The Component Signal #22 — The MLCC Allocation Crisis: Silver, AI Rack Density, and the PDN-to-EMI Cascade

Who We Are

POCONS USA supplies board-level EMI shielding direct from the manufacturer — shield cans, SMT shield clips, precision spring contacts, and custom metal stampings. We're a US company: commercial operations and customer support run from San Diego, while our parts are manufactured in Korea under IATF 16949, with custom shields in roughly 2–3 weeks. Because production sits in Korea, POCONS USA tracks the Asian electronic-component supply chain closely — and this newsletter passes that signal forward.

Supply Chain Alert

The Causal Chain Behind the 2026 MLCC Squeeze

The current MLCC shortage is not a single-point disruption — it is a causal cascade with three reinforcing legs: AI rack density driving structural volume demand, capacity reallocation by the dominant supply base, and a metals-cost shock that has repriced the entire passive component stack simultaneously. Understanding the mechanism matters because the exit timeline depends on all three legs easing, and right now none of them is.

Leg 1: Rack-level demand density

AI servers use ten to fifteen times the number of MLCCs compared to general-purpose servers. A single NVIDIA GB300 rack consumes roughly 440,000 MLCCs. AI servers impose stringent specification requirements — high voltage, high capacitance, high temperature resistance, and low equivalent series resistance (ESR) — meaning these are not standard X7R decoupling parts that any mid-tier manufacturer can substitute in. High-end MLCC manufacturing is complex and yields are significantly lower than those for standard products, maintaining a tight balance on the supply side.

Leg 2: Capacity reallocation and lead-time deterioration

Major Japanese and South Korean suppliers have shifted their production capacity toward components for AI applications, steadily constraining the supply flexibility of consumer MLCCs quarter by quarter. The result is a lead-time cliff. Per TrendForce analyst Chen Weisheng, lead times for AI high-capacitance MLCCs have stretched from 8–12 weeks to 16–24 weeks, with some specifications beginning to see order restrictions. Sourcing desks in Taiwan and mainland China are reporting figures toward the upper end of that range or beyond. Lead times now exceed 20 weeks for high-end grades, with supply constraints expected to persist through 2027. Murata management revised its forecast for the compound annual growth rate of AI server MLCCs from 2025 to 2030 upward from 18% to 30% at the end of 2025 — which is guidance, not capacity on the floor today. Murata's capacity expansion announcements suggest meaningful new output arriving no earlier than late 2026.

The book-to-bill ratio confirms the imbalance is structural, not seasonal. Per TrendForce data, the overall MLCC industry book-to-bill ratio rose from 0.89 in March to 0.92 in April 2026, with leading manufacturers including Murata, Taiyo Yuden, and TDK maintaining BB ratios consistently above 1, indicating that order intake is outpacing shipment speed.

Leg 3: The repricing cascade — Taiyo Yuden opens; the industry follows

On April 15, 2026, Taiyo Yuden notified customers of price increases on a range of passive components — MLCCs, inductors, ferrite beads, RF components, and selected aluminum electrolytic capacitors — effective May 1. Prices of high-capacity products for automotive and server applications were raised by 10–30%, and consumer-grade standard products by 6–13%. This was the opening move of a broader repricing cycle, not an isolated event. Murata reportedly raised prices for MLCC products used in AI servers and high-end automotive applications effective April 1, having expanded increases to MLCCs after production capacity reached full utilization; the Liberty Times reported those increases ranging from 15% to 35%. Goldman Sachs subsequently revised its 2026 MLCC price forecast from "flat" to an increase of 0–5% — that forecast carries revision risk given the pace of announcements since it was issued. As of early June, at least one major Korean manufacturer has notified agents to suspend quoting prices, with the market broadly expecting a synchronized adjustment across its full product line.

The metals cost driver behind these hikes is quantifiable. Silver accounts for approximately 42–58% of MLCC production cost, and per data from the Taiwan Economic Journal, the London Silver Fix price surged from $36.2 per ounce in July 2025 to $103.2 in January 2026 — an increase of nearly 185% in six months. Murata has pointed directly to silver as a primary cost driver, noting its widespread use across solar panels, electric vehicles, semiconductors, AI devices, and medical equipment, where demand has surged in recent years. Panasonic's customer notification similarly cited continued increases in the prices of precious metal raw materials including silver and copper.

Leg 4: Secondary pressure — the tantalum substitution effect

There is a second-order demand vector piling onto the same constrained manufacturers. Panasonic notified distributors and customers that prices for a wide range of its tantalum capacitors would increase 15–30% starting February 1, 2026. KEMET, the global leader in tantalum capacitors with a market share exceeding 40%, is facing severe supply-demand imbalance driven by the extensive adoption of tantalum capacitors in ASIC servers and tightening upstream tantalum ore resources. The consequence is a substitution demand that feeds directly back into the MLCC shortage: tantalum shortages and rising costs have prompted many design teams to adopt hybrid "tantalum + MLCC" solutions, pushing additional demand toward MLCCs. Designs using tantalum alone or in combination with high-end MLCCs are extremely capacity-intensive, with Commercial Times noting that the stage is set for a new wave of widespread tantalum capacitor price hikes that further compresses design teams' alternatives.

What this means for programs outside the hyperscaler tier

Inventory levels at the distribution tier provide a temporary buffer, but as tier-one OEMs consume available safety stock, smaller players face increased competition for remaining allocations. As Astute's Franchise Marketing Manager Damian Semple stated: "The divergence between AI-driven demand and general industrial recovery is creating a dangerous procurement gap for standard hardware manufacturers. Firms must secure allocations for high-capacitance MLCCs now, as lead times will likely deteriorate further when automotive demand aligns with the current server build-out."

The EMI dimension of this squeeze is worth stating explicitly. Decoupling MLCCs — particularly the 10 µF and 100 µF X5R/X7R grades in 0402 and 0603 packages used around power delivery networks on RF and mixed-signal boards — are among the constrained specifications. A board that loses decoupling coverage because the BOM-specified part is on a 30-week tape-out is not merely a supply problem: the resulting PDN impedance rise at tens-to-hundreds of MHz becomes a radiated-emission problem. An EMI shield can attenuate the symptom at the aperture, but it cannot compensate for the conducted noise path that underlies it. The fix is in the decoupling layout; the shield is last-resort mitigation.

POCONS connection: Board-level EMI shield cans are seeing increased RFQ activity from programs experiencing exactly this failure mode — decoupling gaps exposed late in DVT that cannot be resolved before certification. A shielding solution is not a substitute for correct PDN design, but if your schedule has already slipped and you need to contain a radiated-emission exceedance at a specific IC cluster, POCONS USA can turn a custom shield can in roughly two to three weeks from San Diego. That is faster than most constrained MLCC replenishment timelines right now. Contact us for a free PCB layout and shielding review.

Price Watch

Silver Is Doing the Work Your P&L Hasn't Priced In

Silver is not a precious-metals story this cycle — it is a passive-components cost story. It accounts for 42–58% of MLCC bill-of-materials cost (electrode paste and termination metallization), which means every move in the London spot price flows directly into component pricing with roughly a one-to-two quarter lag as supplier contracts roll.

Where spot sits right now. As of June 12, 2026, London spot silver was trading at approximately $68.94/oz. That is a significant pullback from the February peak: silver briefly exceeded $120/oz earlier in 2026 before consolidating, with the February high widely reported in the $76–$84 range. For procurement purposes, the relevant reference point for current MLCC contract negotiations is the Q1 spike: London spot reached $76.204/oz in early February 2026 — the level at which most Q2 price-increase letters were benchmarked. Even at current levels, silver is more than $27/oz higher than it was one year ago. That year-over-year delta is the number your MLCC supplier's finance team is using to justify line-item increases, not the mid-June spot.

The Goldman Sachs MLCC call. Goldman Sachs raised its 2026 MLCC price forecast from "flat" to an increase of 0%–5%. That forecast deserves careful reading. It is a full-year blended average across all MLCC grades. Taiyo Yuden has already announced 6%–13% hikes for mid-to-low capacitance grades. Spot-market data for high-capacitance (≥10 µF, X5R/X7R, 0402–0603) and automotive AEC-Q200-qualified parts is already showing 10–20% premiums over Q4 2025 distributor price books — meaning the Goldman blended forecast is a floor, not a ceiling, for the grades that matter in AI-server and automotive design.

Supply deficit: the structural underpinning. The silver market is heading into its sixth consecutive year of supply deficit, with the shortfall projected at approximately 46 million ounces. By mid-2026, structural supply deficits have intensified, drawing down above-ground stocks to decade lows. Mine supply is inelastic on a 12–18 month horizon; no major greenfield project brings meaningful ounces to market before 2028. That means any demand recovery in PV or electronics tightens the market fast.

The Engineering Cost Exposure: Two Design Cases

The gap between the Goldman forecast (+0–5% blended) and actual spot-market premiums (+10–20% on the grades you actually specify) concentrates in the highest-capacitance, highest-layer-count parts — exactly the decoupling capacitors that dominate AI-server GPU PCBs. Work through two representative boards:

Case 1 — GPU PCB (3,000 MLCCs)

A high-end GPU carrier board running ~3,000 MLCCs is not unusual; as GPU boards swell to 3,000 MLCCs each, supply tightness in 0402 parts is likely to persist. Assume a mix weighted toward 0402 X7R 10µF (power-rail decoupling) at a Q4 2025 distributor price of $0.012/unit for that grade. At 3,000 units, passive BOM cost for MLCCs alone was ~$36. A 15% blended price increase on the high-cap portion (assume 60% of the count) adds roughly +$3.24 per board. At a 10,000-unit GPU build, that is $32,400 in unforecasted MLCC cost — before any spot-buy premium for units currently quoting at 26–40 week lead time.

Case 2 — Automotive ECU (500 MLCCs)

A mid-complexity ADAS ECU runs ~500 MLCCs, predominantly 0402–0603 X7R AEC-Q200, distributed across power conditioning and signal-chain decoupling. Baseline cost ~$0.018/unit average (AEC-Q200 premium baked in). At 500 units, Q4 2025 baseline was $9.00. A 12% increase on AEC-Q200 automotive-grade parts adds **$1.08 per board**. At an annual ECU volume of 250,000 units, that is $270,000 in annual passive-component cost growth — from silver alone, before any allocation-driven spot-buy exposure. Long AEC-Q200 qualification cycles keep near-term supply tight, so automotive procurement cannot simply switch supplier to escape the price action.

The critical point: these are conservative estimates using the low end of the 10–20% spot-market range. If silver reverts toward the Q1 peak — Goldman Sachs projects silver averaging in the $85–$100 range in a base-to-bull scenario — the per-board cost multiplier tightens further, and the Q3/Q4 contract rollover becomes a material P&L event, not a rounding error.

The EMI Complication of Thin Decoupling Coverage

There is a circuit-level consequence hiding inside the cost-avoidance math. When procurement responds to MLCC price pressure by reducing count, substituting lower-capacitance values, or deferring board-layout adjustments to avoid redesign costs, power-distribution network (PDN) impedance rises at the frequencies that matter. The relationship is direct:

Z_PDN(f) ≈ 1 / (2π · f · C_eff)

Cutting C_eff by 20% raises PDN impedance by 25% at the same frequency. On a GPU power rail switching at 1–5 MHz, a 25% impedance rise translates to larger voltage excursions, which couple into adjacent signal layers and radiate from board edges and connector transitions. A shield can attenuate the resulting emissions — but it treats the symptom, not the cause. The fix is adequate decoupling at the source; the shield is a last line of defense, not a substitute for correct PDN design.

Rotation Read: High-Capacitance MLCC (≥10 µF, X7R/X5R, 0402–0603)

Status: 🔴 Allocation

Lead times on this grade family are quoting 26–40 weeks at major distributors. Automotive AEC-Q200 versions are tighter. Surging demand from vehicle electrification, AI infrastructure, and edge computing is placing sustained pressure on legacy supply chains. Murata's new Izumo plant — construction began in February 2024 and is set to open in 2026 — is the most tangible near-term capacity addition, but qualification of new fab output into customer AVLs typically takes two to four quarters after line qualification. Expect allocation conditions on high-cap grades to persist through at least Q1 2027. Spot-buy premiums of 10–20% are the current market-clearing price; long-term agreements with quarterly price adjustment clauses tied to a silver-index benchmark are the structural hedge, not one-time spot coverage.

Copper and tin context for completeness. LME copper spot was $13,480/t on June 9, 2026. Jefferies expects copper prices to stay elevated for longer than previously anticipated, citing an average annual supply deficit of 491,000 tons through 2030. For datacenter wiring and PCB copper weights, this is a sustained headwind on laminate and cable assembly costs. LME tin was trading at approximately $52,848/t (3-month, June 12, 2026 per LME day-delayed close). SAC305 solder (96.5% tin, 3% silver, 0.5% copper) is directly exposed to both the tin and silver vectors simultaneously — assembly-house surcharges on solder paste are already flowing through on some accounts.

Asian Manufacturer Spotlight

Murata and Taiyo Yuden: The Process Lock-In That Defines the 0402 Constraint

To understand the MLCC squeeze squeezing AI server builds right now, you have to understand that this is not a generic shortage. It is a process-technology barrier — and two Japanese manufacturers sit at the center of it.

What Murata Makes and Why It Is the Constraint

Murata Manufacturing is the world's largest MLCC supplier by volume and, more importantly for this issue, by process leadership at advanced nodes. In July 2025, Murata began the world's first mass production of the 0402-inch size (1.0 × 0.5 mm) MLCC with a capacitance of 47 µF. To put the engineering significance plainly: this new part delivers approximately 2.1 times the capacitance of Murata's previous 22 µF product in the same 0402-inch size. The parts — the GRM158R60E476ME01 (X5R, –55 to +85°C) and the GRM158C80E476ME01 (X6S, –55 to +105°C) — are both rated at 2.5 Vdc and target the decoupling rails immediately adjacent to AI accelerator ICs. Compared to Murata's conventional 0603-inch product at the same capacitance, this new capacitor reduces mounting area by approximately 60%.

The enabling technology is a layer count the rest of the industry cannot easily replicate: Murata launched a 47 µF 0402 MLCC with 800 layers — the densest MLCC stack commercially produced. Manufacturing complexity scales sharply below 0402, requiring photolithography-grade clean rooms and laser trim, which concentrates effective capacity among a handful of Japanese and Korean leaders and has extended lead times to 20 weeks in early 2026.

The demand driver is equally concrete. 0402 units are rising 16% annually, supported by AI servers that need 47 µF decoupling capacitors contiguous to 700 W GPUs. At the board level, the scale is extraordinary: a single AI server utilizing the NVIDIA GB300 platform requires approximately 30,000 MLCCs — thirty times the amount found in a standard smartphone — while a full server rack can consume up to 450,000 units.

Murata's own internal outlook has hardened accordingly. Murata management revised its forecast for the compound annual growth rate of AI server MLCCs from 2025 to 2030 upward from 18% to 30% at the end of 2025. On the pricing side, on March 17, 2026, Murata announced a price increase of 15%–35%, effective April 1, 2026, applying to AI server high-capacitance MLCCs, high-end automotive-grade MLCCs, and RF/microwave MLCCs. Driven by AI infrastructure demand, Murata's high-end production utilization exceeds 80%, with strong order growth expected and ongoing capacity expansion across Japan and Southeast Asia.

Taiyo Yuden: The Embedded MLCC Specialist Moving Into Server Infrastructure

Taiyo Yuden, Japan's third-largest MLCC supplier by market share, is executing a deliberate mix shift toward exactly the segments where capacity is tightest. Taiyo Yuden is undergoing a structural transformation to focus on higher-value markets such as automotive and communication infrastructure; with overall utilization around 85%, the company plans moderate capacity growth while improving profitability, and has commercialized embedded MLCCs for AI servers, including a 22 µF 0402 product optimized for close IC placement.

The financial signal is unambiguous. Taiyo Yuden's net profit for fiscal year 2025 surged 536% year-on-year, with its capacitor division's book-to-bill ratio reaching 1.31, and the company forecasts that AI server MLCC revenue will grow by approximately 80% in fiscal year 2026.

On pricing, Taiyo Yuden moved before most of its peers. In April 2026, Taiyo Yuden took the lead by raising prices for low-capacitance consumer and automotive MLCCs by 6% to 13%. That hike covers grades well below the high-capacitance AI server tiers — meaning the price pressure is not confined to exotic SKUs but is spreading into the mid-capacitance catalog that standard industrial and automotive boards depend on.

Taiyo Yuden's new server-capacity build is also on a deferred timeline. The server production capacity of Taiyo Yuden in South Korea is expected to be completed in 2027, meaning that at least before the first half of 2027, the supply-demand gap for high-end, high-capacitance MLCCs is difficult to be fundamentally alleviated. This is a significant data point: even Taiyo Yuden's own capacity expansion specifically targeting AI server demand will not come online until after the current acute phase.

What the Yield Math Actually Means

The reason capacity is so hard to expand quickly comes down to process economics at the sub-0402 node. Although AI servers consume only 2% to 3% of global MLCC units, they occupy nearly 10% of production capacity, because ultra-high-capacitance products have yields of only around 40% and production cycles that are twice as long.

Run the implication forward: a fab line nominally producing 1,000 units of high-capacitance 0402 MLCCs yields roughly 400 shippable parts. The same line running commodity 0603 units might yield 90%+. No amount of incremental capex fixes that ratio on a short timeline — the dielectric layer precision, sintering atmosphere control, and laser-trim tuning required at 800 layers are not problems that purchasing more equipment solves in one cycle. Proprietary technologies in ceramic powders, multilayer printing, and termination materials create high barriers to entry, ensuring that leading suppliers maintain dominance in high-performance AI MLCCs.

Korean Manufacturers in the Same Tier: A Constrained But Growing Position

Korean producers operating in the advanced MLCC tier face the same process demands as Murata and Taiyo Yuden — and increasingly the same customer pull. MLCC suppliers' capacity utilization rates have continued to recover since February 2026, with Japanese and Korean manufacturers aggressively shifting capacity from consumer-grade products toward high-value-added, high-end MLCCs. The overall industry book-to-bill ratio has climbed from 0.89 in March to 0.92 in April, with benchmark manufacturers maintaining BB ratios consistently above 1, indicating that order intake is outpacing shipment speed with a clear supply tightness trend.

Korean manufacturers competing in this tier share the same structural constraint: the clean-room and process-control requirements for advanced MLCCs are effectively identical regardless of geography. AI servers have stringent specification requirements for MLCCs — high voltage, high capacitance, high temperature resistance, and low ESR — and high-end MLCC manufacturing yields are significantly lower than those for standard products, maintaining a tight balance on the supply side.

The market-level picture, per TrendForce, is one of bifurcation: the MLCC market is experiencing a polarized pricing landscape in Q2 2026, characterized by robust AI demand versus sluggish consumer sentiment. Japanese and South Korean manufacturers are strategically focusing on high-end products, causing mid-to-low-end orders to spill over to Chinese firms. That spill-over does not relieve the high-end constraint — it confirms that the capacity reallocation away from commodity grades is deliberate and structural, not temporary.

The EMI Connection: What Decoupling Gaps Mean at the Board Level

This is where the process constraint becomes a board-design problem, not just a procurement problem. When a power-delivery network designed around 47 µF 0402 parts cannot be fully populated — because allocation is short or lead times exceed the build schedule — the designer either substitutes a lower-capacitance part in the same footprint or reduces the number of decoupling sites. Both paths increase supply-rail impedance at the frequencies where AI accelerator ICs switch.

An AI accelerator board can easily integrate tens of thousands of MLCCs across point-of-load converters, decoupling networks, and auxiliary circuits; this is not just a volume story, as data-center power-delivery networks favor high-CV, low-ESR MLC

Korean Supply Chain Intelligence

• Q3 2026 price negotiations closed early — and the signal is unambiguous for second-half allocation. By early May, some equipment makers had already completed their Q3 2026 price negotiations with MLCC suppliers, with prices recovering; as the majority of ODMs began a new round of negotiations in late May, the pivotal question became whether tightening conditions would drive a further rebound for automotive and consumer applications. For IATF 16949-qualified supply chains, this matters practically: contracts settled in that early-May window locked in allocations before the broader negotiation round tightened terms. Procurement teams that waited for late-May are negotiating into a seller's market.

• Korean and Japanese leaders are running high-end lines at capacity — and new output won't arrive until 2027. The utilization rate of high-end production lines has exceeded 90%, and delivery times for tight orders have extended to four months. Capacity expansion from Korean fab conversions is expected to release only in Q4 2026 at the earliest, with additional Korean server-grade output not completing until 2027 — meaning the supply-demand gap for high-end, high-capacitance MLCCs is unlikely to be fundamentally relieved before the first half of 2027. Lead times to bring new ceramic dielectric production lines online run 18 to 24 months from capacity commitment to first output — and the 2023–2024 inventory correction cycle meant capacity investment was deferred precisely when AI server demand was beginning its structural lift.

• The lead-time asymmetry between standard X7R 0402 and high-CV automotive grades is now the defining procurement variable. Per TrendForce analyst Chen Weisheng, the global MLCC industry showed clear polarization in Q1 2026: AI-related high-end demand grew against the trend, while lead times for AI high-capacitance MLCCs stretched from 8–12 weeks to 16–24 weeks, with some specifications beginning to see order restrictions. Standard X7R 0402 parts at moderate capacitance remain comparatively available; it is the X5R/C0G grades in the ≥10 µF, 0402/0603 footprints — the exact decoupling values demanded by AI server power rails and AEC-Q200 automotive ADAS rails — where allocation is acute. Class 1 (C0G/NP0) devices captured 62.28% of the 2025 automotive MLCC market, reflecting the industry's insistence on stable capacitance across −55 °C to +150 °C cycles; Class 2 parts, whose capacitance drops under DC bias, are allocated primarily to lower-criticality rails, leaving safety-critical ADAS and braking circuits dependent on Class 1 supply.

• Silver cost inflation is structurally baked into Korean and Japanese production cost floors for the rest of 2026. Silver accounts for approximately 42–58% of MLCC production cost; as of February 5, 2026, London spot silver reached USD 76.204/oz — significantly above the 2025 Q3 average — sharply increasing manufacturing costs and leaving suppliers with limited room to absorb pressure without price adjustments. While the rise in silver is the visible trigger, the deeper cause is explosive demand for high-end MLCCs from AI servers, where per-GPU MLCC usage has skyrocketed over twenty-fold across four generations of Nvidia platforms. Korean producers, whose internal cost structures mirror this silver exposure, cannot de-link their automotive contract pricing from spot-metal moves — BOM owners should model silver as a forward cost input, not a one-time event.

• Chinese entrants are absorbing mid-to-low-end overflow — but the qualification wall for automotive grades remains intact. As demand for AI servers soars, leading Japanese and South Korean MLCC makers are shifting further upmarket, opening the door for Chinese competitors to gain share in lower-end segments despite lagging their rivals in advanced technology. As Japanese and South Korean manufacturers collectively raise prices and focus strategically on high-end products, the spillover of mid-to-low-end orders has opened a window for Chinese domestic manufacturers — Chinese firm Chaozhou Three-Circle Group reported Q1 2026 revenue of 2.68 billion yuan (~$396 million USD), up 46% year-on-year. However, Chinese entrants still rely on imported sintering equipment and high-purity powders sourced from Japanese suppliers, and technology barriers continue to limit Chinese manufacturers' ability to compete in premium segments, as advanced MLCC technologies requiring ultra-fine layer structures and specialized dielectric formulations remain dominated by established international suppliers with decades of development experience. The practical implication: Chinese-sourced 0603/0805 commodity parts may offer short-term relief on standard rails, but no qualified substitute for AEC-Q200 X5R or C0G 0402 automotive grades exists from domestic Chinese supply at scale today.

Design Corner

Power-Rail ΔV, di/dt Noise, and What Actually Radiates When Your Decoupling BOM Goes on Allocation

When high-capacitance MLCCs move onto allocation — as is happening now across 10 µF and 22 µF X5R/X7R grades in 0402 and 0201 packages — the first instinct is to flag the purchasing problem and move on. The EMI consequence, however, is quantifiable and it arrives before the product ever reaches a pre-compliance chamber.

Step 1: How Capacitance Loss Raises Rail ΔV

The governing equation for transient voltage deviation on a power rail is:

ΔV = I · Δt / C

Where:

• ΔV = peak voltage deviation on the rail (V)
• I = load step current (A)
• Δt = duration of the current transient before the regulator responds (s)
• C = total effective decoupling capacitance at the point of load (F)

Take a representative AI inference accelerator drawing 8 A peak load steps, with a regulator loop response time of 500 ns, and a target decoupling bank of 100 µF (ten 10 µF 0402 X5R capacitors). Substituting:

ΔV = 8 A × 500 × 10⁻⁹ s / 100 × 10⁻⁶ F ΔV = 4 × 10⁻⁶ / 100 × 10⁻⁶ ΔV = 40 mV

Now suppose allocation forces you to populate only four of those ten capacitors — 40 µF effective. Re-substituting:

ΔV = 8 A × 500 × 10⁻⁹ s / 40 × 10⁻⁶ F ΔV = 100 mV

A 2.5× increase in rail deviation, sourced from the same load, same regulator, same board. The power-integrity problem is immediate; the EMI consequence follows directly.

Step 2: ΔV Drives di/dt, and di/dt Radiates

The voltage deviation on the rail is not static. The load current ramp — the di/dt — is what drives current into and out of the reference plane as the decoupling network attempts to supply charge locally. When local capacitance is insufficient, the current loop grows: charge must travel farther along the power and ground planes to reach the switching node, and that larger loop area radiates.

The spectral content of a trapezoidal current transient extends to a first-order corner frequency of:

f_corner = 1 / (π · t_r)

Where t_r is the current rise time. For a modern SoC or FPGA with 200 ps I/O edge rates:

f_corner = 1 / (π × 200 × 10⁻¹²) f_corner ≈ 1.59 GHz

Significant spectral energy therefore extends well into the GHz range — directly into CISPR 25 Class 5's most demanding test band (1 GHz – 2.5 GHz, radiated emissions limit nominally 47 dBµV/m at 1 m for broadband). A 100 mV deviation switching at those edge rates couples a current transient across a ground-plane loop that, even at 10 mm × 10 mm area, produces a loop antenna radiating at frequencies where 30–40 dB of shielding effectiveness is hard to achieve with anything short of a fully gasketed enclosure.

The far-field E-field from a small loop antenna is approximated by:

E = (η₀ · π · A · I · f²) / (c² · r)

Where η₀ = 377 Ω (free-space impedance), A = loop area (m²), I = loop current (A), f = frequency (Hz), c = 3 × 10⁸ m/s, r = measurement distance (m). The f² dependence is the critical point: doubling frequency quadruples the radiated field. A current loop problem that is marginal at 500 MHz becomes a hard CISPR 25 failure at 1.5 GHz.

Step 3: Where a Shield Can Actually Helps — and Where It Does Not

A board-level shield can, properly grounded, suppresses radiated emissions from the enclosed circuitry by providing a conductive boundary that reflects and absorbs fields before they escape into the measurement antenna's sight line. Per POCONS' own shielding design guidance, effective containment requires a continuous low-impedance ground ring with via stitching pitched below λ/20 at the highest frequency of concern — at 1.59 GHz, λ/20 = (188 mm)/20 ≈ 9.4 mm, so via pitch must stay under roughly 9 mm to prevent slot-antenna re-radiation at that frequency.

The shield cannot address two of the three paths through which the noise escapes:

1. Conducted path: Current transients on the power rail couple directly onto any connector or harness pin that shares the power or ground net. This is the primary CISPR 25 conducted-emissions failure mode in automotive ECUs, and a shield can does nothing to interrupt it. A common-mode choke and properly placed bulk capacitance at the board edge are the right tools here.

2. Ground bounce: Insufficient local decoupling causes the reference plane itself to oscillate. When the ground plane bounces, every trace referenced to it becomes a radiating element — including traces outside the shield boundary. A can over one section of the board does not prevent ground bounce from coupling noise out through adjacent I/O or power traces.

3. Radiated from enclosed area: This is the one path a shield can interrupt, and it does so reliably when the ground ring and via pitch are correctly designed.

The Real Fix: Restore Capacitance Before Adding Metal

The engineering priority order, when decoupling BOM availability collapses, is:

1. Substitute footprint, not just part number. A 0603 C0G 1 µF cap (readily available, lower allocation pressure than 0402 X5R) can replace a 0402 X7R 1 µF in many layouts with a simple copper pour adjustment. C0G holds its capacitance under DC bias and temperature — X5R and X7R can lose 50–80% of rated capacitance at rated voltage, so a 10 µF X5R at 3.3 V rail may be delivering 4–5 µF effective. The C0G part delivers its full rated value and may provide equivalent or better effective capacitance in a larger body.

2. Stack where footprint allows. Two 0402 caps in parallel on the same courtyard using a 0402-stacked land pattern doubles capacitance without PCB respinning, provided paste stencil and reflow profiles are reviewed with the CM. This is a faster path than waiting 26–40 weeks for allocation relief.

3. Add bulk at the power entry. A 100 µF polymer aluminum or MLCC in an 0805 or 1206 package at the VRM output extends the hold-up window (increases effective Δt budget before the rail deviates), directly reducing ΔV for the same load step and the same local bank.

4. Then add shielding if radiated margin is still insufficient. After restoring the decoupling target, if the product still fails radiated emissions — because the switching regulator itself, not the power-rail transient, is the dominant emitter — a shield can over the converter is appropriate and effective. That is a different root cause and a different fix.

The math is unambiguous: with the same load and regulator, cutting effective decoupling capacitance from 100 µF to 40 µF raises rail ΔV by 2.5×. That deviation drives a larger current loop at frequencies above 1 GHz, where CISPR 25 Class 5 radiated limits are tightest. A shield can over the affected circuitry, with via pitch held under 9 mm, addresses the radiated portion of that problem. It does not address the conducted portion, and it does not substitute for restoring the decoupling network to its designed capacitance value.

Fix the BOM first. Then close the remaining radiated margin with metal.

One Thing

The MLCC squeeze is not a passive-component problem — it's a signal-integrity problem dressed in a procurement deadline. Every 0402 X5R/X7R part ≥10 µF and every automotive C0G/NP0 part currently under allocation is load-bearing decoupling. Pull one from the BOM without a verified substitute and the PDN impedance rises, switching noise couples into adjacent circuits, and your shield can dampens the symptom while the root cause keeps radiating. The next 30 days are the window: audit those exact grades across your active BOMs, place strategic buffer stock before Q3 2026 contract prices lock in, and schedule a PCB shielding review now — so any decoupling topology change you make under allocation pressure gets checked for new radiated-emission exposure before the layout is frozen, not after a compliance failure at the test house.

— Mike Kwak, POCONS USA

Sources

1. AI drives MLCC shortage ...
2. Epic Price Surge of MLCC Is on the Horizon
3. The 2026 Passive Components Crunch: Why MLCC and Capacitor Lead Times — 773 GROUP LLC
4. MLCC Price Hike Logic Shifts: AI Servers Gobble Up 10% of Capacity—Why Is the 5-Trillion-Unit 'Electronic Rice' Still in Short Supply? — BigGo Finance
5. Automotive MLCC Market Size & Share Outlook to 2031
6. MLCC Spot Prices Surge by Up to 20% as Supply Tightens | Lisleapex
7. As Japanese and Korean giants move upmarket, China’s MLCC firms see their chance | South China Morning Post
8. High Voltage MLCC Market Size & Share Outlook to 2031
9. China MLCC Market– Size, Share, Trends, Growth & Forecast 2025–2034 2025-2034 | Size,Share, Growth
10. Electronic Component Shortage 2026: Memory, MLCC & Op Amp
11. AI Servers Create New MLCC Trend: Will It Be the Next HBM?
12. MLCC Shortages Deepen as AI Demand Extends Lead Times - Astute Group
13. MLCC Price Spikes, Component Selection Playbooks - Semicon electronics
14. Taiyo Yuden to raise prices across passive components from May - SemiMedia
15. Murata and Taiyo Yuden shares surge on price hike reports By Investing.com
16. Passive Component Crisis: MLCC Price Surge, Supply Chain Impact | LDeepAI Tech Hub
17. [News] MLCC Giant Murata Reportedly Confirms April 1 Price Hike on Key Components
18. Capacitor Prices Rise Again, Panasonic and KEMET Increases of Up to 65% - Semicon electronics
19. Panasonic raises tantalum capacitor prices as passive component costs surge - SemiMedia
20. [News] Tantalum Capacitor Rally on AI: Panasonic Reportedly Hikes 15–30%, Yageo Rides Wave
21. MLCC Shortages Return as AI Server Demand Strains Capacity - Astute Group
22. Murata Announces Mass Manufacturing of World’s First 0402 47µF MLCC
23. Murata Begins World’s First Mass Production of 47µF MLCC in 0402-inch Size
24. Top Multilayer Ceramic Capacitor Manufacturers | Unibetter
25. Multilayer Ceramic Capacitor (MLCC) Market Report 2031
26. AI MLCC Market Boom: How AI Servers Drive MLCC Demand
27. AI Servers Drive MLCC Pricing Polarization; High-End Components Set for Price Hikes in Second Half — BigGo Finance
28. MLCC Manufacturers Consider Price Increase as AI Demand Outpaces Supply
29. Silver Price Today - Live Silver Spot Price Charts | JM Bullion
30. www.mexc.com
31. Current price of silver as of Thursday, June 11, 2026 | Fortune
32. Silver Price Outlook May 2026: Stop Chasing the Number
33. Silver price forecast 2026: drivers, risks, base/bull/bear case | Skillings
34. Silver Price Predictions for the Next 5 Years: Data-Backed Scenarios
35. Automotive MLCC Market Size, Growth, Trends 2035
36. LME - Copper Spot Price | Series | MacroMicro
37. Copper - Price - Chart - Historical Data - News
38. LME Tin | London Metal Exchange