Should Investors Worry About Michael Burry's Big AI Stock Shorts

Investor Michael Burry remains short Nvidia, Micron and Palantir, but the article argues his Nvidia thesis is weakened by hyperscaler AI-infrastructure paybacks of roughly 1-3 years and rising rental prices for Nvidia's B200 GPUs. It views Micron as the more credible short because its sub-6x fiscal-2027 forward P/E could prove misleading if conventional-memory pricing, margins and earnings normalize after the HBM-driven supply squeeze. Palantir trades at a steep nearly 57x forward sales multiple, but its approximately 85% gross margin and 62% adjusted operating margin support the view that it is a high-quality AI SaaS business rather than a consultancy.

Analysis

The relevant NVDA debate is no longer whether GPU demand exists, but whether inference utilization converts booked cloud capacity into durable renewals before each hardware generation resets pricing. AMZN and GOOG reduce that risk through captive workloads and custom silicon, creating a second-order headwind for NVDA at the low-end inference layer while preserving its advantage in frontier training and integrated networking. Over the next 1-3 months, hyperscaler capex guidance and evidence of rising depreciation as a percentage of cloud revenue matter more than third-party GPU rental anecdotes; a capex slowdown would hit NVDA's multiple before its reported revenue.

MU has the most asymmetric fundamental risk because HBM economics are intertwined with conventional DRAM/NAND supply discipline. A normalization in commodity-memory pricing would create operating deleverage precisely as consensus extrapolates peak mix and utilization into FY27 estimates; the low headline P/E is therefore not a valuation cushion if earnings are cyclically inflated. Watch monthly contract-price trends, HBM qualification share versus SK Hynix/Samsung, and management's capacity-addition commentary over the next two quarters.

PLTR's valuation leaves little room for a deceleration in commercial net-new bookings, but the more important bear test is cash-conversion quality: sustained receivables growth ahead of revenue and limited remaining-performance-obligation growth would imply deployment-heavy revenue rather than scalable platform adoption. Conversely, model commoditization can strengthen its positioning only if it lifts seat expansion and lowers implementation time; high margins alone do not establish recurring demand durability. This is a 6-18 month execution story, not an attractive outright short solely on valuation.

Key Decisions for Investors

• Maintain a tactical long NVDA / short SOXX hedge through the next hyperscaler earnings cycle; NVDA should outperform if AMZN and GOOG capex and cloud growth remain above guidance. Reduce if either reports cloud-revenue deceleration alongside higher depreciation or explicitly reallocates inference spend toward internal ASICs.
• Initiate a 3-6 month relative-value short MU / long WDC or neutral-weight semiconductor ETF hedge only after two consecutive signs of weakening DRAM contract pricing or lower-than-expected MU gross-margin guidance. Target 15-20% relative downside if commodity-memory margins roll over; cover on evidence of further HBM supply tightness or disciplined industry capex.
• Do not chase PLTR outright. Use a post-earnings watch trigger: consider a 6-12 month short only if commercial bookings/RPO decelerate materially while receivables outgrow revenue; absent that evidence, valuation alone is an insufficient catalyst and short-squeeze risk is high.
• Add AMZN and GOOG selectively on AI monetization evidence rather than capex headlines: accelerating cloud revenue with stable operating margins would validate internal-silicon ROI and support multiple expansion. Falsifier is sustained margin erosion without corresponding cloud-growth acceleration over the next two reporting periods.