AI servers are quietly becoming an MLCC story — not just GPU and HBM.
Samsung Electro-Mechanics is finalizing a 700 billion won long-term MLCC deal with a global power and components firm. If it closes, total AI-server MLCC contracts signed in the past four months will exceed 4 trillion won.
They already locked in a 1.07 trillion won AI-server MLCC agreement this month — their largest MLCC contract ever — running January 2025 through December 2027. About ten global accounts are now under long-term agreements, with more in the pipeline.
The demand shift is real. $AMD's MI450 platform swapped aluminum and tantalum caps for MLCCs, increasing 47 μF parts from 1,440 per board to 10,544 — a 632% jump.
$NVDA's Vera Rubin architecture is scaling 100 μF MLCCs from 320 per board to 500. Google TPU v8, AWS Trainium 4, and Meta MTIA 400/450 are all following the same high-capacitance playbook.
Murata is ramping AI-server-grade production and building out Izumo for 2027. Taiyo Yuden and Kyocera are moving in parallel. Samsung Electro-Mechanics boosted high-cap output in March and plans to invest roughly 15 trillion won into its Busan campus through 2040 for data-center MLCCs and package substrates.
The AI infrastructure build-out runs deeper than most realize.
Samsung Electro-Mechanics is finalizing a 700 billion won long-term MLCC deal with a global power and components firm. If it closes, total AI-server MLCC contracts signed in the past four months will exceed 4 trillion won.
They already locked in a 1.07 trillion won AI-server MLCC agreement this month — their largest MLCC contract ever — running January 2025 through December 2027. About ten global accounts are now under long-term agreements, with more in the pipeline.
The demand shift is real. $AMD's MI450 platform swapped aluminum and tantalum caps for MLCCs, increasing 47 μF parts from 1,440 per board to 10,544 — a 632% jump.
$NVDA's Vera Rubin architecture is scaling 100 μF MLCCs from 320 per board to 500. Google TPU v8, AWS Trainium 4, and Meta MTIA 400/450 are all following the same high-capacitance playbook.
Murata is ramping AI-server-grade production and building out Izumo for 2027. Taiyo Yuden and Kyocera are moving in parallel. Samsung Electro-Mechanics boosted high-cap output in March and plans to invest roughly 15 trillion won into its Busan campus through 2040 for data-center MLCCs and package substrates.
The AI infrastructure build-out runs deeper than most realize.
