Artificial intelligence is triggering the most intense period of supply and demand reconfiguration in the history of the global MLCC (Multi-Layer Ceramic Capacitor) industry. Goldman Sachs characterizes the current boom cycle as the "largest in scale and longest-lasting" upturn for the sector. The explosive demand from AI data center construction has pushed global lead times for in-demand products to a maximum of 24 weeks, approximately double the normal levels.
Recent moves by Samsung Electro-Mechanics are particularly noteworthy. According to reports, the company recently signed a supply contract worth 1.5 trillion won with a major global technology firm, setting a record for its largest single order ever. Revenue from this deal is expected to be recognized starting in 2027. Concurrently, Samsung Electro-Mechanics announced plans to build a new factory in the Philippines to expand its server-grade MLCC production capacity.
The situation facing industry leader Murata Manufacturing similarly reflects the overall market tightness. Data from Mirae Asset Securities indicates that current global lead times for popular MLCCs have stretched to a maximum of 24 weeks from a normal level of around 10 weeks. Murata has announced a capital expenditure plan of 250 billion yen (approximately $1.56 billion) for the 2026 fiscal year. This includes an additional 80 billion yen specifically earmarked for expanding server-grade MLCC capacity, with its current capacity utilization rate already nearing 95%.
This AI-driven supply-demand mismatch is now translating into price pressures. According to Citrini Research, spot and distributor prices for consumer-grade MLCCs have risen 20% to 40% from previous levels. Analysts anticipate that as the industry renegotiates contract prices quarterly, these increases will further spread to long-term agreement customers. Investor earnings expectations for relevant suppliers are being intensively revised upward.
Surge in AI Server Demand Fuels Unprecedented Industry Cycle
The large-scale expansion of AI data centers is the core engine of this MLCC super-cycle. A single AI server requires up to 28,000 MLCCs, roughly 13 times the number needed for a standard server. Nvidia's next-generation Rubin architecture computing platform will increase MLCC usage per board to 12,000 units, up from 6,500 in the current GB200 platform.
Beyond AI demand, electric vehicles and humanoid robots are also providing sustained consumption pull. According to Murata Manufacturing data, an electric vehicle equipped with L2+ level autonomous driving functions requires over 10,000 MLCCs, whereas a smartphone needs only 800 to 1,000 units.
Citrini Research estimates the annual growth rate for AI servers is approximately 80%, a pace far exceeding the expansion rhythm any supplier can quickly match, thereby creating a structural supply-demand gap. The CEO of Japan's Taiyo Yuden, Katsuya Sase, has directly described this demand wave as "frightening."
Samsung Electro-Mechanics Bets on Silicon Capacitor Integration
As competition in traditional MLCCs intensifies, Samsung Electro-Mechanics is betting on a more differentiated technological path—integrating Silicon Capacitors (Si-Cap) with MLCCs and FC-BGA packaging substrates into a unified solution.
Silicon capacitors are passive components made using silicon wafers as the base material. They achieve energy storage by etching microscopic holes into the wafer and placing electrodes within, allowing device thickness to be compressed below 100 micrometers. This technology can provide more stable voltage output in high-power-density chip packaging environments, a key characteristic required for AI servers. Samsung Electro-Mechanics forecasts the silicon capacitor market will grow at an annual rate exceeding 18%, with applications extending from mobile devices to AI servers, automotive electronics, aerospace, and optical communications.
The head of Samsung Electro-Mechanics' Si-Cap development group, Kim Won-ki, stated the company is incorporating silicon capacitors into its packaging substrate product portfolio as part of a comprehensive solution strategy and has begun mass-producing silicon capacitor products embedded in FC-BGA substrates. FC-BGA substrates are widely used in AI data center and PC processors. It is reported that Samsung Electro-Mechanics has supplied silicon capacitor products for Marvell Technology's AI ASIC and Samsung Electronics' Exynos 2600 application processor.
According to industry analyst views cited by a report, Samsung Electro-Mechanics is currently the only company with capabilities across all three businesses: MLCCs, FC-BGA packaging substrates, and silicon capacitors. While Murata Manufacturing and Taiwan Semiconductor Manufacturing Company (TSMC) also produce silicon capacitors, neither offers full-line coverage. Technologically, Samsung Electro-Mechanics employs a DRAM architecture-based solution, leveraging fine-patterning technology accumulated from the DRAM line-width reduction process. TSMC uses a trench structure derived from logic process technology. Samsung Electro-Mechanics conducts mass production on 300mm wafers, outsourcing wafer manufacturing to foundries and operating under a fabless model.
Murata's Major Expansion Fails to Match AI Demand Pace
As the world's largest MLCC manufacturer, Murata Manufacturing is tackling capacity bottlenecks with a 250 billion yen capital expenditure plan for the 2026 fiscal year, of which 80 billion yen is specifically for expanding server-grade MLCC capacity. However, even with this additional investment, Murata expects its capacity expansion on a load basis over the next two years to be only slightly above 20%, a pace that still lags significantly behind AI demand growth.
Raw materials and the upstream supply chain also pose risks. Nano-scale ceramic powder is a key dielectric material for producing high-end MLCCs, requiring extremely high purity. This niche market is also dominated by Japanese chemical companies. Furthermore, the main raw material for MLCC release film is crude oil. Geopolitical tensions in the Middle East have led to significant cost increases, further squeezing manufacturers' cost control margins.
For Samsung Electro-Mechanics, in addition to its silicon capacitor strategy, its reported new factory in the Philippines is set to begin construction this year, specifically to expand high-end MLCC capacity for AI servers.
Price Increases Continue as Analyst Views Diverge
Supply tightness is gradually transmitting to prices. According to Citrini Research, spot and distributor prices for consumer-grade MLCCs have risen 20% to 40% from previous levels, with some of the increase driven by inventory stocking and double-ordering. Taiyo Yuden has notified customers it will raise prices on certain products, including MLCCs, starting in May.
Murata Manufacturing has hinted it may consider raising contract prices if crude oil prices push raw material costs beyond its internal absorption capacity. Morningstar director Kazunori Ito believes the "extremely tight" supply situation could trigger price increases that "exceed market expectations." Mirae Asset Securities analyst Junseo Park has raised the 2027 average selling price forecast for Samsung Electro-Mechanics' MLCCs by 10%, noting that price hikes have already materialized in some distribution channels. As the industry negotiates pricing quarterly, further price increases are likely to continue.
However, Bank of America Securities Japan research analyst Masashi Kubota holds a more cautious stance. In a recent research report, he noted that some price increase forecasts based on supply tightening are already "too aggressive." Even if manufacturers proceed with price hikes, the premium space is primarily used to cover higher fixed costs rather than improve profit margins.
Goldman Sachs: Cycle Still in Early Stages, Could Extend to 2030
A key market question is how long this super-cycle can last. Goldman Sachs' assessment is that the industry remains in the early stages of this AI-driven cycle.
According to a recent report by Goldman Sachs analyst Daiki Takayama, Murata Manufacturing expects AI infrastructure investment to peak around 2028. However, influenced by factors like power supply shortages, the cycle could potentially extend to 2030. Morningstar's Kazunori Ito expects AI server and infrastructure investment to "continue growing at least until 2028," which would provide relatively clear medium-term growth visibility for passive components like MLCCs.
Comments