Last month, Samsung Electro-Mechanics presented its technology direction for AI server MLCCs at the Open Compute Project Asia Pacific Summit in Taipei, Taiwan.
The key issue is why MLCCs, small components, have become important in GPU racks with rising power density.
The intraday strength in the stock price on September 22 was also interpreted as the result of supply shortages coinciding with expectations for earnings, but the outlook needs to be distinguished from confirmed results.
3-Line Summary
1. Samsung Electro-Mechanics presented an MLCC solution for AI servers.
2. It estimated that the MLCC capacitance required per GPU would need to double every year.
3. Securities firms’ earnings forecasts are not confirmed results.
MLCCs Change as GPU Rack Power Increases
MLCCs, or multilayer ceramic capacitors, are components that help supply the electricity needed by electronic devices in a stable manner. At the Open Compute Project (APAC Summit), Samsung Electro-Mechanics proposed the so-called “law of MLCCs,” stating that the MLCC capacitance required for a single GPU doubles every year.
The background is the power density of AI servers. Yonhap Infomax reported that the average power density of a single data-center rack was previously 8~10 kilowatts, while the latest GPU racks are at the level of 100~120 kilowatts. The industry expects GPU racks to reach the level of 1 megawatt in the future. Because more power must be placed in a narrower space, the size and performance of components that handle power are becoming important at the same time.
Samsung Electro-Mechanics presented three solutions: ultra-small, high-capacitance MLCCs suited to the limited area of GPU packages; embedded MLCCs, which place capacitors inside printed circuit boards; and ceramic capacitors with low equivalent series inductance (ESL) performance. The company said it had set next year as a potential timeframe for introducing products applying these solutions. This is the company’s plan and does not mean that the launch has been confirmed.
September 22, When Supply Shortages Met Earnings Expectations
According to a Daum report, Samsung Electro-Mechanics traded at 5.53% above the previous trading day’s close at 150만8,000 won on the morning of September 22 at the Korea Exchange. The report explained that shortages of MLCCs and FC-BGA (flip-chip ball grid array) package substrates, along with expectations for third-quarter earnings, affected investor sentiment.
The earnings figures mentioned that day were estimates from iM Securities. iM Securities forecast Samsung Electro-Mechanics’ third-quarter revenue at 3조9,000억원 and operating profit at 6,500억원, viewing these as increases of 34% and 150%, respectively, from the same period a year earlier. It is important to note that these are securities-firm forecasts, not the results of an actual earnings announcement.
The same report said that after shortages of high-capacitance MLCCs emerged last month, they spread to general-purpose products this month. The explanation was that concentrating production capacity on server products could tighten supply and demand for other products as well. The expansion of AI servers is turning not only finished-product companies’ competition but also the supply capacity of components such as MLCCs and substrates, which handle power and signals stably, into competitive factors.
What to Confirm Is Actual Product Conversion, Not the “Law”
Samsung Electro-Mechanics’ “law of MLCCs” is a technical blueprint suggesting that component demand could grow in AI infrastructure. However, the claim that demand for capacitance will double every year is a direction proposed by the company and cannot be viewed as a result already proven across the entire AI server market.
What deserves attention in this trend is that MLCCs are not being treated merely as general-purpose components. As GPU rack power density rises, the proportion of products simultaneously requiring small size, high capacitance, and low ESL may increase. Whether the solutions Samsung Electro-Mechanics mentioned with next year as the target actually lead to commercial products, and how the supply environment mentioned by the company appears in its results, will determine the weight of the “law.”
References
Tags #SamsungElectroMechanics #MLCC #MultilayerCeramicCapacitor #AIServer #GPU #DataCenter #PowerDensity #EmbeddedMLCC #FCBGA #PackageSubstrate #OpenComputeProject #AIInfrastructure