MLCC long-term contracts are continuously signed, Samsung Electro-Mechanics secures 4 trillion Korean won

The expansion of investments in artificial intelligence (AI) servers is causing supply and demand pressure to shift from Graphics Processing Units (GPU) and High Bandwidth Memory (HBM) to Multi-Layer Ceramic Capacitors (MLCC). As the power consumption of AI accelerators increases, usage of high-capacity and high-reliability MLCCs is also rising, with leading companies such as Samsung Electro-Mechanics and Murata quickly increasing the proportion of their high-end products.

According to industry sources on the 25th, Samsung Electro-Mechanics is in the final stages of negotiations with global power and electronic component companies to sign a long-term supply agreement (LTA) for MLCCs used in AI servers, valued at around 700 billion KRW. If the contract is successfully concluded, the cumulative value of silicon capacitor and AI server MLCC LTAs obtained by Samsung Electro-Mechanics over the past four months is expected to exceed 4 trillion KRW.

Earlier this month, Samsung Electro-Mechanics also signed an AI server MLCC supply contract worth 1.0722 trillion KRW with a global large-scale enterprise. The contract period is from January to December 2027, lasting one year. This is the largest MLCC long-term supply contract ever signed by Samsung Electro-Mechanics. Currently, the company has entered into LTAs with more than ten global corporations and is exploring the possibility of additional contracts.

The consecutive wins of orders are due to the growing demand for MLCCs used in AI servers. MLCCs play a role in stabilizing power supply and reducing interference between components in electronic circuits. Since AI server products must operate for long periods in high-temperature and high-voltage environments, they require greater capacity and reliability than typical IT products.

In particular, as the performance and power consumption of AI accelerators continually increase, the importance of components that ensure stable power supply becomes more prominent. This is because GPUs and Application-Specific Integrated Circuits (ASICs) consume large amounts of power instantaneously, making it necessary to reduce voltage fluctuations. As a result, the number of high-capacity MLCCs used in a single AI server is also rising rapidly.

According to Taiwan-based market research firm TrendForce, AMD, during the validation process of its next-generation MI450, replaced some aluminum/tantalum capacitors with MLCCs, increasing the number of 47μF-level products used per board from 1,440 to 10,544, a surge of 632%. On NVIDIA’s Vera Rubin platform, the number of 100μF-level MLCCs used per board also increased from 320 to 500.

The demand market is also continuously expanding. As cloud service providers (CSPs) such as Google (“TPU V8t/i”), Amazon Web Services (AWS) (“Trainium 4”), and Meta (“MTIA 400·450”) increase production of their self-developed AI accelerators, demand for high-capacity MLCCs also grows. Not only NVIDIA GPUs, but ASICs developed by large technology companies are also adopting high-end products, making AI servers a new major demand market.

Major MLCC manufacturers are responding to this trend by increasing production of high-end products. Since the end of last year, Murata has expanded mass production of high-end MLCCs for AI servers and is also building a new factory in Izumo, scheduled to begin operations in 2027. Taiyo Yuden and Kyocera are also strengthening their production of products used in AI servers.

Samsung Electro-Mechanics has also expanded production of high-capacity MLCCs since March this year. At the same time, it has announced plans to develop its Busan factory into a core production base for high value-added MLCCs and packaging substrates used in AI data centers and servers. To this end, Samsung Electro-Mechanics has decided to invest approximately 15 trillion KRW in the Busan factory (including the packaging substrate business) by 2040.

Industry insiders stated: “MLCC, which previously grew centered on mobile devices and the IT sector, is now gradually expanding its core application areas to high-capacity products.” He added: “Due to the high difficulty of MLCC manufacturing, it is not easy to expand production capacity in the short term. Therefore, as AI infrastructure investment continues, related demand is expected to remain strong.”