Michael Burry's 2026 Market Warning: The Bear Case Explained

Michael Burry’s bearish case in the summer of 2026 is not that stocks are expensive. It is that the way money now moves through the market makes a calm, rising tape into a mechanism for building leverage — and that the AI trade at the centre of it is financed in a loop. On 4 August, the day the S&P 500 closed at a record for the first time since June, he wrote that it is “possible we are near a major top, and possible a 1987-type fall.”

That quote has travelled a long way without the argument underneath it. This page is about the argument: what he actually claims, the mechanism he describes, which parts of it can be checked against a primary source, and where he has been wrong so far. For the positions themselves and how they are performing, see our Michael Burry stock picks roundup . Nothing here is investment advice.

What is Michael Burry warning about in 2026?

That the market may be near a major top, and that if it breaks, it could break the way 1987 did — fast, and without a recession to explain it. The full sentence, from a Substack post on 4 August 2026 reported by CNBC, is more careful than the headline version:

“I continue to believe it is possible we are near a major top, and possible a 1987-type fall, but the S&P 500 making new highs likely will bring new money into the market.”

Three things in that sentence are worth separating, because they usually get collapsed into one.

It is conditional, twice. “Possible” appears twice, and the second clause concedes the rally can keep running — that new highs themselves attract money. This is a statement about fragility, not a date.

The 1987 reference is about shape, not size. The 1987 crash is the reference case for a market that fell violently without a recession attached, driven substantially by mechanical selling — portfolio insurance strategies that were required to sell as prices dropped, which pushed prices down further. Reaching for that year rather than 2000 or 2008 tells you his concern is market structure rather than economic fundamentals or credit quality.

The timing was pointed. He wrote it on a day the S&P 500 rose 1.9% to its first record close since June, with the Nasdaq up 2.7% and nearly 5% across two sessions, lifted by earnings beats and falling oil. Saying the top may be in on the day of a record close is either the whole point or the whole problem, depending on your view of him.

Why does he think the rally itself is the risk?

Because a rising market with falling volatility mechanically forces a particular set of funds to borrow more — and nobody at those funds has to decide to do it. This is the part of his case that is genuinely his, and he stated it plainly in the same post:

“Remember, the market going up on falling volatility forces vol-targeting funds to leverage up, and brings leverage from other momentum strategies into play.”

The mechanism, unpacked: a large pool of money runs volatility-targeting strategies — risk parity funds, some commodity trading advisers, parts of the insurance world. These strategies size their exposure to hit a fixed level of portfolio risk. When measured volatility falls, the same dollar of capital supports a larger position, so the strategy adds exposure. It is a rule, not a judgement. Momentum strategies compound it from the other side by adding to whatever has already gone up.

So a calm, rising market generates leverage automatically. The concern is symmetry: the same rules that add exposure into falling volatility are required to cut it when volatility spikes, and they cut into the same names at the same time. That is the 1987 analogy doing its work — not “stocks are overvalued” but “the seller of last resort is an algorithm with a risk limit”.

He has been building this argument in public all summer. His own archive shows the sequence: a July post on the VIX, an instalment on offshore insurers and the hyperscalers, and then on 31 July a long piece titled Foundations: Market Structure, Volatility Targeting, Pod Shops & Other Gremlins, abridged into a shorter version on 3 August. The multi-strategy funds — the “pod shops” — are in that title because they are the other half of the crowding problem: many books running similar momentum exposures with tight risk limits, which forces simultaneous de-grossing when something moves.

⚠️ The full texts of those posts are behind a paywall. Their titles and subtitles are public and are the basis for the description above; the specific quotes in this article come from press reporting of the posts, not from the posts themselves.

What is the circular financing argument, and does the BIS back it?

Yes, and this is the one part of his case that can be checked against a primary document rather than taken on trust.

Burry’s AI argument is not primarily about valuation either. It is that a meaningful share of AI demand is not end demand — that chipmakers and hyperscalers take equity stakes in AI labs and neocloud providers, which then commit to buying chips and compute from them, so capital goes out one door and comes back as revenue. In July he explicitly credited the Bank for International Settlements on this point, titling his 23 July post 3 Stock Buys & the BIS Weighs in on AI’s Rampant Circular Financing.

We read the BIS report. It backs him.

The BIS Annual Economic Report 2026, published on 28 June 2026, states that the opacity of AI sector financing compounds the sector’s vulnerabilities, that hyperscalers, chipmakers and AI labs are linked through a complex web of private arrangements, and that the most prominent of these is circular financing — which it defines as a reciprocal structure in which hyperscalers take equity stakes in AI labs in exchange for those labs’ purchase commitments, rechannelling capital back to investors as revenue. One of its chart panels is headed “Circular AI financing is widespread.”

The report goes further than a definition:

• It shows AI firms’ investment increasingly financed by debt, and their credit risk rising.
• It notes that data centre construction is increasingly outsourced to third parties that lease the facilities back to the hyperscalers.
• It says AI-related financing has become more concentrated and circular within the ecosystem, and that these interlinkages could turn into powerful amplifiers at times of stress.
• It runs a scenario in which the unwind involves partial dissolution of those partnerships and debt fire sales.
• In a separate discussion of AI’s long-run effects it describes a case where the binding constraint turns out to be demand — the demand needed to justify further capacity expansion simply not being there.

None of that is a market call. The BIS is describing a structure and its stress behaviour, not predicting a crash, and its report also credits AI investment with contributing roughly a percentage point to US real GDP growth. But it does mean the factual spine of Burry’s AI thesis — that the financing is circular, concentrated and debt-funded — is the mainstream central-banking view as of June 2026, not a contrarian invention.

Separately, and from earlier: in November 2025 Burry argued that hyperscalers understate depreciation by writing AI chips down over five or six years when their useful life is closer to two or three, and estimated roughly $176 billion of unrecognised capital decay across Microsoft, Amazon and Google over 2026–2028, with Oracle’s and Meta’s earnings overstated by roughly a quarter and a fifth respectively by 2028. That argument is dated to 2025, the per-company figures vary slightly between reports of it, and depreciation schedules are disclosed in company filings and are a judgement companies are entitled to make. It is background to this summer’s case, not part of it.

What has he actually done about it?

Extended the bet rather than pressed it — which is the strongest evidence about what he really believes.

Reporting of his 4 August trading post is consistent across several outlets: he rolled his Nvidia puts out to June 2027, with strikes in the low $100s, and extended a short against the Nasdaq-100 tracker QQQ to February 2027. At the same time he cleaned up: exiting a Microsoft long, closing an Oracle short, and closing out a Palantir position — the outlets do not fully agree on whether what closed there was the stock short or an options leg, and his Palantir short is still described as open.

Per CNBC’s account of the same post, the short book he continues to hold is the iShares Semiconductor ETF (SOXX), Micron, Nvidia, Caterpillar, Palantir, Tesla and Applied Materials. He said all of them were profitable except the Nvidia bet — and added that he would cut his losses if the trades moved decisively against him.

Read those two facts together. Rolling options out by ten months costs money and buys time; it is what someone does when they think they are right about the destination and wrong about the schedule. Saying out loud that he will cut if the market disagrees decisively is the opposite of a conviction-at-all-costs short. Both cut against the “Burry is calling a crash” reading that his quotes usually get compressed into.

Where has he been wrong so far?

On the single position he has talked about most, and on timing generally.

Nvidia is the one that has not worked. It is the name he has argued hardest about in public — the circular-financing case is fundamentally a case about Nvidia’s demand — and it is the trade running against him, which is why the puts needed rolling rather than closing.

The broader pattern is more important than any one name. Burry’s record is of being early rather than wrong, and early is expensive in short positions in a way it is not in long ones. The housing bet that made his reputation was placed well ahead of the payoff and he faced redemptions from his own investors before it worked. A short that is profitable five weeks in has not been proved right; a short that is losing has not been proved wrong. This is why his own framing — a fragility argument with no date, financed out to 2027 — is more informative than the headline.

There is also a category of thing he has simply not said. He has not named a level, a catalyst, or a date. He has not said the AI capex cycle is fake, only that a meaningful share of it is financed in a loop. And the paid nature of his newsletter means the public sees a subset of his book: his July posts are titled by trade count, so more positions exist than have been reported anywhere.

Should you act on any of this?

He answered that himself, and the answer was no. From the same 4 August post:

“Again, shorting is not for everyone. I must short. Most should not.”

We are not financial advisers and this is not a recommendation. But there are a few things worth understanding about the shape of what he is doing before treating it as a signal:

The single most useful takeaway is not directional. It is that his argument identifies something specific and checkable — leverage that accumulates automatically in calm markets, and AI financing that loops between a small number of counterparties. Those are conditions, and they can be watched by anyone, whether or not you think they will resolve the way he does.

Sources

Checked on 5 August 2026.

His depreciation estimates from November 2025 are widely reported but are outside this article’s window and are described as approximations rather than quoted to the decimal.

This is an explainer, not investment advice. Drawpie is not a financial adviser and nothing here is a recommendation to buy, sell, hold or short anything. Short selling carries theoretically unlimited risk and options can expire worthless. Every position described is a second-hand report of a past disclosure and may already have been closed. Do your own research and consider a licensed professional before acting on anything here.