Burry closes $187 million Nvidia put as AI short gets costlier

Michael Burry closed his December 2026 Nvidia put options without rolling them into later-dated contracts, cutting the most-watched leg of a bearish AI wager that had reached roughly $1 billion.

The Scion Asset Management founder said in a Substack post cited this week that he "pulled in risk" in September and now holds cash he is "happy to sit on." The Nvidia position he exited was worth $187 million in put options, according to the disclosure cited by 24/7 Wall St. He did not replace it.

"This will be an interesting market this fall," Burry wrote, adding that the exit should not be read as optimism about AI valuations.

The stated reason was time decay, not a change of thesis. Short-dated puts lose premium quickly when the underlying refuses to fall, and Nvidia has done the opposite. The stock is up 19.16% year to date and 30.16% over the past 12 months, closing near $221.97 with a market capitalization around $5.4 trillion. The full-chain put/call ratio sits at 0.55, meaning bullish call positioning still dominates options flow in the name.

Burry kept his 2027 puts on Palantir and the Invesco QQQ Trust, which tells you where the remaining bearish exposure sits. The AI infrastructure complex is no longer part of it, at least not through Nvidia.

Nvidia's August quarter made the short painful to hold

The Q2 FY2027 report on Aug. 26 is the kind of quarter that squeezes shorts. Nvidia delivered revenue of $96.22 billion, up 105.8% from a year earlier, with non-GAAP earnings per share of $2.22 against a $2.09 consensus. The Data Center segment grew 117% to $89.02 billion on the Blackwell Ultra ramp, and non-GAAP gross margin expanded to 75.0%. Management guided Q3 revenue to $108 billion, plus or minus 2%.

Supply commitments jumped to $279 billion, largely tied to memory procurement for Vera Rubin production, according to the Q2 press release filed with the SEC. Chief Executive Jensen Huang framed the setup in one line: "AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue."

Betting against that print with a four-month put was an expensive hold. The deeper problem for the bear case is that Nvidia's customer base is not just hyperscalers anymore — it includes the GPU cloud operators that Burry also targeted.

CoreWeave carries the same trade, with worse cash flow

CoreWeave sits directly downstream of Nvidia's roadmap. It was the first to bring up and validate the Nvidia Vera Rubin NVL72, and Nvidia holds an equity stake in the GPU cloud operator. CoreWeave's Q2 FY2026 revenue reached $2.58 billion, up 112.3% year over year, and its backlog swelled to roughly $104 billion, with more than $25 billion in additional net new customer commitments in early Q3 2026.

The bear case is not hidden. CoreWeave posted a GAAP net loss of $1.14 per share, free cash flow of negative $5.74 billion on $6.42 billion of capital expenditure, interest expense of $640 million, and a debt-to-equity ratio of 8.94. Options traders reflect that split view: the full-chain put/call ratio is 1.15, with October and November expirations skewing heavily to puts. Shares trade at $90.48, up 27.48% year to date but down 8.91% over the past 12 months, against a 52-week range of $60.55 to $153.20.

One detail cuts against the depreciation argument Burry made last November, when he contended that hyperscalers were stretching GPU useful-life assumptions. CoreWeave has signed A100 contracts extending through 2029, which challenges the claim that older-generation accelerators lose their value too quickly. If last-generation silicon still commands multi-year contracts, the write-down thesis gets harder to sustain.

What the trim does and does not signal

For investors, the read-through is narrower than the headlines suggest. Burry closing a four-month put is a statement about the cost of carrying a hedge, not a statement about AI capital expenditure. The remaining 2027 Palantir and QQQ puts show the broader thesis is intact — he is still positioned for a market-level drawdown, just not through the single name that has been compounding fastest.

The positioning signal matters because Burry's AI short was the most publicized bearish bet in the complex. Its removal takes a known overhang off Nvidia and CoreWeave, and it may pull other short sellers toward the exit. But the actual price effect depends on the size of the trimmed stake relative to total float, and the next 13F disclosure will show how much bearish exposure genuinely remains.

Two dates decide the fall. Nvidia's Q3 guide of $108 billion sets the bar for the December quarter, and CoreWeave's ability to convert its $104 billion backlog into positive free cash flow is the test of whether the leverage that funds the buildout is sustainable. Huang's framing — compute is revenue — is now the consensus. Burry's remaining puts are a bet that it stops being true.

This article is for informational purposes only and does not constitute investment advice.