The AI Trade Just Hit a Wall. The AI Buildout Hasn't.

The numbers today are ugly across every time zone.

South Korea's KOSPI plunged 9-10%, triggering circuit breakers for the second time in a week. The index has now fallen roughly 45% from its June highs. SK Hynix dropped 13%. Samsung fell 10%. Japan's Nikkei slid nearly 4%. Taiwan's benchmark dropped 4%. A Bloomberg gauge of Asian semiconductor shares fell 7.5% — its worst session since April 2025.

In the U.S., the Nasdaq 100 officially entered correction territory, falling 10% from its record high. The Philadelphia Semiconductor Index retreated for a fourth straight session in its longest losing streak of the year. Nvidia fell 4.3%. ASML dropped over 7%. Dell cratered 13%. Micron and AMD were both down 5%+.

The word being used across every trading desk: capitulation.

But is it?

What's Actually Driving the Selloff

Three catalysts converged this week, and together they've forced a repricing of the entire AI semiconductor complex.

First, China's chipmaking progress is no longer theoretical. A Shanghai-based company has begun mass-producing immersion deep ultraviolet lithography machines — the critical equipment that ASML has held a near-monopoly on for years. Meanwhile, CXMT debuted on Shanghai's STAR Market at an $85 billion valuation and immediately became China's most valuable listed company, with an implied valuation approaching $490 billion. The narrative that Western companies control the semiconductor supply chain is being directly challenged.

Second, AI debt levels are becoming impossible to ignore. Nvidia's $750 billion in AI infrastructure deals — including the $250 billion financing guarantee for OpenAI's Ohio data center — raised questions about debt sustainability across the ecosystem. When the company enabling the AI buildout is also financing it, investors start asking who holds the risk if returns don't materialize on schedule.

Third, the math between AI spending and AI revenue is diverging. Capex is scaling roughly 46-50% faster than revenue across the hyperscaler complex. EV/EBITDA multiples for U.S. tech and AI equities are near 25x — close to historical extremes and uncomfortably near the telecom valuations that preceded the 2000 dot-com peak. The market isn't questioning whether AI is real. It's questioning whether the current prices already assume profits that remain years away.

The Correction in Context

Here's the part that requires perspective.

The Nasdaq 100 is down 10% from its record high. The KOSPI is down 45% from June. These are real drawdowns. People are losing real money. But secular technology bull markets have never advanced in a straight line.

The dot-com era produced multiple 20-30% corrections before the actual top. The 2020-2021 pandemic tech rally included a 12% Nasdaq drawdown in September 2020 that felt like the end of the world. It wasn't. The AI trade itself already survived a sharp correction in early 2025 that temporarily wiped trillions off semiconductor valuations before resuming its uptrend.

What's happening today is a repricing of the pace of returns, not a repudiation of the underlying thesis. The market got ahead of the revenue curve, and it's correcting back toward it. That's normal. It's also painful.

$725 Billion Doesn't Get Refunded

Here's the bull case in one number: $725 billion.

That's the combined 2026 capital expenditure guidance from the four largest U.S. hyperscalers. Amazon at $200 billion. Google at $175-185 billion (recently raised to a potential $205 billion). Meta at $125-145 billion. Microsoft at $120 billion. Up 77% from approximately $410 billion in 2025. JPMorgan estimates total AI-related corporate spending approaching $870 billion by year-end 2026.

This money is being spent right now. The data centers are under construction. The GPU orders are placed. The power purchase agreements are signed. Nvidia's supply is sold out through calendar year 2026 and into 2027. This spending doesn't reverse because the Nasdaq had a bad week.

Every hyperscaler reports that their markets are supply-constrained, not demand-constrained. Azure is turning away AI revenue it cannot serve. Google raised its capex guidance in consecutive quarters because demand kept exceeding forecasts. Meta is exploring a cloud business specifically to sell excess AI compute to external customers.

Analysts at Evercore and Bank of America project hyperscaler capex exceeding $1 trillion in 2027. Goldman Sachs models an upside scenario of $1.4 trillion. Morgan Stanley estimates the tech sector may need to issue $1.5 trillion in new debt over the next several years to finance continued construction.

The scale of committed capital makes this fundamentally different from a speculative bubble. The dot-com era involved companies spending millions on sock puppet Super Bowl ads. This era involves the five largest companies in the world pouring hundreds of billions into physical infrastructure that takes years to build and decades to depreciate. You don't spend $725 billion on a thesis you're about to abandon.

The Real Risk Nobody's Pricing

The actual risk isn't that AI is overhyped. It's that the capex-to-revenue gap keeps widening.

Investment is growing 46-50% faster than revenue across the sector. That doesn't mean revenue isn't growing — it is, significantly. OpenAI ended 2025 at roughly $20 billion in annual recurring revenue, a threefold increase from the prior year. Google Cloud AI revenue is accelerating. Amazon's AI-related AWS revenue is scaling. The money is coming in.

But it's not coming in fast enough to match the pace of spending. The payback period is being pushed further out with every quarter. Hyperscalers that used to fund capex entirely from internal cash flow are now raising external capital at scale. Alphabet priced an $84.75 billion equity raise in June 2026. Oracle plans $40 billion in combined debt and equity for fiscal year 2027. Incremental annual debt across the big five rose from 9% of capex in 2024 to 32% by mid-2026.

This is the transition that matters: from internally funded to externally financed. As long as the capex was coming out of free cash flow, the risk sat with the hyperscalers. Now it's being distributed across credit markets, equity investors, and sovereign partners. That doesn't kill the thesis — but it changes who bears the consequences if monetization takes longer than expected.

The China Variable

China's semiconductor progress deserves its own section because the market is struggling to price it correctly.

The DUV lithography development is significant but measured. BofA estimates a Chinese shift toward domestic DUV tools would reduce ASML's sales by roughly $1.4 billion, or about 2.4%. ASML's real moat — extreme ultraviolet lithography — remains unchallenged. No Chinese company has demonstrated EUV capability, and the technology gap there is measured in years, not months.

CXMT's IPO and its rapid DRAM capacity expansion are a more immediate concern, particularly for commodity memory pricing. But as covered in our SK Hynix analysis, CXMT does not currently produce HBM — the high-bandwidth memory driving the highest margins and strongest demand in the semiconductor industry. The competitive threat is real in commodity segments and largely irrelevant to the AI memory supply chain in the near term.

The market is treating "China progress" as a binary negative for all semiconductor stocks. The reality is more granular: commodity DRAM faces new competition, while advanced HBM, EUV lithography, and leading-edge logic remain firmly in Western and allied hands.

What History Says About Corrections Like This

Every major infrastructure buildout in history has produced violent corrections in the stocks of the companies building it.

Railroad stocks crashed repeatedly in the 1870s and 1880s. The infrastructure they built connected a continent. Telecom stocks collapsed 80%+ in 2000-2002. The fiber optic cables they laid still carry the internet today. Shale oil stocks were decimated in 2014-2016. The wells they drilled made America the world's largest oil producer.

The pattern is always the same: massive capital deployment, speculative excess, a violent correction that wipes out the weak hands, and then the infrastructure keeps operating and generating returns for decades. The correction punishes the investors who bought at the top. It doesn't erase the infrastructure they financed.

$725 billion in data centers and GPU clusters doesn't disappear because the stock market corrects. It processes AI workloads for the next 15-20 years regardless of what happens to Nasdaq this month.

The On-Chain Angle

For on-chain traders tracking semiconductor and AI stocks on Stockgecko, today's selloff is exactly the kind of event where 24/7 markets create structural advantages.

The selling started overnight in Asia — KOSPI triggering circuit breakers, SK Hynix collapsing 13%, Samsung down 10%. By the time U.S. pre-market opened, the move was already in progress. On-chain traders on Hyperliquid were pricing the selloff in real time through HIP-3 equity perps while traditional U.S. investors were asleep.

The reverse will play out tomorrow morning in Asia. Whatever happens in the U.S. session today — whether the selling accelerates or finds a floor — Asian markets will react. On-chain traders can position ahead of that reaction.

SK Hynix reports earnings Wednesday. If the report matches the trajectory Micron just set (revenue up 345% YoY, earnings doubling consensus), it could mark the bottom of this correction. If it disappoints, the selloff has further to run.

The Bottom Line

The AI trade is correcting. Nasdaq is officially in correction territory. Semiconductor stocks are in freefall across three continents. Circuit breakers are being triggered. The fear is real.

But $725 billion in committed hyperscaler capex doesn't get refunded. $1 trillion in projected 2027 spending doesn't get cancelled. The data centers under construction don't stop being built because KOSPI had a bad week.

What's dying is the assumption that every AI stock goes up in a straight line forever. What's not dying is the largest infrastructure buildout in the history of capitalism. The correction punishes the timing. It doesn't change the trajectory.

Prices and data referenced as of July 28, 2026. This is not financial advice.