Is Michael Burry Waving a White Flag? ‘Big Short' Investor Reduces Bearish Bets on NVDA, CRWV

Michael Burry closed his December 2026 NVIDIA puts, part of roughly $1 billion in bearish AI wagers that had included $187 million of NVDA put exposure, citing rapid time decay rather than a bullish reversal. NVIDIA's Q2 FY2027 revenue rose 105.8% year over year to $96.22 billion, EPS beat consensus at $2.22 versus $2.09, and management guided Q3 revenue to $108 billion plus or minus 2%. Burry retains 2027 puts on Palantir and QQQ, while CoreWeave remains a high-growth but highly leveraged AI-infrastructure risk, with $2.58 billion in Q2 revenue, a roughly $104 billion backlog, negative $5.74 billion of free cash flow, and debt-to-equity of 8.94.

Analysis

Burry's position management is not a fundamental signal: disclosed option notionals obscure strike, premium paid, offsetting positions, and timing. The more relevant read is that persistent index/mega-cap strength raises the cost of expressing an AI-cycle short through dated puts; crowded downside hedging can itself reduce near-term downside as dealers buy delta into rallies. Over the next days to 1-3 months, this is modestly supportive of NVDA's multiple, but it does not validate the economics of every levered GPU buyer.

The cleaner dispersion is NVDA versus CRWV. NVDA monetizes demand before its customers prove utilization and carries substantially less financing sensitivity, whereas CRWV must fund capacity years ahead of cash receipts; a delay in customer deployments, lower GPU rental pricing, or higher funding costs can turn backlog growth into a liquidity problem. NVDA's large component commitments also create a 6-18 month risk: if hyperscaler and neocloud capacity additions normalize, memory and system suppliers will feel the adjustment first, while NVDA faces potential inventory/working-capital pressure and gross-margin debate.

Consensus is likely too focused on whether AI demand exists rather than who captures the return on capital. Productive inference workloads can sustain accelerator demand while still compressing cloud-GPU economics as supply broadens and major customers internalize capacity. PLTR is a separate valuation-duration risk rather than a direct GPU-cycle hedge: any broad long-duration equity de-rating would hit it, but its downside will depend more on commercial conversion and margin durability than on semiconductor shipment data.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: long NVDA / short CRWV, sized beta-neutral. The thesis is margin and balance-sheet dispersion rather than an outright AI call; target 15-25% relative outperformance. Stop if CRWV demonstrates sustained positive free cash flow or materially reduces leverage without equity dilution, or if NVDA cuts gross-margin/shipments guidance.
• Do not chase NVDA upside immediately after a sentiment-driven rally. Add on post-results volatility only if forward revenue guidance remains intact and gross-margin commentary does not imply a sharper-than-expected Rubin transition dilution; use a 8-10% downside risk limit on a cash position.
• For portfolios needing AI downside protection, prefer a 3-6 month QQQ put spread over single-name NVDA puts. It captures the broader duration/liquidity risk implied by expensive AI valuations while avoiding idiosyncratic squeeze risk; reassess if real rates fall materially or QQQ earnings revisions continue broadening upward.
• Set a CRWV watch trigger around quarterly capex-to-revenue, interest expense, and backlog conversion rather than headline backlog. A second consecutive quarter of deeply negative free cash flow alongside weakening net-new commitments would support increasing the short; improving utilization and financing terms would invalidate it.