Michael Burry Thinks the AI Bubble Will Burst Sooner Than Later, Including Big Drops in Micron and Nebius by Mid-2027

Michael Burry shifted toward more leveraged bearish positions on the AI trade, buying June 2027 put options on Micron and Nebius that imply potential share-price declines of more than 50%. His thesis is that AI-related capital expenditure, particularly for chips and networking equipment, is priced for an unrealistically low risk of disappointment. A slowdown in hyperscaler spending could compress Micron memory-chip pricing and utilization while pressuring Nebius' compute-capacity premiums and leveraged data-center model.

Analysis

The actionable signal is not a famous investor’s positioning; it is the asymmetric earnings sensitivity embedded in the AI infrastructure stack. MU’s HBM/DRAM profit pool is vulnerable to a modest reduction in hyperscaler server-build assumptions because memory pricing can reverse before physical supply additions are absorbed. A 5-10% cut to 2027 AI-server unit expectations could produce a materially larger downgrade to MU EBITDA and FCF as utilization, mix, and spot/contract pricing all move against it simultaneously.

NBIS is the more fragile expression: leased compute capacity has limited product differentiation, while debt, fixed facilities costs, and rapid GPU depreciation create a high operating-breakeven business. The first 1-3 month catalyst is any evidence of slower GPU-cluster bookings, lower contracted backlog conversion, or rising idle capacity; the 6-18 month risk is refinancing or equity dilution if utilization fails to support the asset base. Short interest, borrow availability, and option implied volatility need checking before expressing the view, since NBIS can be structurally expensive to short.

The contrarian case is that a capex "pause" can shift spend rather than destroy it: sovereign AI, enterprise inference, and second-tier cloud demand could absorb capacity with a lag. That outcome would hurt merchant memory less than feared if HBM supply remains qualification-constrained, while an AI semiconductor selloff could be most acute in stocks already priced for near-perfect utilization. The thesis is falsified by sustained MU gross-margin expansion alongside upward HBM contract-price commentary, or by NBIS reporting improving utilization and contracted revenue without incremental balance-sheet stress.

A second-order watch item is private-credit financing of data-center assets. If utilization expectations weaken, lender spreads and loan-to-value assumptions can reset before equity earnings estimates do; ARES is not a clean short, but widening data-center credit spreads would validate a broader de-rating risk across levered neoclouds and equipment vendors.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: short MU versus long NVDA in equal beta-adjusted dollars. The objective is to isolate memory-cycle downside from broad AI demand; reassess if MU guides HBM bit growth and gross margin above consensus for two consecutive reporting periods.
• For NBIS, use defined-risk put spreads 6-12 months out only after confirming liquid open interest and acceptable implied volatility; avoid naked equity shorts if borrow costs exceed the modeled downside carry. Target a 2:1 minimum payoff profile, with invalidation on utilization/backlog metrics improving while net leverage remains contained.
• Set an earnings-season alert for hyperscaler capex guidance, GPU lead times, and HBM contract pricing. A downward revision to 2027 capex or a sequential decline in memory pricing is the trigger to add risk; absent those data, do not chase a sentiment-driven selloff.
• Monitor data-center private-credit spreads and any NBIS financing announcements over the next 1-3 months. Spread widening or equity-funded capex would strengthen the short thesis; stable financing terms and pre-funded expansion would reduce urgency.