Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
For three years the loudest voices declaring an AI bubble were the ones nobody had to take seriously: permabears who have been wrong about everything since 2009, and retail sceptics shouting into the void while the index ground higher. That is precisely why the market ignored them. A bubble call only matters when it comes from someone with capital on the line and a record of being early in exactly the right way. This week, that person showed up. Several of them did, on three continents, and the market is finally listening.
When the People Who Take Your Money Start Refusing It
The signal that moved sentiment this week did not come from a financial-news talking head. It came from two of China’s most respected hedge fund managers, and the most important thing they did was not what they said but what they stopped doing. Wealspring Asset, run by Yang Dong, told investors global AI stocks have formed a “super bubble” whose collapse point may not be far away. Yang halted new fund subscriptions back in November. Chen Guangming’s Foresight Fund did the same around the same date. These are managers whose entire business model is gathering and deploying capital, and they have chosen to turn fresh money away rather than buy at current levels.
[Also See: Index Worship is Creating a Bubble]
That detail is the whole story. Anyone can publish a bearish note. Closing the door to new subscriptions is a decision that costs the manager fee income and reputational standing if they are wrong. It is the difference between a forecast and a price. When someone with Yang Dong’s record, he called the tops before China’s 2007 crash, the 2015 unwind, and the 2021 renewable-energy correction, decides the level is too high to enter, that is not a prediction you file away. It is a position taken with real consequences.
The Tell Is the Pattern, Not the Prophecy
There is a comfortable counter-argument doing the rounds, and it deserves a fair hearing because it is the consensus view among Western allocators right now. It runs like this: every pullback in this cycle has been a buying opportunity, AI adoption is real and transformative, and the dip-buyers have been rewarded every single time. Each of those statements is true. None of them tells you anything about what happens next.
The 1990s internet build-out was also real and transformative. The companies that vanished in 2001 were not building fake technology. The mistake was never about whether the internet mattered. It was about the speed at which prices assumed that mattering would convert into earnings. The current divergence sharpens that lesson. Domestic Chinese investors, who have lived through repeated valuation resets since 2020, now lean toward balance sheets and cash yields over long-dated promises. American investors, where the AI story sits at the centre of passive flows, structured products, and corporate messaging, see each drop as a positioning reset. East sees froth. West sees inevitability. Both cannot be right, and the side with the better recent memory of what a real bursting looks like is the one quietly stepping back.
Korea Put a Number on It
If you want to know what the unwind actually feels like rather than what it sounds like in an investor letter, look at Seoul. South Korea’s market is one of the most leveraged proxies for the global AI trade on earth, because Samsung and SK Hynix together represent an enormous share of the index and supply the memory chips that every data-centre build-out depends on. That concentration cuts both ways. When orders flow, the KOSPI flies. When the narrative cracks, it implodes faster than anywhere else.
This week it imploded. The KOSPI sank nearly 10 percent in a single session, tripping a market-wide circuit breaker and a 20-minute trading halt, while Micron plunged more than 13 percent and Nvidia slid as investors began questioning whether infrastructure spending can keep producing earnings to match the price. What makes Korea the canary is the structure underneath it. Retail investors there were handed easy access to two-times and three-times leveraged products, a wonderful arrangement on the way up and a catastrophe on the way down. Leverage does not just amplify losses. It compresses the time available to react, which is how a market goes from record highs to a trading halt in the span of a few sessions.
Follow the Money Out the Door
The most concrete evidence that this is a turn rather than a wobble is in the fund flows, not the headlines. After a roughly 27 percent surge in megacap tech over three months left positioning stretched and valuations brittle, the air came out fast across the global AI complex: the Nasdaq fell more than 2 percent in a session and investors pulled money from US equity funds for the first time since March, with record withdrawals from technology funds specifically. That is the part the bull case cannot easily explain away. Sentiment can stay irrational for a long time, but redemptions are not sentiment. They are capital physically leaving the building.
The contagion is showing up in the places most exposed to the AI windfall narrative. SoftBank, which had effectively become a single-name bet on an OpenAI listing, fell as much as 13 percent in Tokyo after reports the IPO could slip to 2027. The logic there is brutal in its simplicity. SoftBank’s roughly $65 billion OpenAI stake had pushed its market value above Toyota’s on the expectation of a near-term, transparent payday. Take away the certainty of “soon” and you take away the premium. The position is not impaired, but the timing of the liquidity event is suddenly unknowable, and a holding company carrying that much leverage cannot afford an unknowable timeline.
The Real Economy Is Already Sending the Bill
Markets can dismiss a sentiment story for months. They have a harder time arguing with layoffs and price hikes, and both arrived this week from companies that have nothing to do with the AI trade directly but everything to do with where the broader economy actually sits. Volkswagen, Europe’s largest carmaker, is reportedly preparing to cut up to 100,000 jobs and end production at four German plants, the most radical overhaul in its 89-year history, as Chinese rivals take roughly one in ten new-vehicle sales in Europe and US tariffs add billions in annual cost. That is one in six of its global workforce.
On the consumer side, Apple raised hardware prices, sending its own stock down more than 5 percent, and Microsoft pushed through increases too. When the most valuable companies in the world are raising prices into a softening demand environment, and a 657,000-person industrial giant is preparing to shed a sixth of its people, the question stops being whether AI valuations are stretched and becomes whether the underlying economy can carry valuations of any kind at these levels. The super-bubble call is not really about a single sector. It is about the gap between what prices assume and what the world can actually deliver, and this week that gap stopped being theoretical.
What Actually Resolves This
None of this tells you the timing, and anyone claiming otherwise is selling something. Yang Dong himself did not short the market or set a date. He simply stopped buying. That is the honest position, and it is the one worth copying. The thesis from here is not “sell everything.” It is that the burden of proof has flipped. For three years, the AI trade only had to keep existing to keep working, and stories alone carried stocks. That regime is over. From here the market wants math: orders that hold, margins that stay wide, and capital expenditure that converts into returns on a schedule prices can live with.
If those numbers show up in the second half, the leaders reassert and the dip-buyers are vindicated one more time. If they do not, the phrase “super bubble” stops being a provocative headline and becomes the label on a repricing that moves outward from a handful of chipmakers into the entire market narrative that has defined the past three years. The people best placed to judge which way it breaks have already voted with their subscription books. They are not buying. That is the single most informative data point available right now, and it costs nothing to take it as seriously as they do.
