원문 보기
In late September 2026, Michael Burry returned to the market spotlight, but this time his actions are more specific and urgent than ever before. The investor, who rose to fame for shorting the U.S. housing market in 2008, is expressing his judgment on the AI frenzy in a vastly different way. He is no longer content with holding short positions in common stocks; instead, he has converted most of his AI-related shorts into put options, focusing on expiration dates within the next year.
The core significance of this shift was summarized clearly by Burry himself: " From a fundamental perspective, I am moving up the timeline." When a contrarian investor known for his patience actively shortens his time window, the market should pause to consider what he sees that makes the cost of waiting seem too high.
From sooner or later to much earlier—a proactive advancement of the timeline. Burry’s warnings about AI are nothing new. In May 2026, he compared the current market to the final months before the 1999-2000 dot-com bubble burst. At that time, the Nasdaq was still climbing, and the market responded to his assessment with polite attention before continuing to hit new highs.
But this September adjustment is different. In his investor letter, Burry explicitly stated that the AI bubble might burst sooner than expected . It is no longer a distant prophecy, but a judgment on a compressed timeline . He uses options to replace shorting —the financial expression of this judgment.
Options grant him the right to sell at a specific price within a specific time window. If the bubble bursts before expiration, the leverage effect will amplify gains. If the time window passes and the market continues to rise, the options will expire worthless, with losses limited to the premium paid.
The brilliance of this move lies in the asymmetry of the risk structure. Shorting common stocks faces theoretically unlimited risk, as stock prices can rise indefinitely, while put options lock the risk within the premium, all while retaining significant downside profit potential. One of Burry’s reasons is that, with volatility indicators like the VIX currently at record lows, put options are relatively cheap.
Low volatility leads to low option pricing, which is the technical condition that allows him to front-load his leverage. Specific position allocations reveal his timeframe: put options on Micron Technology and Nebius are concentrated for June 2027 expiration, while put options on the SOX semiconductor ETF and PER extend to September 2027.
This means Burry expects the turning point in the AI narrative to occur within the next year, no later than the fall of 2027. For an investor known for being " too early," he is clearly trying hard this time to avoid repeating his past mistakes. A 3 trillion dollar invisible burden. Berry’s judgment on AI is built upon a figure he believes the market is systematically ignoring: 3 trillion dollars.
According to his calculations, the five hyperscale cloud providers—Amazon, Alphabet, Meta, Microsoft, and Oracle—have accumulated approximately 3 trillion dollars in purchase commitments, future lease obligations, and other off-balance-sheet items related to AI infrastructure. The composition of this figure includes about 1.2 trillion dollars in future lease commitments and approximately 1.5 trillion dollars in commitments to purchase AI-related hardware equipment.
Under current accounting standards, the latter can remain off the balance sheet until the products are finally delivered. When the music stops, these off-balance-sheet commitments will quickly turn into real liabilities. Berry warns that his logic is not complex. The scale of these commitments is already large enough to have a material impact on corporate earnings and cash flow when they come due.
If AI revenue fails to grow at the expected pace, companies will face a dilemma: either continue to fulfill commitments and endure declining returns, or seek to renegotiate or pay penalties, though existing legal agreements already leave room for such adjustments. This judgment echoes research from ARS Management; Berry specifically cites the analysis from this asset management firm.
The AI investment frenzy relies on the premise of sustained AI capital expenditure, which in turn depends on AI revenue that has yet to be fully validated. The ARS report points out that it would only take one quarter of AI revenue falling short of expectations to support their capital spending for some company boards to reallocate funds to other high-certainty projects.
The keyword here is "one quarter." Both ARS and Berry are suggesting that the sustainability of AI capital expenditure is not as solid as the market assumes. It is built on a series of mutually reinforcing expectations: today's investment generates tomorrow's revenue, and tomorrow's revenue justifies today's investment.
But if this cycle breaks at any link, the chain reaction could be faster and more severe than expected. Berry also places this logic within a longer capital cycle framework. He points out that net capital investment by S&P 500 companies reached 2.07%of U.S. GDP by the end of June 2026. This is the highest level in nearly 40 years, excluding the period following the Nasdaq peak in March 2000 .
He expects even higher records in the coming quarters, potentially surpassing the levels seen after the 2000 tech stock peak. In the logic of the capital cycle, overinvestment creates capacity, capacity suppresses returns, and deteriorating returns eventually force companies to slash assets. Berry predicts the retrenchment could occur in 2028 or 2029, but the market will price in this prospect ahead of time, highlighting the overlooked cyclical cracks.
Beyond the AI hype, Burry has locked onto a more specific battlefield : memory chips. He replaced his short position in Micron Technology with put options, driven by a simple yet shrewd judgment. The supply shortage of memory chips cannot last forever. He cites public remarks from Acer CEO Jason Chen as evidence; when asked if he noticed a supply shortage, Chen replied bluntly: "How could a persistent shortage exist?
" China's production capacity has been consistently increasing; there is absolutely no shortage. Contract prices are currently fluctuating at high levels, with some rising and some falling. The value of this quote lies in the fact that it comes from within the industry, not from Wall Street analysts.
As the CEO of a major global PC manufacturer, Chen has a direct pulse on the supply and demand for memory. His judgment implies that the current high prices of memory are not the result of a structural shortage, but a product of cyclical mismatch. China's continuous expansion of capacity is changing the global supply landscape, and the historical cyclical patterns of the memory market dictate that after every boom, a correction must follow.
Burry’s chain of reasoning is clear: the AI boom has pushed up demand for high-bandwidth memory and DRAM. Profits for memory manufacturers like Micron have soared, but such high margins inevitably attract capacity expansion. And China's production growth will only accelerate this process. When supply catches up to demand, prices will fall, and the super-cycle narrative of memory will give way to a standard cyclical downturn.
It is worth noting that Burry is not denying the existence of AI demand; he is questioning whether current memory profitability is sustainable and whether the market has priced in the inevitable cyclical reversal. Micron's stock price is about 16%lower than its historical high. However, this decline may be far from reflecting the magnitude of a potential reversal.
Regarding the parallels between the "Big Short" and the present , any assessment of Burry's current moves cannot avoid his experience from 2005 to 2007. At that time, he shorted subprime mortgage bonds via credit default swaps, persisting for over two years while the market continued to rise, enduring immense psychological pressure and funding costs.
Ultimately, he waited out the crisis and reaped massive returns. But that experience also became a classic case study of being "too early," a risk he is clearly aware of. Replacing short positions with put options is a direct response to the "too early" problem. Put options limit maximum loss and don't face margin call pressures like shorts, though they come with their own cost: time decay.
If the market does not decline before the options expire, the entire premium will be lost. Berry’s choice of a time window centered on June and September 2027 suggests he believes a turning point is approaching . This judgment likely stems from his observation of several key variables: the validation window for AI revenue is narrowing, and hyperscaler capital expenditure is translating into physical infrastructure that must generate revenue to justify the investment.
If the first-half 2027 earnings reports fail to show accelerated AI revenue growth, market patience could quickly evaporate, with the macro environment exerting additional pressure. As noted for the same period, U.S. Treasury yields are at 20-year highs, with the 10-year breaking 4.2%and the 30-year topping 4.5%, as high rates exert persistent pressure on high-valuation growth stocks.
AI-related stocks remain one of the highest-valued sectors in the market. If interest rates climb further or remain at current levels longer than expected, the valuation support for AI stocks will become increasingly fragile . An inflection point in memory chip supply may arrive sooner than anticipated.
The expansion of production capacity in China is a process already underway. Its impact will become increasingly apparent over the coming quarters. If memory chip prices soften in the first half of 2027 , earnings expectations for companies like Micron will face downward pressure , though the challenges Berry faces are equally real.
The Nasdaq Composite just hit a record high last week, and revenue growth for AI-related companies remains strong. Palantir’s P/E ratio exceeds 100x, yet its revenue growth is equally impressive. Market pricing may already reflect a high degree of optimism, but "expensive" and "about to crash" are two different judgments.
The signaling function of put options and the market’s pricing dilemma: Berry’s choice to use put options instead of shorting is, in itself, a signal worth interpreting. If he believed a bubble burst was imminent, shorting stocks directly would yield more direct returns; while put options offer higher leverage, they require premium payments and face time decay.
Choosing options over shorting shows he wants to amplify gains while controlling risk, suggesting that while his timing is early, there is still uncertainty. The deeper question is whether the options market has already priced in an AI crash? Berry mentions that volatility indicators like the VIX are extremely low and puts are relatively cheap, meaning current market pricing implies low expected volatility.
If cracks appear in AI, volatility will spike, and the value of these put options will inflate accordingly. From this perspective, Berry is betting not only on a direction but also on the mispricing of volatility. On the other side of the market, there is ample justification; AI infrastructure construction is supported by real commercial demand.
Cloud computing, model training, and inference workloads continue to grow. Big tech cash flow remains strong, and even with rising capex, they can still sustain investments, credit ratings, and shareholder returns. The definition of a bubble is inherently ambiguous. Prices far exceeding fundamentals are characteristic of one, but fundamentals themselves are evolving rapidly.
As the scarcest resource, Melberry's aggressive bet forces a question onto the market: how much time is left in the AI window? His answer is one year, at most until the autumn of 2027. This judgment may be correct, or it may be wrong. But its value lies in forcing the market to face a variable masked by optimism: time itself.
AI infrastructure construction is consuming capital at an alarming rate, and this capital needs returns to sustain the cycle. If returns arrive slower than capital is consumed, an adjustment will be inevitable. The essence of Barry’s success in 2008 was his ability to identify an unsustainable structure.
The logic behind mortgage stimulus relied on the assumption that housing prices would continue to rise. When that premise was broken, the entire structure collapsed rapidly. What he sees in AI investment might be a similar dependent structure: capital expenditure relies on revenue growth, revenue growth relies on continued adoption, and increased adoption takes time—longer than the market expects.
His accelerated timeline suggests he believes this structure is more fragile than most people imagine. Put options are his tools; 2027 is the date he’s circled on the calendar. As for whether the market agrees, only time will tell, and time is precisely the scarcest resource that Barry is betting against.
Comments 0