Michael Burry believes the AI bubble 'may burst' sooner than he first believed

Michael Burry accelerated his bearish AI thesis by replacing short positions in Micron, Nebius, SOXX and Palantir with put options, largely targeting June expirations for Micron and Nebius and September 2027 for SOXX and Palantir. He said low implied volatility, including tight VIX measures, made puts a cost-effective way to add leverage and indicated the AI bubble could burst "sooner than later," potentially by next summer. Burry cited risks that AI revenue may fail to justify capital-expenditure commitments, while warning of renewed memory-chip cyclicality as Chinese production capacity expands; Micron and Palantir remain 16% and about 10%, respectively, below their record highs.

Analysis

The relevant transmission channel is not an AI-demand collapse per se, but a change in hyperscaler capex hurdle rates. Semiconductor multiples can de-rate before reported revenue weakens if investors conclude that incremental GPU, memory and data-center spending is being financed against optimistic utilization assumptions. MU is particularly exposed to a memory pricing reversal: incremental Chinese capacity and inventory normalization would compress both HBM scarcity premia and conventional DRAM/NAND profitability, making consensus margin estimates more vulnerable than revenue estimates over the next 2-4 quarters.

PLTR and NBIS represent different forms of duration risk. PLTR's valuation requires sustained commercial growth plus durable operating leverage, so even a modest deceleration in net-new contract value can cause disproportionate multiple compression over 1-3 months; NBIS has greater financing, execution and customer-concentration sensitivity, making it the higher-beta downside expression. Conversely, a broad SOXX short is less clean because analog, auto, industrial and diversified equipment holdings can benefit from an AI-capex rotation rather than suffer from it.

The contrarian point is that tight index volatility makes downside convexity inexpensive only if the eventual catalyst arrives before decay overwhelms the thesis. Public disclosure of one prominent bear's positioning is not itself a fundamental catalyst and could create a reflexive squeeze in the highest-short-interest AI names. The thesis is falsified near term by upward revisions to hyperscaler 2027 capex, sustained HBM contract-price increases, and quarterly evidence that AI workloads are converting into recurring enterprise revenue rather than pilot activity.

ARES is a second-order watch rather than an immediate short: fee earnings can remain resilient, but private-credit marks and deal deployment would be exposed if data-center project financing is repriced. Watch for widening spreads in digital-infrastructure debt, covenant amendments, or reduced sponsor-equity contributions; those would matter more to the credit complex over 6-18 months than an equity-market correction alone.

Key Decisions for Investors

• Do not mirror disclosed put positions without premium, strike and open-interest data; establish an alert for a 10% SOXX decline accompanied by falling semiconductor forward EPS revisions, which would validate a broader de-risking rather than a sentiment-only move.
• Express the bearish valuation view through a 3-6 month PLTR put spread, funded only after a failed rebound into resistance or a commercial-growth guidance miss. Target roughly 2:1 payout-to-premium; exit if management raises full-year revenue guidance while maintaining or expanding operating-margin targets.
• Prefer a relative trade over an outright SOXX short: short MU versus long a diversified semiconductor-equipment basket such as AMAT/LRCX, sized beta-neutral, over 1-3 months. The trade isolates memory-price and China-supply risk; stop out on renewed HBM pricing strength and upward MU gross-margin revisions.
• For NBIS, maintain a downside watch rather than initiate solely on narrative risk. A trade becomes actionable if financing disclosures reveal materially higher funding costs, customer concentration, or delayed capacity monetization; absent that evidence, borrow availability and squeeze risk can dominate fundamentals.
• Monitor ARES and private-credit peers for digital-infrastructure underwriting stress over 6-18 months. Consider reducing exposure only if spreads widen materially and portfolio-company valuation marks or non-accrual commentary deteriorate; equity volatility without credit impairment is insufficient confirmation.