AI Server Component Boom
Samsung Electro-Mechanics Wins $293 Million MLCC Contract for AI Servers Through 2027
Key Points
The AI-server MLCC contract is valued at 454 billion won (about $293 million) and runs through 2027 (January 1 to December 31); the customer is undisclosed, but industry sources say it is a major U.S. cloud provider.
Industry observers note that such a large, one-year MLCC supply contract is unusual, signaling tightening AI-server component supply amid strong demand.
AI servers use far more MLCCs than conventional servers—more than 20,000 MLCCs per GPU and up to 600,000 MLCCs per server rack.
Samsung Electro-Mechanics holds about 25% of the global MLCC market and leads AI-server MLCCs with more than a 40% share.
AI-server MLCCs must withstand harsh operating conditions, including temperatures of 105°C or higher and voltages up to 100V, and must be densely packable on server boards due to space constraints.
Samsung Electro-Mechanics has signed a $293 million deal to supply a critical AI server component to a major U.S. cloud company. The contract, worth 454 billion won, covers the full calendar year of 2027 — from January 1 to December 31 — according to GuruFocus and The Elec.
The components at the center of the deal are multilayer ceramic capacitors, or MLCCs — tiny parts that stabilize power inside AI servers. The customer was not named, but KED Global and other industry sources say it is a major North American cloud service provider, with speculation pointing to Amazon, Google, or Meta.
A Contract Size That Signals a Supply Crunch
One-year, fixed-price MLCC contracts of this scale are almost unheard of. The industry normally runs on quarterly pricing or spot-market buys. Experts say the unusual terms reveal how desperate U.S. big tech has become to lock in supply, according to The Elec and MK News.
The deal represents about 4% of Samsung Electro-Mechanics' total annual revenue and roughly 9% of its core components division, The Elec reported. Production lines for high-end MLCCs are already fully booked, with a two-year lead time needed to add new capacity. That bottleneck could keep AI hardware prices elevated through 2027.
AI Servers Devour 270 Times More MLCCs Than Conventional Ones
A standard server uses around 2,200 MLCCs. A single AI GPU needs 20,000 to 30,000. A full AI server rack can require up to 600,000 MLCCs, according to MK News and industry research firm 7SEtronic. That staggering difference explains why AI is reshaping demand for what was once a commodity part.
MLCCs act as shock absorbers for electricity. AI chips draw thousands of amperes at very low voltages — around 0.8 volts. Without MLCCs to smooth that out, chips suffer noise and slowdowns. AI-server MLCCs must also survive temperatures above 105°C and voltages up to 100V, far beyond what a smartphone part can handle, according to The Korea Herald.
Samsung Leads the AI-Server MLCC Race With 40% Share
Samsung Electro-Mechanics is the world's second-largest MLCC maker overall, with about 25% of the global market. But in the specialized AI-server sub-segment, it leads with more than 40% share, ahead of Japan's Murata Manufacturing, according to BusinessKorea. CEO Chang Duck-hyun said the deal shows "our MLCC technology has been recognized as a key component for the AI era."
The company has been pushing beyond MLCCs. On May 20, 2026, it signed a separate 1.5 trillion won ($1.1 billion) deal for silicon capacitors for AI infrastructure through 2028, according to Samsung Newsroom. It also launched full-scale silicon capacitor production on June 14, aiming to be a one-stop shop for AI hardware makers. AI-related MLCC demand is projected to grow 3.3 times between 2025 and 2030, per 7SEtronic.
Bull Case vs. Manufacturing Reality: Two Views on the Upside
Morgan Stanley analysts have raised their target price for Samsung Electro-Mechanics to 2.56 million won, calling AI components a structural "Super Cycle" rather than a temporary boom, according to HTX Research. Analysts also forecast the company will rejoin the "1 trillion won operating profit club" by late 2026, per The Korea Herald.
Not everyone is fully bullish. Former SEMCO research engineer Phillip Yoon warns that pushing capacitance limits into tiny footprints is extremely hard to manufacture reliably. Any defects in these ultra-dense parts could lead to costly recalls or thinner-than-expected margins, The Korea Herald reported. The physics of making them, he argues, may not keep pace with the soaring demand.
Sentiment
Mostly PositivePublishers
11
Articles
8
Reach
19
