Michael Burry's Nvidia Hedge Worked. His Thesis Hasn't.

Michael Burry went into Nvidia's latest earnings report betting the stock would fall. Then he bought call options just in case it rose.

It rose. The stock surged 8.7% after Nvidia reported $96.2 billion in quarterly revenue — more than double the year-ago quarter. A tech columnist on X pointed out that Burry's hedge "made him a killing", and Burry publicly agreed.

But the Nvidia short — the main bet, the one that represents 3.5% to 4% of his portfolio — is still losing money. It has been Burry's only unprofitable short position for months. The calls were insurance, not a confession. He added to his short the same day and kept putting pressure on the stock after the beat.

The real story here isn't about Burry's options game. It's about what Nvidia just proved, what questions remain unanswered, and what this means for investors who are neither shorting nor hedging — just trying to decide whether the stock is still worth owning.

Nvidia reported $96.2 billion in revenue for the quarter ended July 26, 2026 — up 106% from a year ago and up 18% from the previous quarter. Data center revenue, the engine of the business, reached $89 billion, up 117% year over year. Non-GAAP gross margin held at 75%. Non-GAAP earnings per share were $2.22. The company returned roughly $26 billion to shareholders through buybacks and dividends in the quarter alone.

Then came guidance: $108 billion in revenue for the next quarter, plus or minus 2%. That's another 12% sequential increase on a $96 billion base. Jensen Huang declared that "AI has reached its inflection point," with AI now doing "useful work" that generates "productive and profitable tokens."

None of this is close to the bubble Burry is describing.

What should you make of it? Nvidia is executing at a scale most investors have never seen: $96 billion in one quarter, with $108 billion expected next. The Vera Rubin platform is ramping into full production at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud, and Nebius. Nvidia has formed compute financing platforms with Apollo, BlackRock, Blackstone, Goldman Sachs, and KKR to mobilize over $500 billion in third-party capital. The company is sitting on a $5.5 trillion market cap with a trailing P/E of roughly 28 — a multiple that would look absurd for almost any other company.

But the P/E doesn't tell the whole story. Forward P/E is closer to 61, which means the market is pricing in continued acceleration. That acceleration has to come from somewhere.

Here's where Burry's argument actually matters, even if his bet against Nvidia hasn't worked yet.

His thesis isn't that demand is fake. It's that demand is circular — that chipmakers, cloud providers, and AI companies are financing and guaranteeing business for one another, making the demand look bigger and more durable than it is.

The evidence is harder to dismiss than a simple "AI bubble" claim.

Nvidia announced an investment of up to $100 billion in OpenAI, tied to OpenAI's plan to deploy at least 10 gigawatts of data-center capacity using Nvidia systems. The capital is fungible: funding a buyer relieves them of raising money elsewhere, which increases their capacity to buy the sponsor's hardware. Nvidia also holds an 11.5% stake in CoreWeave and agreed to purchase up to $6.3 billion of CoreWeave's unsold cloud capacity of CoreWeave's unsold cloud capacity — a backstop that functions as a revenue guarantee.

Four customers now account for roughly 61% of Nvidia's revenue, up from 36% a year earlier. Nvidia refers to them as "Customer A," "Customer B," and so on, making it impossible for investors to determine exactly how much of the top-line growth comes from independent demand versus these reciprocal arrangements.

All four major hyperscalers — Microsoft, Google, Amazon, and Meta — are developing their own AI chips to reduce their dependency on Nvidia. Microsoft's Maia 200 is designed specifically to cut third-party reliance. Google has the most mature custom silicon lineage with its TPUs. Amazon is deploying Trainium 2 and Inferentia 3. Meta has an inference accelerator roadmap.

Burry's point, stripped of the options positioning, is structural: the biggest buyers of Nvidia's chips are also Nvidia's biggest competitive threats. They have the capital, the engineering talent, and the incentive to build their way out of a monopoly that costs them billions.

Here's the mechanism that makes this relevant to your investment judgment.

The AI compute market is shifting from training to inference. Training is characterized by large, lumpy capital expenditures and high per-unit prices — Nvidia's sweet spot, where the CUDA ecosystem creates a real moat. Inference scales incrementally with traffic and favors cost optimization. In inference, customers care less about raw performance and more about the cheapest silicon that meets their latency and throughput requirements.

Custom silicon excels at inference for stable model architectures. As model architectures stabilize, the share of total accelerator spend that goes to general-purpose GPUs shrinks. Nvidia retains its moat in training, but the faster the industry moves from training to inference, the more pricing pressure builds on the exact products that carry 75% gross margins.

This is not a prediction that Nvidia will collapse. It's a mechanism that explains why Burry — wrong on timing, potentially right on direction — keeps doubling down.

So what does this mean if you own Nvidia, are watching it, or are considering a position?

The numbers are extraordinary. $96.2 billion in quarterly revenue with 75% margins and $108 billion in next-quarter guidance is not a company that's about to fail. Nvidia generates $127 billion in free cash flow over the trailing twelve months, with $117% return on equity and 87% return on invested capital. The company is a financial machine.

But the forward P/E of roughly 61 means the market is assuming this machine keeps accelerating. That assumption requires several things to hold simultaneously: hyperscaler capex translates into Nvidia orders without meaningful substitution to custom silicon, inference-era per-unit revenue declines are offset by enough volume growth to maintain aggregate revenue, and the hyperscalers' AI investments eventually generate enough return to justify the spending.

If all three hold, the stock has further to run. If even one doesn't, the forward multiple compresses and the stock price adjusts.

Burry's hedge is a lesson in market mechanics, not investment wisdom. He bought insurance against a stock he believes is overvalued, and the insurance paid off because Nvidia beat again. That doesn't mean his thesis is wrong — it means the near-term numbers keep printing and the structural questions keep accumulating.

The debate is not about whether Nvidia stays important. It is about whether the return profile is still as compelling as what can be found elsewhere, given that the stock is already pricing in sustained acceleration from a $5.5 trillion base. The answers to that question will come from customer concentration disclosures, inference-era margin trends, and whether hyperscaler AI spending proves durable once the initial build-out phase ends.

Until those answers arrive, the machine keeps running. But machines that depend on four customers to account for 61% of their revenue, and whose biggest customers are building escape routes, carry a different kind of risk than the headline growth rate suggests.