Over Thanksgiving week, a lone investor turned a family‑table argument about holiday leftovers into market drama. Michael Burry, famed for shorting the housing bubble, launched a Substack and took bearish positions worth roughly $1 billion against Nvidia and other AI names, arguing the AI-hardware story rests on overstated economics and accounting quirks.

This matters because Nvidia sits at the center of modern AI: its GPUs power the data centers that train and serve large language models and generative systems. If one well-known contrarian convinces enough traders that Nvidia’s economics are fragile, that skepticism can cascade into falling valuations, higher funding costs for AI startups, and a sudden re‑rating of the sector. The spat-played out on Substack and in an investor memo response from Nvidia-arrives as the industry reports explosive growth and plans multibillion‑dollar infrastructure builds.

Counting the claims and the counters

Burry’s move into the public square is detailed and specific. According to reporting, his campaign includes bearish put positions exceeding $1 billion across Nvidia and peers, and a newsletter called “Cassandra Unchained” that reached tens of thousands of subscribers in days. His headline accusation: Nvidia’s shareholder dilution and stock‑based compensation have leached tens of billions from owners-a figure he put at roughly $112.5 billion in one public filing.

Nvidia pushed back in a seven‑page memo to analysts, saying Burry’s arithmetic is faulty, that his cost figures “incorrectly included RSU taxes”-which reduces the buyback‑adjusted number to about $91 billion-and that employee compensation is “consistent with peers.” The company also emphasized surging data‑center demand. Both sides use exact numbers; the dispute now hinges on accounting conventions rather than vague rhetoric.

How GPU economics translate into corporate accounting

The differences matter because they change the headline math investors use to price the company. Stock‑based compensation is recorded in several ways on corporate financials; including tax impacts or not can swing headline costs by tens of billions. For active investors and algorithmic models that scan filings, that swing alters buyback‑adjusted EPS calculations and valuation multiples overnight.

At the center of Burry’s critique is depreciation: how long does a data‑center GPU remain economically useful? He argues customers are overstating useful life to justify ballooning capital expenditures. If GPUs lose useful life faster than firms claim, straight‑line depreciation understates annual costs and overstates current profits.

Market mechanics: can a single contrarian spark a cascade?

From an engineering and ML‑deployment standpoint, GPU useful life is nuanced. Trainers usually prefer newer microarchitectures-H100 over A100-because they can halve training time for some workloads. But many production inference loads tolerate older cards. The result is a mixed secondary market and a graduated depreciation curve rather than a uniform five‑year write‑off.

That distinction explains why hyperscalers and startups diverge in accounting and procurement. Cloud providers often expense chips over longer lives and amortize facility costs differently from an AI startup that buys racks outright. Those differences are lawful but create comparison problems for analysts trying to value demand for Nvidia silicon across the entire ecosystem.

What’s at stake for fairness, access, and the AI ladder

History gives precedent for idea‑driven market shifts. TechCrunch’s coverage cites cases such as Jim Chanos and David Einhorn, whose public critiques helped accelerate doubts about susceptible companies. Burry’s argument is similarly granular and public; he deregistered his fund earlier this month, freeing him to publish and agitate without prior compliance constraints, and his newsletter is paywalled at $400 a year-an odd modern megaphone for a lone analyst.

Two mechanisms can amplify his case. First, algorithmic funds and quant models monitor changes in sentiment and key metrics; a sudden spike in bearish options positions against a single name can trigger risk rebalancing across funds. Second, retail and private institutional audiences that follow Burry treat his calls as a catalyst, producing position squeezes and feedback loops. If selling begets selling, a narrative can become reality-regardless of the underlying engineering facts.

But there are limits. Nvidia’s business is diversified across gaming, automotive, and data center, and its data‑center revenue has grown in recent quarters. Market concentration in one supplier creates systemic reliance, but it also creates a cohort of customers-cloud providers, telecom carriers, and enterprise AI labs-who have incentives to continue buying if workloads demand more compute. Dislodging that demand requires either a real decline in technical need or an equally real liquidity shock.

What’s at stake for fairness, access, and the AI ladder

Sources

• This Thanksgiving's real drama may be Michael Burry versus Nvidia - TechCrunch, 2025-11-27

• Vercel Security Checkpoint - VentureBeat, 2025-11-28

• Moving toward LessOps with VMware-to-cloud migrations - MIT Technology Review, 2025-11-27