Goldman Sachs has identified multilayer ceramic capacitors as the next supply-chain squeeze in the AI buildout, and a handful of Asian component makers have already tripled off their spring lows before pulling back sharply.

The AI infrastructure trade is migrating down the bill of materials, from GPUs to memory chips to passive components most investors have never heard of. For metals and hard-asset readers, the pattern matters: each new bottleneck leg concentrates capital in a narrower set of names, stretches valuations further from fundamentals, and raises the stakes for the eventual unwind.

A multilayer ceramic capacitor, or MLCC, is a tiny component that smooths voltage, filters electrical noise, and prevents chip glitches when GPUs and custom silicon ramp power demand up and down during AI processing. A single advanced AI server rack can require hundreds of thousands of them. That density is the core of the investment thesis: if AI server deployments keep scaling, MLCC demand could outstrip the capacity of the handful of companies that make them, creating the kind of pricing power that turned memory-chip maker Micron into an 1,800% winner off its April 2025 lows.

Goldman’s Thesis and the Trade That Followed

In late May, Goldman Sachs published research placing MLCCs behind only GPUs and memory among the biggest cost items in AI server builds, as Yahoo Finance detailed in its coverage of the trade. The bank estimated the AI-server MLCC market could grow more than fourfold from fiscal 2025 to fiscal 2030. That kind of growth projection, attached to a component most generalist investors had never considered, was enough to set off a sharp rally in a cluster of Asian-listed names.

Related Posts
Russia’s gold shipments through Hong Kong surge to all-time high amid shifting trade routes

Taiyo Yuden, Walsin, Samsung Electro-Mechanics, Yageo, Murata, and TDK all surged off March 30 lows. Several of those stocks more than tripled before the latest pullback. Then, in early July, Goldman issued fresh upgrades across the MLCC and capacitor supply chain, naming Yageo, Samsung Electro-Mechanics, and LG Innotek specifically. The bank pointed to AI server demand, rising component content per server, and tighter supply as drivers of higher earnings estimates.

The pullback that followed tells its own story. Several of those MLCC stocks now sit more than 20% below their 52-week highs. A few are off more than 30%. That kind of volatility in a trade built around a component smaller than a grain of rice is worth understanding, not because the capacitor market is a metals story in itself, but because the pattern it reveals about AI capital allocation has direct implications for risk appetite, sector rotation, and the durability of the equity rally that has been suppressing safe-haven demand.

The Bottleneck Playbook

Wall Street has been running a version of this trade for years now. First it was GPUs. Then memory. Now passive components. Each time, the thesis follows the same arc: identify the next constrained input in the AI supply chain, model out the demand growth, find the companies with the most concentrated exposure, and ride the re-rating.

Micron is the poster child. Its nearly 1,800% surge off the April 2025 lows showed what happens when a bottleneck narrative meets a real demand cycle. The MLCC trade is an attempt to replicate that outcome further down the bill of materials, in a market dominated by a small number of Asian manufacturers.

The logic is simple enough. But the further you move down the component stack, the thinner the float, the less liquid the stocks, and the more violent the swings in both directions. That is precisely what the March-to-July round trip in MLCC names demonstrated. A tripling followed by a 20-to-30% drawdown is not the behavior of a stable, well-understood market. It is the behavior of a momentum trade layered on top of a supply-chain thesis that has not yet been confirmed by actual pricing power in the physical component market.

The question Goldman’s research does not fully resolve is whether MLCC pricing power has already materialized or remains an anticipation. The thesis depends on demand exceeding supply capacity. If capacity additions keep pace, the squeeze never arrives, and the stocks that tripled on the expectation give back most of those gains. That distinction matters enormously for anyone sizing the trade.

What the AI Trade Rotation Means for Metals

For readers of this publication, the MLCC story is less about which Asian component maker to own and more about what the pattern reveals. The AI equity trade has been running on an increasingly narrow track, concentrating capital in fewer names and pushing valuations into territory where any disruption to the growth narrative creates outsized downside risk.

Each new bottleneck leg extends the trade’s reach but also its fragility. When the market is chasing the third or fourth derivative of AI infrastructure demand, you are a long way from the original thesis about transformative technology. You are in component-level supply-chain speculation, and the risk profile looks very different.

That fragility has implications for gold and hard assets. The AI trade has been one of the primary engines of equity-market confidence and risk appetite. As long as investors believe the next bottleneck play will deliver Micron-like returns, capital flows toward equities and away from defensive positioning. But when the rotation within AI itself starts signaling exhaustion, the capital that flowed in looking for 1,800% gains does not exit in an orderly fashion.

The Leverage Behind the Buildout

The infrastructure spending powering AI server deployments is not coming from retained earnings alone. Tech companies have been borrowing heavily to fund the buildout, layering debt onto balance sheets in a bet that AI revenue will eventually justify the capital expenditure. That credit dimension adds another layer of risk. If the revenue takes longer to materialize than the debt schedule allows, the unwind ripples through bond markets, credit spreads, and eventually into the kind of risk-off positioning that benefits gold.

The MLCC trade sits at the far end of that chain. It is a bet on a bet on a bet: that AI demand will keep scaling, that server deployments will accelerate, that MLCC supply will lag, and that pricing power will flow to a handful of Asian manufacturers. Each link in that chain has to hold for the trade to work. When one breaks, the correction is fast and steep, as the 20-to-30% drawdowns from 52-week highs already suggest.

Reading the Signal, Not the Component

Goldman’s fourfold growth estimate for the AI-server MLCC market between fiscal 2025 and fiscal 2030 is a projection, not a fact. It may prove conservative or wildly optimistic. What it tells us with more certainty is where Wall Street’s attention is moving and how far down the supply chain the search for the next trade has gone.

When the market’s best-resourced analysts are writing upgrade notes on capacitor makers, the AI trade has entered a late-cycle dynamic. That does not mean it ends tomorrow. Late-cycle trades can run for months or even years. But the risk-reward shifts. The easy money has already been made in the names that tripled off the March lows. What remains is a more contested trade with thinner margins for error.

For metals investors, the practical question is simple: how much of the current equity-market structure depends on the AI growth narrative continuing to deliver new legs? The answer, judging by the concentration of returns and the increasingly exotic nature of the bottleneck plays, is quite a lot. When the market is pricing capacitors like growth equities, the gap between narrative and fundamentals is wide enough to matter. We saw a version of this dynamic when chip stocks shed a trillion dollars in a single session and gold caught a bid.

The MLCC trade may work for nimble traders who got in near the March lows. For the rest of the market, it is a signal worth reading carefully. The further the AI bottleneck thesis extends, the more fragile the structure becomes, and the more valuable a position in assets that do not depend on the next upgrade cycle turns out to be.

When Wall Street’s hottest trade fits on the head of a pin, the case for owning something that has held its value for five thousand years gets a little easier to make.