Burry Shorts NVIDIA: The Korea Semiconductor Reckoning

The Trade That Has Everyone Talking Again

On June 30, 2026, Michael Burry did what he does periodically: he posted his short positions on Substack and let the internet argue. His entry on NVIDIA came at $198.09; his short on the iShares Semiconductor ETF (SOXX) at $642.80.^[1][2] Both trades have moved in his favor since. As of July 11, 2026, NVIDIA trades around $194.83^[3], SOXX has fallen 12.57% from his entry, and Applied Materials (AMAT) — another short in the same cluster — is off 21.79%. Bloomberg reported on July 8 that NVIDIA’s market cap has shed roughly $1 trillion over the prior two months.^[4]

The headlines write themselves. The man who called 2008 is betting against the most important chip company on earth. But before drawing conclusions, it’s worth separating what Burry actually argues from the mythology that trails him.

The Full Short Book: Five Names

Burry’s disclosed short positions as of June 30, 2026 span five names that, taken together, paint a coherent macro picture: NVIDIA, SOXX, AMAT, Tesla, and Caterpillar. The common thread isn’t simply “AI is a bubble.” It’s a view that capital cycle excess — what Burry calls Supply-Side Gluttony — has run ahead of sustainable demand across semiconductors, electric vehicles, and heavy machinery alike.

The Cisco Analogy: Compelling, But Not Conclusive

Burry’s central AI bear argument is the Cisco comparison. In 2000, Cisco was the undisputed backbone of the internet buildout — dominant, profitable, and universally beloved by analysts. Its stock still fell roughly 90% from peak to trough because investor expectations had detached from any plausible earnings trajectory, even as the internet itself kept growing. Burry argues NVIDIA occupies the same structural position today.

The analogy has real force. On the valuation side, SOXX’s price-to-sales ratio exceeded 16x as of July 1, 2026^[5] — territory that historically precedes painful mean reversion. The Philadelphia Semiconductor Index (SOX) is trading at a premium to its 200-day moving average that hasn’t been seen since the dot-com era. These aren’t Burry’s numbers; they’re the market’s own data.

The Supply-Side Gluttony argument goes a step further. Even granting that AI demand is real, the thesis holds that data center operators have over-ordered GPUs relative to their near-term utilization needs, creating an inventory overhang that will compress both volumes and pricing when the buying frenzy normalizes. Burry specifically labeled South Korea’s semiconductor spending surge the “Beginning of the End” in his June 30 Substack post — the idea being that maximum capex commitment by suppliers historically signals a cycle top, not sustained growth.

The $725 Billion Counterargument

Burry’s bears have a serious problem on the demand side. The four major hyperscalers — Amazon, Microsoft, Alphabet, and Meta — have collectively committed to approximately $725 billion in AI infrastructure spending in 2026.^[6] This isn’t analyst forecast; it’s disclosed capex guidance from companies with balance sheets large enough to absorb years of elevated spending without distress.

The Cisco comparison also breaks down at the earnings layer. In 2000, Cisco’s multiples were built on extrapolated growth from companies that were burning cash. Today’s AI capex is funded by hyperscalers with operating margins comfortably above 30–40%. When Amazon builds a data center, it pays for it from AWS profits. The speculative froth that doomed Cisco’s valuation came from the ecosystem around it — the CLECs, the fiber overbuilders, the dot-com clients — not from Cisco’s own customers going bankrupt.

None of this means NVIDIA’s current valuation is justified. A business can be extraordinary and still be priced for perfection. The question Burry leaves unanswered is timing. The market can remain richly valued for years while fundamental demand continues to grow — and that gap between “expensive” and “about to crash” is where most short sellers get carried out.

Burry’s Actual Track Record

The Big Short is a remarkable film. It is not a reliable guide to Burry’s ongoing predictive accuracy. An analysis of his public calls from 2017 through 2023 found that approximately 29% of his predictions proved accurate.^[7] That’s not a damning indictment — forecasting markets is hard, and even being right 29% of the time can generate alpha if position sizing is correct. But it does mean treating every Burry disclosure as a high-conviction signal is a category error.

In November 2025, Scion Asset Management deregistered from the SEC.^[8] Burry is no longer running a regulated fund that files 13F reports. His current position disclosures are voluntary Substack posts, not regulatory filings. The information has value, but it lacks the accountability structure that made his pre-2008 thesis trackable and verifiable in real time.

The fair read: Burry is a serious investor with an exceptional single call and a mixed subsequent record. His current short book identifies real valuation stress in semiconductor markets. It does not tell you when — or whether — that stress resolves through a crash rather than a slow grind or a rotation.

What This Actually Means for Korean Investors

This is where the analysis gets specific. Korean investors face a structural exposure to this debate that their American counterparts largely don’t.

SK Hynix’s NVIDIA dependency. According to Counterpoint Research data from Q3 2025, SK Hynix holds 62% of the global HBM market, with Micron at 21% and Samsung at 17%.^[9] HBM — high-bandwidth memory — is the component that sits directly on NVIDIA’s AI GPUs. SK Hynix is not just exposed to the AI cycle; it is the AI cycle’s memory architecture, full stop. If NVIDIA’s GPU shipment trajectory bends down, SK Hynix’s HBM revenue follows with minimal lag.

On July 10, 2026, SK Hynix listed its ADR on Nasdaq under ticker SKHY. The IPO priced at $149 and closed up 17% on its first trading day^[10] — a vote of confidence from US institutional investors who clearly don’t share Burry’s bearish conviction in the near term. The Korea Stock Exchange price stood at ₩2,180,000 that same day. The ADR listing improves SK Hynix’s global visibility, but it also means that when global semiconductor sentiment shifts, US-based sellers now have a liquid vehicle to express that view directly.

KOSPI concentration risk. On June 8, 2026, KOSPI fell 8.29% in a single session, touching 7,484 points. That kind of single-day move reflects how few names drive the index. Samsung Electronics and SK Hynix together represent a disproportionate share of KOSPI’s total market cap. A synchronized correction in AI-linked semiconductors doesn’t just affect those two stocks — it affects Korean equity beta as an asset class.

Margin leverage. As of May 2026, Korean retail investors carried margin leverage of approximately 75% of their equity positions.^[11] In a falling market, that leverage ratio is the mechanism that turns a moderate correction into a cascade. When KOSPI drops 8% in a single day and margin calls begin firing, forced selling amplifies the move regardless of whether any individual investor’s fundamental thesis was right or wrong.

None of this requires Burry to be right about the timing of a semiconductor crash. The structural exposure exists whether or not NVIDIA falls another 20%. The question for a Korean investor holding SK Hynix, Samsung, or KOSPI-tracking ETFs is simpler: How much of my portfolio depends on AI capex continuing at current rates, and what is my plan if it doesn’t?

Two Facts That Coexist

SOXX trading at 16x price-to-sales is a fact. The $725 billion hyperscaler capex commitment is also a fact. Markets don’t resolve contradictions tidily or on schedule. What Burry’s trade does — regardless of whether it ultimately profits — is force a specific question onto the table: at what point does “AI demand is real” stop being sufficient justification for “semiconductor valuations are fair”?

The dot-com analogy that Burry favors is instructive precisely because it didn’t require the internet to be fake. The internet was very real. The companies that built the internet’s infrastructure were often real businesses. The problem was valuation, timing, and the speed at which competitive supply overwhelmed first-mover advantages. If the analogy holds even partially, it’s the valuation gap — not the demand narrative — that matters most to monitor.

For investors in Seoul watching SK Hynix trade near its all-time high on the day of its Nasdaq ADR debut, the relevant discipline isn’t abandoning the semiconductor trade. It’s knowing the specific numbers: 62% HBM market share, 75% retail margin leverage, an index concentration that makes semiconductor sentiment and Korean equity performance nearly synonymous. Those numbers don’t tell you to sell. They tell you exactly how exposed you are if the selling starts.

This article is not investment advice. All investment decisions are made at your own judgment and risk. / 이 글은 투자 자문이 아닙니다. 모든 투자는 자신의 판단과 책임으로.