MLCC value surges by 182%, NVIDIA's new rack reveals the next "shortage king" in AI

Title: Goldman, Morgan Stanley Sound the Alarm: MLCCs Become the Scarcest Puzzle Piece in the AI Supply Chain, Nvidia’s Next-Gen Rack Usage Jumps 182%

Author: BitGo Finance

Compiled by: Peggy

Editor’s note: Supply bottlenecks in AI infrastructure are spreading beyond GPUs, memory, data centers, and power systems, moving further down into more fundamental hardware components. Goldman and Morgan Stanley’s latest attention is turning to MLCCs—multilayer ceramic capacitors, long viewed as ordinary passive components.

In AI servers, MLCCs stabilize current, filter noise, and are key components that help ensure high-speed chip operation. As Nvidia’s new-generation rack architecture drives up the MLCC usage per single rack, the value share is rising rapidly. Goldman expects the AI server MLCC market to grow more than fourfold between 2025 and 2030, while industry capacity will grow at only slightly above 10% annually; the resulting supply-demand mismatch is becoming the core variable of this round of market action.

More importantly, the pricing cycle has already started. Japanese leaders such as Murata and Taiyo Yuden have taken the lead in raising prices, and Japanese export data is also beginning to validate the strength of demand. For capital markets, the logic of MLCCs is not complicated: demand comes from AI servers and high-end automobiles, supply expansion is constrained, and price increases can significantly amplify profit elasticity.

From chips to capacitors, pricing power in the AI supply chain is transferring to more segmented—and more hidden—layers. Whether MLCCs will become “the next storage chip” still depends on whether AI server demand can continue to materialize; but what is certain is that this once-unremarkable core component has already reached the starting point of a new cycle of rising both volume and prices.

Below is the original article:

Supply bottlenecks in the AI arms race are igniting opportunities across various hardware sectors. After data centers, energy infrastructure, and storage chips became the focus of capital attention, Wall Street giants Goldman Sachs and Morgan Stanley, in their latest reports, pointed to a long-undervalued fundamental component: multilayer ceramic capacitors (MLCCs). Both institutions expect MLCCs to become the next “key battleground” for rising volume and prices, and this AI-driven growth cycle could be the largest one in history.

Goldman analyst Daiki Takayama notes in the report that the MLCC market for AI servers is expected to grow from about 215 billion yen (about $1.4 billion) in fiscal 2025 to about 920 billion yen (about $5.8 billion) in fiscal 2030, an increase of more than fourfold, with a compound annual growth rate of 34%. Goldman is direct in stating that the current AI-driven MLCC cycle “will be the largest and longest-lasting in history, and we believe it is still in its early stages.”

MLCC: The “Invisible Heart” That Keeps AI Servers Running

MLCC (Multi-layer Ceramic Capacitor, multilayer ceramic capacitor) can be understood as an extremely miniaturized, ultra-fast charge/discharge unit. Unlike ordinary batteries that store large amounts of energy and release it slowly, MLCCs store only a small amount of energy, but they can complete charge/discharge within extremely short periods—milliseconds or even shorter. Its core function is to smooth power fluctuations and filter noise: it absorbs instantaneous voltage spikes, or quickly replenishes current when voltage drops, providing stable current to sensitive chips and blocking electrical interference that could disrupt digital signals.

The operating characteristics of AI servers make MLCCs indispensable. When AI models perform large-scale computing, the power demand of processors can surge in microsecond-level bursts, then quickly fall back to near zero after computation ends. The power system itself is unable to respond quickly enough to such drastic fluctuations. MLCCs are typically installed directly near the AI chips; when power peaks occur, they release energy instantly to prevent the server from crashing. Because AI chips such as Nvidia GPUs need to process billions of tasks at the same time, a top-tier AI server rack may require as many as 600,000 MLCCs working together to maintain system stability.

Goldman analyst Nelson Armbrust further points out that MLCCs have become the third most expensive component in an AI server bill of materials (BOM), after GPUs and memory. At present, the overall MLCC market is about $15 billion, of which the server-related market is about $1.3 billion and is expanding at an 80% compound annual growth rate. By contrast, demand growth in other application areas such as automotive and smartphones has clearly slowed. Daiki Takayama expects the MLCC cost share in an AI server BOM to gradually rise from about 0.5% to about 1%.

Structural Supply-Demand Contradiction: Annual Capacity Growth Is Only 10%, Unable to Withstand a Fourfold Demand Shock

The core factor driving market attention is that the MLCC industry is facing severe structural supply-demand imbalance. Goldman analyst Allen Chang clearly states that the industry’s annual capacity growth rate is only slightly above 10%. In addition, because both equipment and materials rely heavily on manufacturers’ internal production, the pace of capacity expansion is constrained by internal engineering resources, making it difficult to accelerate significantly. However, the demand shock from AI servers is in an entirely different league. Goldman expects that between fiscal 2025 and fiscal 2030, MLCC demand driven by AI servers will grow by about 4.3 times.

What worries the market even more is that demand for high-voltage, high-capacitance MLCCs driven by electric vehicle electrification remains strong, and MLCC usage per vehicle continues to increase. The two major demand pillars—AI servers and electric vehicles—are jointly consuming limited incremental capacity that was already scarce. This also explains why, even if consumer electronics demand declines, related customers are still actively seeking long-term supply agreements to hedge the risk of future shortages.

Signs of tightening in the market are already appearing on multiple fronts: delivery lead times for high-end MLCCs (high-capacitance, high-voltage specifications) have exceeded 20 weeks; due to stockpiling and repeat orders, the prices in spot and distribution channels for low-capacitance and consumer-grade MLCCs have risen by 20% to 40%; and the prices of key raw materials such as nickel and silver remain high, putting pressure on costs across various product categories.

The Price Increase Cycle Has Officially Begun: Japan’s “Two Giants” Lead Price Hikes, Official Data Confirms the Trend

Price signals are strengthening rapidly. The price hikes initiated by Japan’s two leading companies, Murata Manufacturing (Murata Manufacturing) and Taiyo Yuden (Taiyo Yuden), mark the official start of the MLCC price increase cycle. Since April 1 of this year, Murata has raised MLCC product prices used in AI server and high-end automotive applications by 15% to 35%. Taiyo Yuden has also informed customers that it will adjust prices for multiple product lines starting in May, including MLCCs, inductors, RF devices, FBAR/SAW devices, and aluminum electrolytic capacitors. The reason is that the costs of various raw materials such as precious metals have continued to rise.

Trade statistics released by Japan’s Ministry of Finance on May 28 have validated this upward pricing trend from a macro perspective. The data shows that in April, the average export price of MLCCs increased 3% month-over-month and 16% year-over-year; export volume rose 10% year-over-year; and export value surged 28% year-over-year. Goldman believes this data confirms the signals released in recent financial reports from Japanese MLCC manufacturers: all companies confirmed that order momentum remains strong.

From the timeline of the entire AI supply chain, Goldman’s analytical framework shows that MLCC price increases are lagging behind those of key AI components such as DRAM, NAND flash, ABF substrates, and copper-clad laminate (CCL). Therefore, Goldman judges that among all AI components and materials, MLCCs have the longest price-increase runway and the strongest durability. Goldman has raised its forecast for year-over-year MLCC price changes in 2026 from the previously approximately 0% to a range of 0% to +5%, and emphasized that the actual future price increase could be far higher than this level.

Incredible Profit Elasticity: A 5% Price Hike Could Raise Operating Profit by Up to 37%

For investors, profit elasticity arising from MLCC supply-demand mismatch should not be underestimated. Daiki Takayama estimates that even a 5% increase in product prices could theoretically push Murata’s operating profit in fiscal 2027 up by about 13%, and Taiyo Yuden’s operating profit by as much as 37%.

Goldman expects Murata’s sales in fiscal 2027 will reach 1.05 trillion yen (about $6.6 billion), up 13% year over year; Taiyo Yuden’s sales will reach 286 billion yen (about $18 billion), also up 13% year over year. Goldman maintains “Buy” ratings on Murata, Taiyo Yuden, and TDK. Its constructed Asian MLCC thematic stock portfolio has recently begun to strengthen, but compared with other popular AI themes, there is still clear room for catch-up.

Morgan Stanley Breaks Down Nvidia’s New Rack: Peripheral Components’ Importance Rises, MLCC Usage Soars 182%

Another major catalyst comes from Nvidia’s next-generation Vera Rubin AI racks. After dissecting Nvidia’s latest VR200 racks, Morgan Stanley found that the importance of peripheral components in the new BOM is rising rapidly.

The value of MLCCs in a single rack has increased from about $1,530 in the previous GB300 era to approximately $4,320, a jump of 182%. Although the absolute value remains lower than that of GPUs, memory, and PCB, the growth rate among peripheral components is exceptionally prominent.

Morgan Stanley’s channel research also shows that MLCC usage on compute boards and switch boards has increased significantly, with the increase more pronounced on compute boards. In addition, the newly introduced BlueField and ConnectX modules will further raise the total MLCC usage per rack. To some extent, this explains why current demand for MLCCs in high-end AI servers is so strong, and has prompted multiple ODMs to proactively stock up in preparation for mass production and delivery of Rubin racks in the second half of 2026.

Morgan Stanley’s teardown of Nvidia’s Vera Rubin rack shows the following changes in the value of key components:

Market intelligence indicates that in the infrastructure arms race of the AI supercycle, the sequential rotation of supply bottlenecks has produced wave after wave of market winners. Goldman Sachs’ latest assessment describes MLCCs as “the new storage chip”—a passive component subsegment that is standing at the beginning of a cycle of rising volume and prices.

As demand from AI servers and Nvidia Rubin racks brings an exponential shock, lead times for high-end MLCCs have already exceeded 20 weeks. Japanese industry leaders have initiated price hikes, official export data remains strong, and all signals point to one conclusion: this AI-driven MLCC supercycle is just getting started.

Original link