Nvidia CEO Jensen Huang Just Put a Nail in Michael Burry's Bear Case

Nvidia's older A100 GPUs, introduced in 2020, remain in commercial use and are contracted by CoreWeave through 2029, challenging Michael Burry's claim that AI-chip useful-life assumptions of roughly six years overstate profits. Rental pricing for newer H100 GPUs reportedly rose 22% month over month, while CEO Jensen Huang characterized Nvidia compute as durable, fungible and revenue-generating. The article argues sustained demand for both legacy Ampere/Hopper and current Blackwell/Vera Rubin products supports Nvidia's earnings durability; shares trade at about 28x earnings.

Analysis

The relevant distinction is economic life versus accounting life: continued operation or a single long-dated lease does not establish that an accelerator retains sufficient cash yield to justify a six-year depreciation schedule. The underwriting variable is realized GPU contribution margin after power, networking, maintenance, downtime and financing costs—not quoted spot rental rates. This makes CRWV materially more exposed than NVDA to any future useful-life reassessment, because a depreciation reset would reduce reported EBITDA/FCF conversion optics and could tighten leverage capacity precisely when capex remains elevated.

For NVDA, durable prior-generation demand is a double-edged signal. It supports the installed-base value of CUDA-compatible compute and lowers buyer concerns around residual values, improving willingness to finance new systems; however, a liquid secondary market can also defer upgrades at the low end and shift demand toward used capacity. The key 1-3 month catalyst is whether H100/H200 utilization and contracted pricing remain firm after additional Blackwell/Rubin supply arrives. Over 6-18 months, the bullish case requires new-generation workloads to deliver enough total-cost-of-ownership advantage that legacy supply complements rather than cannibalizes new silicon demand.

Consensus is too quick to treat a company promotional post and an unverified rental-rate observation as dispositive evidence against the depreciation bear case. A rate spike can reflect short-term capacity scarcity, region-specific power constraints, or contract mix rather than durable asset economics. The more actionable accounting risk is not an immediate NVDA earnings hit, but a potential de-rating of highly levered GPU lessors if auditors, lenders, or equity investors begin underwriting shorter refresh cycles.

Key Decisions for Investors

• Maintain a selective long NVDA position, but do not add solely on this narrative; add over the next 1-3 months only if quarterly data-center guidance, Blackwell/Rubin supply conversion, and disclosed demand for prior-generation capacity remain intact. Falsify on a material guide-down tied to customer upgrade deferrals or evidence that used-GPU availability is pressuring new-system pricing.
• Use a relative-value expression: long NVDA / short AMD in equal beta-adjusted dollar amounts for 3-6 months. Durable legacy CUDA demand raises switching costs and residual-value confidence for NVDA customers, while AMD has less installed-base monetization support. Exit if AMD demonstrates sustained AI accelerator share gains or NVDA signals meaningful pricing concessions.
• Treat CRWV as an accounting-and-financing watch item rather than a direct long from higher rental rates. Monitor GPU useful-life assumptions, debt covenants, contracted-revenue duration versus depreciation schedules, and cash interest coverage at the next filing; a shortened-life charge or weaker financing terms would be a short catalyst.
• Avoid using NVDA's stated earnings multiple as a valuation anchor until consensus EPS, stock-based compensation treatment, and the sustainability of gross margin are independently refreshed. The risk/reward improves only if forward estimates hold while supply broadens without a corresponding decline in accelerator pricing.