In the AI compute investment boom of 2026, the market’s attention is locked on the supply-demand battle between GPUs and HBM memory chips. Yet, a more fundamental "hidden bottleneck" is emerging beneath the surface. Multilayer ceramic capacitors (MLCCs), often called the "rice of the electronics industry," are shifting from their traditional role as basic passive components to become a critical variable in the cost structure of AI servers.
In May 2026, Japanese passive component giant Taiyo Yuden issued an industry warning: MLCC demand for high-end AI servers has reached "alarming" levels, pushing production capacity to its limits. The global high-end MLCC supply chain is facing unprecedented supply pressure. With a single AI server rack consuming nearly 600,000 MLCCs, and the value per capacitor in high-end applications climbing steadily, this industry—long seen as a supporting actor—is undergoing a structural value re-evaluation driven by AI.
TrendForce data shows that the global server shipment growth rate for 2026 has been revised upward from 14.1% to 17%, with AI server annual growth exceeding 28%. This double-digit growth is expected to continue through 2027. According to Sigmaintell, global AI server shipments are projected to reach about 3.7 million units in 2026, a year-on-year increase of 51.3%, and double-digit growth is expected to persist through 2027 and 2028. The industry consensus is clear: the hardware race for AI compute infrastructure is accelerating, and MLCCs are becoming an unavoidable core beneficiary in this process.
Comparison Chart: MLCC Value in AI Servers
| Platform/Type | MLCCs per Unit | MLCC Value (USD) | BOM Ranking |
|---|---|---|---|
| Standard Server | ~2,000–4,000 | ~$60–120 | Outside top 15 |
| NVIDIA GB300 | ~30,000 | ~$1,530 | 6th–8th |
| NVIDIA VR200 NVL72 | ~600,000 | ~$4,320 | 3rd |
Data sources: Murata Manufacturing public disclosures, Morgan Stanley VR200 NVL72 rack BOM analysis (May 2026), Goldman Sachs research. Standard server data is industry average; AI server data reflects NVIDIA platform specifications.
AI Server Shipments Surge, MLCC Demand Grows Exponentially
To understand the current value transformation in the MLCC sector, we must first establish the basic quantitative coordinates for AI server market growth. TrendForce has raised the global server shipment growth rate for 2026 from 14.1% to 17%, with AI server annual growth exceeding 28%. Double-digit growth is expected to extend through 2027, reflecting the sustained acceleration of AI infrastructure buildout over the past year.
This expansion in shipment volume is only the first layer of demand growth. The more critical change lies in the geometric leap in MLCCs per device. Data from Japanese leader Murata Manufacturing highlights this difference: a standard server requires only 2,200–4,000 MLCCs per unit, while a NVIDIA GB300 AI server uses about 30,000. In March 2026, NVIDIA launched the VR200 NVL72 next-generation compute rack, with MLCC usage reaching 440,000–600,000 per unit. This means a high-end AI rack consumes tens to hundreds of times more MLCCs than a traditional server.
Industry forecasts for total demand show equally dramatic growth. CICC estimates that AI servers will require 72.6 billion MLCCs in 2026, an 87% year-on-year increase; demand will further rise to 136.7 billion in 2027, up 88% year-on-year. CITIC Securities projects global server MLCC shipments could exceed 400 billion units by 2030, with a compound annual growth rate of about 40%. This explosive growth stems from AI server architecture evolving from single motherboards to rack-level high-density compute platforms, where every additional GPU or HBM chip means dozens to hundreds more MLCCs.
From the perspective of high-performance computing power density, this trend has deep technical inevitability. MLCC usage per board in NVIDIA’s Rubin platform nearly doubled to 12,000 compared to the previous generation. Each leap in power density directly drives exponential increases in passive component usage, and every compute upgrade brings a corresponding rise in capacitor requirements.
From "Supporting Role" to "Leading Role": MLCCs Become the Third Largest AI Server Cost
Explosive demand is only one dimension of change. What truly drives the MLCC industry’s value re-evaluation is its rise in the AI server bill of materials (BOM) cost hierarchy.
Goldman Sachs analyst Nelson Armbrust recently noted that MLCCs have become the third largest cost item in current AI server BOMs, trailing only GPUs and memory chips. This conclusion has been widely cited and recognized in global electronic component industry research.
Morgan Stanley’s BOM analysis of the NVIDIA VR200 NVL72 rack provides more precise quantification. MLCC value per rack is about $4,320, up 182% from the previous generation GB300’s $1,530. This surge is driven by both increased volume and higher unit prices—a "quantity and price double hit."
At the industry scale, the global MLCC market is currently about $15 billion, with the AI server segment around $1.3 billion and growing at a robust 80% compound annual rate, while growth in automotive and mobile sectors has slowed. Goldman Sachs’ latest report suggests the AI-driven MLCC supercycle is just beginning, projecting market size to grow about 4.3 times from 2025 to 2030. This growth trajectory is rare in the passive component industry, marking a historic shift in MLCC value positioning.
In contrast, growth in traditional MLCC demand categories like smartphones and consumer electronics has noticeably slowed. This means the current MLCC cycle is structurally different—AI compute infrastructure is replacing consumer electronics as the new anchor for MLCC demand growth.
Global Leaders: Oligopoly and Structural Capacity Constraints
The global MLCC market exhibits a typical "oligopoly + domestic catch-up" structure, with the top five players (CR5) controlling over 80% of the market in 2026 and extremely high barriers to entry in the high-end segment.
Global MLCC Supply-Demand Balance Transmission Path
In terms of market tiers, Japanese and Korean firms dominate the first echelon: Murata holds a 25%–34% market share and about 70% in high-end fields like AI servers; Samsung Electro-Mechanics has an 18%–24% share; Taiyo Yuden and TDK together account for 15%–20%. These four leaders collectively command 85% of the AI server and high-end automotive sectors. Taiwanese manufacturers (Yageo, Walsin) hold 10%–15%, focusing on mid-to-high-end general and consumer electronics. Chinese firms such as Sanhuan Group, Fenghua Advanced, and Microgate Electronics have a combined share of 10%–12%, rapidly penetrating mid-to-high-end AI and automotive segments.
This highly concentrated supply landscape means capacity bottlenecks are almost systemic. On the demand side, leading companies have successively implemented price hikes. In June 2026, the MLCC industry saw its third price increase of the year, with Murata, Samsung Electro-Mechanics, Taiyo Yuden, and Panasonic all raising prices simultaneously. High-end AI/automotive MLCCs saw increases up to 35%, while standard consumer grades rose 6%–30%. Manufacturers are shifting capacity toward AI compute and automotive, with standard MLCC supply shrinking and a clear capacity split between high-end and conventional products.
On delivery cycles, Murata’s high-end MLCC production lines are running at 95% utilization, with lead times stretched beyond 20 weeks and some scarce models now on limited order. Taiyo Yuden’s high-capacity MLCCs have lead times of 16–24 weeks, with Malaysian plant capacity constrained and spot inventory low. Samsung Electro-Mechanics lead times exceed 18 weeks, with spot prices rising monthly.
| Manufacturer | Global Market Share | AI/High-End Share | Latest Developments |
|---|---|---|---|
| Murata | 25%–34% | ~70% (AI servers) | Raised prices 15%–35% in April; issued another price hike notice June 9, AI/automotive MLCCs up 10%–40%, effective July 1 |
| Samsung Electro-Mechanics | 18%–24% | Strong in automotive/5G | Plans to raise consumer MLCCs 5%–10%; AI high-capacity models up another 30% |
| Taiyo Yuden | With TDK, 15%–20% | Automotive/industrial high-end | Raised prices 6%–15% May 1; CEO warns demand is "alarming" |
| Fenghua Advanced | Leading domestic player | Accelerating in AI/automotive | Xianghe Industrial Park project completed April 2026; mid/high-voltage, high-temp, high-capacity MLCCs now used in AI servers |
Looking ahead, the pace of high-end capacity expansion is expected to lag downstream demand growth. High-end MLCC production line expansion cycles typically last 18–24 months, with core equipment dependent on a handful of Japanese machinery suppliers, limiting supply elasticity. This structural feature closely mirrors the supply-demand logic of HBM memory chips.
Capacity Expansion Pace and Supply-Demand Gap Evolution
Despite aggressive expansion by industry leaders, there remains a clear lag between capacity coming online and demand surging. Japanese and Korean firms are ramping up: Murata is investing about ¥80 billion in capital expenditure, with a new plant in Izumo, Shimane set to open in 2026 and AI capacity share expected to rise from 30% to over 45%. Samsung Electro-Mechanics is expanding its Tianjin plant by about 20%, with its new Philippines facility sized at 1.5 times current capacity, focused on AI server and automotive MLCCs. Taiyo Yuden plans to invest about ¥270 billion over five years to expand capacity, but its CEO describes this as a "forced acceleration" rather than proactive forward planning.
However, these expansion plans will take time to fully materialize. In the second half of 2026 through 2027, the high-end, high-capacity MLCC supply gap is expected to be 15%–20%, possibly widening to 30% in 2027. General consumer-grade products are also tightening as high-end capacity crowds out standard production, signaling that regular MLCC supply will remain constrained for the long term.
From the perspective of upstream core materials, MLCC supply constraints run even deeper than capacity. JPMorgan’s June 10 report notes that the true bottleneck in the MLCC supply chain is upstream nanometer-scale ceramic powders—high-end dielectric powders must be reduced to about 100nm particle size with 99.99% purity. This segment was historically dominated by Japanese firms like Sakai Chemical, but after Guoci Materials’ breakthrough, domestic market share has reached about 80%, with Samsung Electro-Mechanics among its top clients. However, ultra-fine powders (≤80nm) and 5N grade (99.999% purity) powders are still in validation or pilot stages, not yet fully replacing high-end imports. These upstream material constraints further limit high-end MLCC capacity expansion flexibility and prolong the supply-demand mismatch.
From ABF Substrates to MLCCs: Structural Transmission of Compute Investment
The MLCC boom is not an isolated phenomenon—it’s a key link in the full transmission chain from core chips to basic components in AI compute infrastructure. In this chain, ABF substrates offer a logical reference point: both exhibit similar supply-demand mismatch dynamics, though their market scale and industry impact differ.
ABF substrates are the critical bridge connecting CPUs, GPUs, and other logic chips to external circuits, playing an irreplaceable role in advanced packaging. According to IEK, the global ABF substrate market is expected to reach about $10.02 billion in 2026, expanding at a 22.9% CAGR from 2024 to 2028. Technically, ABF substrate sizes used in NVIDIA Rubin and Rubin Ultra platforms have increased to 100×91mm² and 153×77.5mm², with layer counts rising from 12–14 to 18–20, and unit substrate area consumption reaching 5–10 times that of traditional PC substrates.
Both ABF substrates and MLCCs face highly similar structural constraints: technical specifications keep rising, driving up per-unit capacity consumption; Japanese and Korean firms dominate; and expansion cycles last 12–24 months. In Q1 2026, leading ABF substrate makers increased utilization from 75%–80% in Q3 2025 to about 90%. HSBC models predict the ABF substrate supply gap will break -27% for the first time in 2027. Upstream, Ajinomoto—the core raw material supplier for ABF films—is considering price hikes of at least 30%, and T-Glass low thermal expansion glass fiber shortages reached 50% in late 2025 to early 2026. While MLCCs show even greater elasticity in per-unit usage, this comparison highlights a core trend: supply tension across the compute infrastructure supply chain is systemic, with MLCCs being the earliest and most elastic segment.
Strategic Window for Domestic Substitution and Long-Term Industry Outlook
With global high-end MLCC supply remaining tight and Japanese and Korean leaders needing time to expand capacity, domestic manufacturers face a crucial market entry window. Supply chain security driven by geopolitics, combined with ongoing industry price cycles, presents an unprecedented strategic opportunity for domestic substitution.
On the supply side, Japanese and Korean leaders are focusing expansion on high-end, high-margin products, causing spillover effects as mid- and low-end orders are released. CITIC Securities believes domestic firms stand to benefit from overseas leaders’ capacity crowding in AI, as evidenced by domestic MLCC sector revenue growth of 19%–46% in Q1.
From a technology and capacity perspective, substantial progress in domestic substitution is accelerating. Fenghua Advanced’s Xianghe Industrial Park high-end capacitor base was fully built by the end of 2025 and completed in April 2026, with mid/high-voltage, high-temp, high-capacity MLCCs now used in AI servers. Sanhuan Group, leveraging 100% self-supply of ceramic powders, has monthly MLCC capacity exceeding 90 billion units, with high-capacity products accounting for 70%. The company has entered the Tesla supply chain and successfully supplied NVIDIA AI servers.
However, breakthroughs by domestic firms in high-end MLCCs for AI servers remain in early stages. Core technologies like ultra-fine dielectrics (<1μm), high-reliability formulations, and others still lag behind Japanese competitors, and leading end brands remain cautious about adopting domestic brands. Additionally, localization rates for critical equipment such as high-precision laminators and high-temperature sintering furnaces are below 20%, with delivery cycles for imported equipment measured in years. Thus, the main benefit for domestic substitution currently comes from the supply gap created by Japanese and Korean leaders’ focus on high-end AI, rather than the highest-end MLCC categories for AI servers, where Japanese and Korean dominance is unlikely to be challenged in the short term.
Looking at the length of this supply-demand cycle, multiple institutions offer consistent long-term assessments. Murata’s president points out that AI investment will remain strong for the next 3–5 years, with next-generation AI chips requiring tens of times more high-end MLCCs. Goldman Sachs forecasts MLCC market size will grow about 4.3 times from 2025 to 2030, with this cycle representing a 3–5 year long-term upswing rather than a traditional short-term supply-demand pulse. CICC projects AI server MLCC demand will reach 72.6 billion units in 2026, up 87% year-on-year, and 136.7 billion units in 2027, up 88%.
The supply-demand balance turning point is expected in the first half to mid-2027, marking the peak of this mismatch cycle. Expansion projects launched by industry leaders like Murata in early 2026 are expected to start coming online from mid-2027 to early 2028, while mass production ramp-up for NVIDIA’s Rubin platform is accelerating. This lag between capacity release and demand surge underpins the sustained nature of this boom cycle.
Conclusion
From Taiyo Yuden’s CEO warning of "alarming" demand, to Murata’s June 9 announcement of the third price hike of the year, and MLCCs rising to the third spot in the AI server BOM—these signals together draw a clear conclusion: MLCCs are no longer just the "rice of the electronics industry" following consumer electronics cycles, but are now a structural bottleneck in AI compute infrastructure investment that cannot be ignored.
Much like the supply-demand narrative for GPUs and HBMs, MLCCs face the contradiction of oligopolistic supply and exponential demand growth. With a single AI rack consuming nearly 600,000 capacitors and the value per capacitor in high-end applications rising, this industry—long seen as a supporting role—is undergoing an AI-driven value re-evaluation.
