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

Michael Burry increased leverage on his AI-bubble bearish view by replacing certain short positions with June 2027 put options on Micron and Nebius, implying potential share-price declines of more than 50%. His thesis is that AI-linked valuations underprice the risk of a hyperscaler capital-expenditure slowdown, which would hit memory-chip demand and neocloud compute pricing first. Micron also faces potential memory-price pressure from expanding Chinese production, while highly leveraged Nebius is vulnerable to lower utilization and declining premiums for GPU capacity.

Analysis

The relevant fault line is not an outright AI capex contraction but a deceleration versus embedded expectations. MU is a high-beta expression of that risk because memory pricing and utilization amplify small demand forecast changes into disproportionately large gross-margin and EPS revisions; however, Chinese supply is initially more disruptive to commodity DRAM/NAND than leading-edge HBM, so the bearish case requires either HBM qualification progress or weaker accelerator-server builds. The first confirmation points over the next 1-3 months are hyperscaler capex guidance, HBM contract-price commentary, and MU's bit-growth and gross-margin outlook rather than headline AI sentiment.

NBIS has the more asymmetric downside if GPU-cloud capacity is brought online ahead of contracted demand: lower utilization simultaneously compresses realized compute pricing, raises depreciation burden per sold GPU-hour, and worsens financing optics. That is a balance-sheet/operating-leverage trade, not simply a semiconductor-cycle trade. Conversely, a broad "AI bubble" short is too blunt: NVDA can remain supply constrained even if marginal neocloud economics weaken, while established hyperscalers can redirect infrastructure spend internally; the likely dispersion is short leveraged GPU lessors versus selective long suppliers with contractual demand and scarce components.

The contrarian risk is that consensus may already understand capex circularity, while enterprise inference demand begins absorbing capacity in 2027. A continued HBM supply shortage, firm MU pricing through two quarterly prints, or NBIS disclosing multi-year take-or-pay contracts at attractive economics would falsify the near-term bear thesis. Treat reported fund positioning as a sentiment catalyst rather than independently verifiable evidence of an impending demand break.

Key Decisions for Investors

• Prefer a defined-risk MU bearish structure over an outright short: buy 6-12 month put spreads only after MU guidance implies decelerating HBM revenue or gross-margin expansion. Target a 2:1 payoff profile; exit if management raises full-year HBM supply commitments or pricing expectations on the next earnings call.
• Maintain NBIS as a downside watch/short candidate, contingent on verifying borrow availability, liquidity, net-debt maturities, and contracted-versus-uncontracted GPU capacity. Initiate only after utilization or revenue-per-GPU-hour weakens; cover on evidence of material take-or-pay customer commitments or equity financing that removes refinancing risk.
• Express dispersion rather than index-level AI bearishness: long NVDA versus short NBIS over a 3-6 month horizon, sized beta-neutral. The trade monetizes a divergence between scarce-chip supplier economics and commodity-like rented-GPU economics; stop out if NBIS utilization and pricing improve for two consecutive reporting periods.
• Set event alerts for hyperscaler capex revisions, MU HBM pricing/bit-growth guidance, and China memory-export data. Absent deterioration in those inputs, avoid adding AI shorts solely on prominent-investor positioning.