Global semiconductor stocks have shed $3.3 trillion in market value since June 22, and the bleeding continued Friday as the PHLX Semiconductor Index closed out its worst week in recent memory with a decline of more than 9%. Investors bought the dip late in the session, trimming intraday losses across the sector, but the damage to the AI trade’s credibility is accumulating faster than the recoveries can repair it.

The semiconductor selloff is no longer a one-day event. It is a rolling repricing of AI exuberance, and the capital fleeing the most crowded trade in markets has implications for risk appetite, portfolio concentration, and the case for hard assets.

For metals investors, the question is not whether Nvidia had a bad Friday. The question is what happens to the broader market structure when the single narrative holding equity valuations together starts to fracture. A $3.3 trillion drawdown in one sector over less than a month is not a garden-variety rotation. It is a stress test of the thesis that AI spending can justify the prices investors have been paying.

What Happened Friday

As Yahoo Finance reported, semiconductor stocks opened under heavy selling pressure on July 17 before trimming losses as buyers stepped in later in the session. Nvidia dropped 2%. Applied Materials fell more than 5%. Lam Research declined more than 2%. The iShares Semiconductor ETF (SOXX) closed at 521.81, down 1.64% on the day.

A few names managed to claw back above the flatline. Marvell and Qualcomm trimmed their losses enough to finish slightly positive. Micron Technology and Sandisk flipped between positive and negative territory throughout the session. AMD, Broadcom, and Intel all closed lower but off their worst levels of the day.

The partial recovery does not erase the weekly picture. The PHLX Semiconductor Index lost more than 9% for the week, a punishing stretch that reflects something deeper than a single catalyst.

Related Posts
Ultra-rich investors cut stocks and build cash reserves

The Catalysts Stacking Up

Several developments converged to pressure the sector. Chinese AI startup Moonshot unveiled Kimi K3 at the World Artificial Intelligence Conference in Shanghai on Friday. The model is believed to be the world’s largest publicly available AI model that developers can download and run themselves. That kind of competitive pressure from China rattles the assumption that U.S. chipmakers will capture the lion’s share of AI infrastructure spending indefinitely.

The competitive threat from China arrived against a backdrop of rising costs and delayed timelines at home. Bloomberg reported Thursday that Alphabet was behind schedule in delivering Gemini 3.5 Pro, described as its most powerful AI model. When the biggest spenders in the AI ecosystem start missing deadlines, investors start asking harder questions about the return on all that capital expenditure.

TSMC added another layer of concern. The foundry giant guided to higher-than-anticipated capital expenditures during the week, driven in part by higher tool prices. That is good news for equipment makers in theory, but it also signals that the cost of building out AI infrastructure is rising faster than expected. Equipment stocks did not take it well. Applied Materials and Lam Research bore the brunt of Friday’s selling.

Meanwhile, approximately a month before the Kimi K3 launch, the U.S. government withdrew Anthropic’s Fable and Mythos models over security concerns. The specific agency behind the decision was not identified in reporting, but the move highlighted the increasingly tangled relationship between AI development, national security, and the regulatory environment these companies operate in.

The AI Trade Under Pressure

The selloff that began in late June has been global in scope. Newsmax reported that the initial wave of selling originated in South Korea, where foreign investors dumped over $2.5 billion in Kospi shares amid forced liquidations and margin calls. The Philadelphia Semiconductor Index plunged roughly 8% during that early rout, and Micron shares fell 12.2% ahead of its earnings report.

Chris Low of FHN Financial captured the mood at the time: “The risk-off trade reflects fear AI exuberance may be overdone.” Julian Emanuel of Evercore ISI framed the stakes more bluntly: “Earnings will be the proof of the pudding.”

That was late June. Nearly a month later, the pudding has not set. The sector has continued to bleed, shedding $3.3 trillion in market value since June 22 heading into Friday’s session. The dip-buyers keep showing up, but the dips keep getting deeper.

Seema Shah, chief global strategist at Principal Asset Management, told Yahoo Finance earlier in the week that the foundation of the AI investment case rests on continued spending by the hyperscalers:

“What we definitely need to see continue, though, is hyperscalers’ capex. Their earnings need to be strong. They are really the foundation for the entire AI ecosystem.”

That framing is worth sitting with. If the entire AI ecosystem depends on a handful of companies continuing to spend at current or accelerating rates, the trade is concentrated, not diversified. And concentrated trades unwind in concentrated ways.

Why Metals Investors Should Care

Gold does not compete with semiconductors for the same buyers on a normal day. But the AI trade has been the primary engine of equity market returns for an extended period. When that engine misfires, the effects ripple outward. As we explored in our analysis of how the S&P 500 rally runs on a thin AI track, the concentration of market gains in a narrow set of tech and AI names creates fragility that affects the entire index.

A $3.3 trillion drawdown in semiconductor stocks does not happen in isolation. It changes risk appetite. It changes margin calculations. It changes the willingness of leveraged participants to hold positions across asset classes. The forced liquidations that kicked off the June selling in South Korea are a reminder that leverage unwinds do not respect sector boundaries.

For capital-preservation-focused investors, the relevant signal is not whether Nvidia recovers next week. The signal is whether the market’s dominant narrative can absorb repeated blows without cracking. Rising AI costs, competitive pressure from China, delayed model deliveries, and regulatory uncertainty are not temporary headwinds. They are structural questions about whether the investment case matches the valuations.

The pattern is familiar. We have seen it before when chip stocks shed a trillion dollars in a single session and gold caught a bid as risk capital rotated toward safety. The mechanics are simple: when the most crowded trade in equities starts repricing, some portion of that capital looks for assets that do not depend on the next earnings call.

The Concentration Problem

Wall Street remains broadly positive on the AI trade, according to the Yahoo Finance report. The consensus view is that earnings season will provide the catalyst to drive the sector and the overall market higher. That may prove correct. But the consensus view also held firm through the first $3.3 trillion in losses.

The deeper issue for portfolio construction is what happens if the AI narrative does not deliver on its promises in the quarters ahead. The market has priced in enormous future returns from AI infrastructure spending. If those returns arrive more slowly than expected, or if competitive dynamics compress margins, the repricing has further to run.

Readers who followed Michael Burry’s dot-com bubble warning about AI mania will recognize the pattern: a transformative technology can be real and still produce a valuation bubble. The internet changed the world. Most internet stocks still lost investors money.

What to Watch Next

The key variables for the semiconductor sector in the near term are:

  • Hyperscaler earnings and forward capex guidance, which Shah identified as the foundation of the AI ecosystem
  • Competitive developments from Chinese AI firms, particularly whether models like Kimi K3 reduce the pricing power of U.S. chip designers
  • Equipment cost inflation, as flagged by TSMC’s higher-than-expected capex guidance
  • Regulatory actions affecting AI model availability and export controls

For gold and silver investors, the transmission mechanism runs through risk appetite and liquidity conditions. A sustained repricing of the AI trade would pressure the equity indices that have been propped up by semiconductor gains. That kind of broad equity weakness tends to increase demand for assets outside the credit-money system.

It also matters for the Fed’s calculus. Equity market weakness, if it persists, changes the political and institutional pressure on monetary policy. A stock market that was making new highs gave policymakers room to hold rates. A stock market shedding trillions in its most important sector changes that dynamic.

As we noted in our coverage of warning signals in the Nasdaq options market, the derivatives complex often prices stress before the headlines catch up. The options market has been flashing caution for weeks. The spot market is now confirming it.

The Bigger Picture

Dip-buying is a reflex, not a strategy. The fact that semiconductor stocks trimmed their losses on Friday tells us that muscle memory still favors buying weakness in the AI trade. But muscle memory works until it doesn’t. The question is whether the buyers stepping in at current levels are right about the fundamentals, or whether they are catching a falling knife because it is the only trade they know.

Three-point-three trillion dollars in lost market value is a price signal, not noise. Whether it is a buying opportunity or the early innings of a larger repricing depends entirely on whether AI spending translates into AI revenue at the pace the market has already priced in. That answer will come from earnings reports, not from Friday afternoon dip-buying.

When the most crowded trade in a generation starts losing altitude, the old assets that do not need a narrative to hold their value tend to look a little more attractive. That is just how capital works when confidence gets tested, not a prediction.