCCs, and the industry structure is expected to steadily improve.