AI Server Demand Drives Revival of Long-Term MLCC Supply Agreements
0xBroomberg
Surging AI server demand is forcing buyers to lock in long-term supply agreements — at a premium — for high-end MLCCs; Samsung Electro-Mechanics booked nearly ₩750 billion in two months, industry book-to-bill ratios hit five-year highs, and capacity expansion is back in full swing.
What is an MLCC, and why can't AI servers do without it?
An MLCC — multilayer ceramic capacitor, a fingernail-sized component that stabilizes chip power delivery and filters signal noise — is a basic building block on every circuit board. A single AI server uses several times more MLCCs than a standard server.
This means → the faster AI compute capacity scales, the faster high-end MLCC supply tightens.
In plain terms = if the GPU is the AI server's "brain," the MLCC is the "blood vessel" keeping it stable — a powerful brain is useless if the plumbing can't keep up.
Who is racing to sign contracts, and how much are they paying?
Samsung Electro-Mechanics landed AI-server MLCC contracts two months running: ≈ ₩450 billion in June, ≈ ₩295.1 billion in July — a combined total near ₩750 billion.
Taiwan's Yageo (2327.TW) said more customers are actively seeking long-term agreements, concentrated in AI compute, enterprise servers, SSDs, and memory products.
Some buyers are willing to pay a price premium to secure capacity. This reflects a shift from "push price down" to "guarantee supply" — in the AI hardware chain, a supply disruption is scarier than overpaying.
What do the book-to-bill ratios tell us?
The book-to-bill ratio (B/B — above 1.0 means new orders exceed shipments, signaling a supply deficit) has surged across the board: as of end-June, Murata hit 1.30, Taiyo Yuden 1.25, Samsung Electro-Mechanics 1.31 — all five-year highs.
Yageo's B/B jumped even higher to 2.2; Walsin Technology stood at roughly 1.3–1.4.
This means → order backlogs are approaching levels last seen in the previous shortage cycle, and suppliers' pricing power is strengthening fast.
How are manufacturers responding — how big is the expansion?
Yageo: utilization for both specialty and standard products is expected to exceed 90% in Q3; its Kaohsiung Dafa Plant 3 is near full capacity, with additional equipment in H2 and incremental output arriving in Q3–Q4. H1 capex totaled roughly NT$3 billion.
Walsin Technology: June resistor utilization at 85%–90%, MLCC utilization at 80%–85%; plans to expand both product lines by 10%–15% in 2026, with further expansion in 2027. Capex budget raised sharply from NT$0.5–1 billion to over NT$3 billion.
In plain terms = factories are running near-full and it's still not enough. The money is already flowing into new lines — but fresh capacity won't ship for several quarters at least.
Can this upcycle last?
Two variables will decide: whether AI server shipments sustain their current growth rate, and how fast new capacity actually comes online.
If AI server demand keeps climbing while new capacity lags, suppliers' pricing power will strengthen further — the second half of 2026 is the proving window.
This means → for investors, next quarter's earnings and utilization data are the core signals for judging whether this MLCC upcycle is a "real cycle" or a "short pulse."
Content is for reference only, not financial advice.