MLCC Value Reassessment: How the Explosion of AI Servers Is Driving a Super Cycle in Passive Components?

In the booming AI computing power investment landscape of 2026, market attention is highly focused on the supply and demand game between GPUs and HBM memory chips, but a deeper "invisible bottleneck" is emerging. Known as the "grain of electronic industry rice," multilayer ceramic capacitors (MLCCs) are rising from traditional passive components to become a key variable in the cost structure of AI servers.

In May 2026, Japan’s passive component giant Murata issued an industry warning: the demand for MLCCs for high-end AI servers has reached a "stunning" level, with capacity approaching its limit. The global high-end MLCC supply chain is facing unprecedented supply pressure. When a single AI cabinet consumes nearly 600k MLCCs, and the value of each capacitor in high-end applications continues to rise, the passive component industry, long regarded as a "supporting role," is undergoing a structural value reassessment driven by AI.

TrendForce data shows that in 2026, the global server shipment growth rate has been revised upward from 14.1% to 17%, with AI server annual growth exceeding 28%. The double-digit growth trend is expected to continue into 2027. Cinda Securities predicts that global AI server shipments in 2026 will reach approximately 3.7 million units, a year-on-year increase of 51.3%, and will maintain double-digit growth through 2027 and 2028. Industry consensus is highly aligned—hardware competition for AI computing infrastructure is accelerating across the board, and MLCCs are becoming an unavoidable core beneficiary in this process.

Comparison Chart of MLCC Value in AI Servers

| Platform/Type | MLCC Usage per Unit (pieces) | MLCC Value (USD) | BOM Rank | | --- | --- | --- | --- | | Ordinary Server | About 2,000-4,000 | About 60-120 | Outside top 15 | | NVIDIA GB300 | About 30k | About 1,530 | About 6th-8th | | NVIDIA VR200 NVL72 | About 600k | About 4,320 | 3rd place |

Data sources: Murata public disclosures, Morgan Stanley BOM analysis of NVIDIA VR200 NVL72 cabinet (May 2026), Goldman Sachs research reports. The data for ordinary servers is industry average estimates; AI server data corresponds to NVIDIA platform specifications.

Explosive Growth in AI Server Shipments and Exponential Demand for MLCCs

Understanding the current value restructuring in the MLCC sector first requires establishing basic quantitative benchmarks for the growth of the AI server market. TrendForce has revised the 2026 global server shipment growth rate from 14.1% to 17%, with AI servers growing at over 28% annually, and this double-digit growth is expected to persist into 2027. This data reflects a sustained acceleration in AI infrastructure development over the past year.

This expansion in shipment volume is only the first driver of demand growth. More critical is the geometric increase in MLCC load per device. Japan’s leading manufacturer Murata’s comparative data vividly illustrates this scale difference: a standard server requires only 2,200 to 4,000 MLCCs per unit, while NVIDIA’s GB300 AI server consumes about 30,000; in March 2026, NVIDIA officially released the new VR200 NVL72 compute cabinet, with MLCC usage reaching 440k to 600,000 pieces. This means that a high-end AI cabinet consumes MLCCs at a level dozens or even hundreds of times higher than traditional servers.

Industry estimates of total demand also show astonishing growth. CICC estimates that in 2026, the total MLCC demand for AI servers will reach 72.6 billion pieces, an 87% increase year-on-year; in 2027, demand will further rise to 136.7 billion pieces, an 88% increase. CITIC Securities estimates that global server MLCC shipments could continue expanding beyond 4 trillion pieces by 2030, with an average annual compound growth rate of about 40%. The root of this explosive growth lies in the architectural upgrade of AI servers from traditional single-motherboard designs to high-density cabinet-level computing platforms. Each additional GPU or HBM chip requires tens to hundreds of extra MLCCs.

From the perspective of high-performance computing power density evolution, this trend has deep technological inevitability. For example, NVIDIA’s Rubin platform single board uses nearly double the MLCCs of its predecessor, reaching 12,000 pieces. The generational leap in power density directly translates into a multiplication of passive component usage, with each upgrade in computing power demanding a proportional increase in capacitor configurations.

From "Supporting Role" to "Main Player": MLCC’s Rise as the Third Cost Driver in AI Servers

The explosive demand volume only reflects one dimension of change. The real driver behind the valuation reassessment of the MLCC industry is its rising position in the material bill of quantities (BOM) of AI servers.

Goldman Sachs analyst Nelson Armbrust recently pointed out in a research report that in the current BOM of AI servers, MLCCs have risen to become the third-largest cost item, behind only GPUs and memory chips. This conclusion has been widely cited and recognized in the global electronic component industry.

Morgan Stanley’s BOM analysis of NVIDIA’s VR200 NVL72 cabinet provides more precise quantitative support. The value of MLCCs per rack is about USD 4,320, a 182% increase from the previous generation GB300’s approximately USD 1,530. This surge in value results from both increased usage and higher unit prices—a "double boost" of volume and price.

Looking at the overall industry scale, the current global MLCC market is about USD 15 billion, with the AI server segment accounting for roughly USD 1.3 billion, growing at a robust compound annual rate of 80%. Meanwhile, growth in other key sectors like automotive and mobile phones has slowed. Goldman Sachs’ latest report suggests that the AI-driven MLCC supercycle has just begun, with the market expected to grow approximately 4.3 times from 2025 to 2030. This growth rate is rare in the passive component industry and signifies a historic shift in MLCC’s value positioning.

In contrast, demand from traditional sectors such as smartphones and consumer electronics has noticeably slowed. This indicates that the current MLCC market cycle features a structural shift—AI infrastructure is replacing consumer electronics as the new growth anchor for MLCC demand.

Global Leading Players: Oligopoly and Capacity Structural Constraints

The global MLCC market exhibits a typical "oligopoly + domestic catch-up" structure, with the top five players (CR5) accounting for over 80% of the market in 2026. The technological and capacity barriers in the high-end segment are extremely high.

Global MLCC Supply-Demand Tight Balance Pathway Diagram

In terms of market segmentation, the first tier is dominated by Japanese and Korean manufacturers. Murata’s market share is approximately 25%-34%, with about 70% of its high-end segments (like AI servers). Samsung Electro-Mechanics holds about 18%-24%. TDK and Sunlord together account for about 15%-20%. The four leading Japanese firms collectively hold about 85% of high-margin segments such as AI servers and automotive-grade products. Taiwanese manufacturers (like Walsin, Walsin, and Microchip) hold roughly 10%-15%, mainly targeting mid-to-high-end general and consumer markets. Chinese domestic players (such as Sanhua, Fenghua, and Weirong) have a combined market share of about 10%-12% and are accelerating penetration into AI and automotive high-end segments.

This highly concentrated supply pattern implies systemic capacity bottlenecks. On the demand side, leading companies have already begun raising prices. In June 2026, the MLCC industry experienced its third annual price increase, with Murata, Samsung, TDK, and Panasonic all raising quotes simultaneously. The highest increases in high-end automotive and AI categories reached 35%, while general consumer types increased by 6%-30%. Manufacturers are shifting capacity toward AI and automotive segments, with supply of standard MLCCs continuing to tighten. The capacity differentiation between high-end and general categories has become essentially established.

In terms of delivery cycles, Murata’s high-end MLCC production lines operate at 95% utilization, with lead times extending beyond 20 weeks; some critical models are already in limited order status. Sunlord’s high-capacity materials have lead times of 16-24 weeks, with limited inventory. Samsung’s delivery times have extended beyond 18 weeks, with spot prices rising month by month.

| Manufacturer | Global Market Share | AI/High-End Segment Share | Latest Developments | | --- | --- | --- | --- | | Murata | 25%-34% | About 70% (AI servers) | April price hike 15%-35%; June 9th price increase letter, AI/automotive MLCC up 10%-40%, effective July 1 | | Samsung Electro-Mechanics | 18%-24% | Strong in automotive/5G | Consumer MLCC planned price increase 5%-10%; AI high-capacity models up 30% separately | | TDK & Sunlord | 15%-20% combined | Automotive/Industrial high-end | May 1 price increase 6%-15%; CEO warns demand has reached "stunning" levels | | Fenghua High-Tech | Leading domestic player | Accelerating into AI/automotive | Shihhe Industrial Park project completed in April 2026; high-voltage/high-temperature MLCC in mass production for AI servers |

Looking ahead, the release of high-end capacity is expected to lag behind downstream demand growth. Expanding high-end MLCC lines typically takes 18-24 months, relying on a few Japanese machinery suppliers, with supply elasticity constrained by rigid factors. This structural characteristic is highly similar to the supply-demand logic of HBM memory chips.

Capacity Expansion Pace and Supply-Demand Gap Evolution

Despite active capacity expansion by industry leaders, there remains a significant time lag between capacity release and demand explosion. Japanese and Korean manufacturers are intensively expanding: Murata plans to invest about 80 billion yen, with a new plant in Izumo City, Shimane Prefecture, coming online in 2026, increasing AI capacity share from 30% to over 45%. Samsung’s Tianjin plant is expanding by about 20%, and a new plant in the Philippines will roughly double existing capacity, focusing on AI server and automotive MLCC production. Sunlord plans to invest around 600k yen over five years to expand capacity, but its CEO views this as "necessary acceleration" rather than proactive foresight.

However, these major expansion plans still have a considerable lead time before full capacity release. In the second half of 2026 and into 2027, the shortage of high-end high-capacity MLCCs is expected to be between 15% and 20%, possibly expanding to 30% in 2027. The supply of general-purpose consumer products is also tightening due to high-end capacity constraints, with manufacturers actively signaling that supply of standard MLCCs will remain tight for the long term.

From the perspective of upstream core materials, supply constraints are even more profound than capacity. Morgan Stanley’s June 10 report pointed out that the real bottleneck in the MLCC supply chain lies in upstream nano-scale ceramic powders—high-end dielectric powders with particle sizes around 100 nm and purity of 99.99%. Historically dominated by Japanese firms like Sakai Chemical, Chinese companies like Guocera Materials have made breakthroughs, capturing about 80% of domestic market share and securing key clients like Samsung and others. However, ultra-fine powders (≤80nm) and 5N-grade (99.999% purity) powders are still in validation or pilot stages, not yet capable of fully replacing high-end imported powders. Constraints at the upstream material level further narrow the capacity expansion elasticity of high-end MLCCs and prolong the duration of supply-demand mismatches.

From ABF Substrate to MLCC: Structural Transmission of Computing Investment

The boom in MLCCs is not an isolated phenomenon but part of a comprehensive supply chain transmission from core chips to underlying components in AI infrastructure. Within this chain, ABF substrates offer a reference perspective—both exhibit similar supply-demand mismatches, though their market sizes and industry impacts differ significantly.

ABF substrates are critical interconnects between CPUs, GPUs, and external circuits, playing an irreplaceable role in advanced packaging. Industry research firm IEK estimates that the global ABF substrate market will reach about USD 10.02 billion in 2026, expanding at a CAGR of approximately 22.9% from 2024 to 2028. Technological evolution shows that NVIDIA’s Rubin and Rubin Ultra platforms use ABF substrates sized at 100×91mm² and 153×77.5mm², with layer counts increasing from 12-14 to 18-20, and unit area consumption reaching 5-10 times that of traditional PC substrates.

Both ABF substrates and MLCCs face similar structural constraints: escalating technical specifications lead to significantly increased unit consumption, oligopolistic dominance by Japanese and Korean firms, and expansion cycles of 12-24 months. In Q1 2026, top ABF substrate manufacturers’ utilization rates have risen from 75-80% in Q3 2025 to around 90%. HSBC models forecast that the ABF substrate shortage could first break through -27% in 2027. Extending this view to upstream materials—such as the core raw material for ABF films, with companies like Asahi Kasei considering price hikes of at least 30%, and T-Glass’s low-thermal-expansion glass fiber shortages reaching 50% in H2 2025 to H1 2026—underscores that the entire computing infrastructure supply chain faces systemic tightness, with MLCCs being one of the earliest and most demand-elastic links.

Strategic Window for Domestic Substitution and Long-Term Industry Outlook

Against the backdrop of persistent tightness in high-end MLCC supply and the time needed for Japanese and Korean giants to expand capacity, domestic manufacturers face a critical market entry window. Geopolitical factors driving supply chain security, combined with the industry’s ongoing price increase cycle, create an unprecedented strategic opportunity for domestic substitution.

On the supply side, the expansion efforts of Japanese and Korean leaders are increasingly focused on high-margin high-end products, leaving mid- and low-end orders to spill over. CITIC Securities believes domestic firms are poised to benefit from overseas giants’ capacity squeeze driven by their focus on AI. Early industry performance reflects this—first-quarter revenues for domestic MLCC companies grew by 19% to 46%.

From a technological and capacity-building perspective, substantive progress in domestic substitution is accelerating. Fenghua High-Tech’s Xianghe Industrial Park high-end capacitor project was fully completed by late 2025, with mass application in AI servers by April 2026. Sanhua’s vertical integration of ceramic powders with 100% self-sufficiency has enabled monthly MLCC capacities exceeding 90 billion pieces, with high-capacity products accounting for 70%, entering Tesla’s supply chain and successfully integrating into NVIDIA’s AI server supply chain.

However, breakthroughs in high-end MLCCs for AI servers by domestic firms are still in early stages. Ultra-fine dielectric powders (