Three Signals, 48 Hours: The AI Trade Cracks
Three signals collided in two trading sessions, cracking the "AI demand is infinite" thesis that drove semiconductor stocks for two years.
The Triple Catalyst — July 1-2, 2026
- Meta lit the fuse. Reports emerged that Meta plans to commercialize excess AI compute through a new cloud business. Investors asked: if hyperscalers have spare compute to sell, did they build too much?
- Burry poured gasoline. Michael Burry disclosed fresh bearish 13F positions against NVDA, AMAT, TSLA, CAT, and SOXX — then publicly called Korea's $100B+ semiconductor expansion "the beginning of the end."
- The selloff was immediate and broad. AI chipmakers, memory stocks, and equipment names cratered — while software and cloud platforms caught bids.
AI Semiconductor Selloff — July 1-2, 2026
Burry's late-2025 put options against PLTR have already paid off — Palantir significantly underperformed from the levels when he disclosed his bearish thesis. This track record makes his semiconductor bearishness harder to dismiss. Importantly, Burry has repeatedly stressed: the problem is not AI itself — it's the valuations investors are paying for AI infrastructure companies.
From "Buy Everything" to "Prove the ROI"
This is not a routine selloff. It is a narrative reset. For two years, the AI trade was simple: buy anything that supplies the AI buildout. Meta just made that formula look incomplete.
Old thesis: AI demand will always outrun supply — buy all AI infrastructure, no questions asked.
New thesis: Hyperscalers may have overbuilt — which infrastructure is actually scarce?
Burry turned a rotation into a referendum. His "beginning of the end" framing isn't about AI dying — it's about AI infrastructure valuations that discount years of perfect execution finally hitting a wall.
Burry is not shorting AI. He is shorting what investors are paying for AI infrastructure.
This shifts the debate from "AI growth" to "AI ROI" — two very different questions. The stocks he targeted (NVDA, AMAT, SOXX) are all capex beneficiaries, not AI monetizers.
A Classic Cycle, Dressed in AI Clothing
Burry's logic is not novel — it follows the classic semiconductor cycle playbook:
What's different this time: AI demand is genuinely real and massive — HBM shipments are growing at triple-digit rates. The question isn't whether AI needs memory. It's whether supply is growing faster than even AI demand can absorb. Historically, semiconductor capacity booms precede price peaks by 2–4 quarters. Korea's expansion, if executed on schedule, would hit just as the hyperscaler capex cycle may be maturing.
Where the Pain — and the Gain — Travels
The Korea mega-expansion, and the market's bearish reaction, ripples through the AI hardware chain in distinct ways.
AI Hardware Supply Chain — Directional Impact by Tier
It starts with equipment makers. Korea's expansion means near-term orders for the likes of AMAT. But if this is the last big capacity wave, those orders dry up after delivery. AMAT's −16.6% two-day drop captures "short-term boost, longer-term cliff" anxiety.
It flows to HBM memory makers. More HBM supply → pricing power erosion. MU sits at the epicenter: one of only three HBM suppliers, deeply levered to AI demand, and directly exposed if Korea's capacity floods the market.
It ends with AI buyers and AI apps. Cheaper compute benefits everyone who uses AI rather than builds it — which is why software and cloud stocks caught bids during the selloff.
MU is one of only three suppliers of high-bandwidth memory (HBM3E), alongside SK Hynix and Samsung. Every NVIDIA GPU deployed for AI training requires HBM. MU's revenue from HBM has grown from near-zero to a multi-billion-dollar business in two years. But with Korea's expansion targeting HBM capacity specifically, a triopoly becomes a race to fill fabs — and historically, memory triopolies have never held pricing discipline when capacity exceeds demand.
Who Gains, Who Bleeds
The rotation has clear directional bets — but separating real business exposure from sympathy moves is the key.
June 30 – July 2 Selloff — Impact by Ticker
| Ticker | Relationship | Why It Moved |
|---|---|---|
| AMAT | Equipment supplier to Samsung/SK Hynix | Short-term order boost, longer-term demand cliff fear |
| MU | HBM supplier to NVDA (HBM3E) | HBM oversupply risk if Korea expansion materializes |
| NVDA | GPU monopoly; Burry directly shorted | Valuation compression + direct bearish target |
| SOXX | Semiconductor ETF; Burry shorted | Broad sector repricing |
Indirect Beneficiaries — The Rotation Trade
- Software and cloud platforms gained as investors rotated toward companies that monetize AI rather than build it. The logic: cheaper compute = better margins for AI-native apps.
Rotation, Not Collapse — But the Tails Are Fat
Three paths diverge. The market is pricing Base — but the next hyperscaler earnings calls will decide.
🟢 Bull (20%) — "This Was a Shakeout"
- AI hyperscaler capex next quarter surprises to the upside. HBM demand accelerates faster than Korea can build. Semis rebound, rotation reverses.
🟡 Base (50%) — "Rotation, Not Collapse"
- AI infrastructure multiples normalize modestly. Software and AI apps lead the next leg. Semis are not broken — they're no longer the only game in town.
🔴 Bear (30%) — "Burry Is Right"
- Capex growth slows materially. HBM supply hits faster than demand. Pricing power erodes — MU and AMAT reprice most sharply.
Next hyperscaler earnings calls — specifically FY27 capex outlook
If the growth rate decelerates, the Bear case gains probability. MSFT, GOOGL, and AMZN guidance is the market's next tripwire.
Bull triggers: any hyperscaler raises FY27 capex guidance; new HBM contract prices hold or rise; GPT-5 requires 2x HBM per GPU.
Bear triggers: any hyperscaler cuts capex; HBM spot prices decline; Korea fab timeline accelerates.
Base triggers: mixed capex signals; HBM prices stable; rotation continues but semis find a floor.