Michael Burry is still betting against the AI boom. He is just doing it with a longer fuse now.
The investor famous for calling the housing crash has covered his outright short on the iShares Semiconductor ETF (NASDAQ: SOXX) and swapped it for September 2027 puts struck in the low $400s. That is a meaningful shift in posture. Instead of a near-term wager that chip stocks fall, Burry is now paying for time, giving his thesis more than a year to play out.
He has made similar moves across a cluster of AI-adjacent names: Nvidia Corp (NVDA), Micron Technology (MU), Palantir Technologies (PLTR), Nebius Group (NBIS) and Oracle Corp (ORCL). In a Substack post, Burry said he has grown more confident the AI bubble could burst "sooner than later."
SOXX's Q3 Pullback Puts the Debate In Focus
The timing is notable. SOXX fell 11% in the third quarter, its weakest quarterly performance since March 2025.
Yet the semiconductor story remains closely tied to real AI demand. SOXX, which holds 30 companies, has about $48.4 billion in assets and traded at a trailing P/E of nearly 68. Semiconductors account for 82.3% of the portfolio, with another 17.6% in semiconductor equipment.
That exposure makes SOXX a useful gauge of the broader economics of the AI buildout. Burry's argument is that increasingly powerful models could eventually require less computing, challenging the assumption that more AI adoption automatically means exponentially more chips.
The Capex Machine Is Still Running
For now, spending remains enormous. S&P Global estimates that Alphabet Inc (GOOGL), Amazon.com, Inc (AMZN), Meta Platforms, Inc (META), Microsoft Corp (MSFT) and Oracle will collectively spend about $750 billion on capital expenditure in 2026, equal to 38% of their revenue.
But the latest data offers a counterpoint to Burry's thesis. Micron said Wednesday that customers have increased commitments under long-term supply agreements to $32 billion, while the company has secured most of its 2027 HBM output. Data-center SSD revenue also nearly topped $10 billion in fiscal Q4, more than 10 times the year-ago level. Micron said larger AI models, longer context windows and higher user concurrency are driving greater memory and storage requirements.
For now, the data is pulling in two directions. Burry is betting that more efficient AI could eventually undermine the massive hardware buildout, while Micron's latest results show customers locking in billions of dollars of memory capacity as AI workloads continue to expand. That puts SOXX at the center of a bigger debate: whether AI efficiency will ultimately reduce chip demand or simply make room for the next wave of AI applications.
