AI Bull Market Trap? The Big Short Strikes Again
The recent concentrated release of both bullish and bearish signals in the AI track has greatly intensified the market's sentiment of concern.
Goldman Sachs recently released its 2027 Global Market Outlook report, pointing out that the accelerating wave of AI capital expenditure will most likely peak in early 2027, the market trend will shift from unilateral rise to two-way volatility, and it is recommended to increase diversified allocation and hedging;
The Federal Reserve also explicitly named artificial intelligence in its latest Financial Stability Report, warning that it poses potential threats to asset valuation and financial stability.
The well-known short seller Michael Burry chose to directly express his stance with positions, doubling his put options on the Nasdaq 100, and re-establishing short positions on semiconductor ETFs at the same time. He explicitly stated that this is a "bull market trap", especially compared to the top of the dot-com bubble in 2000, and Jensen Huang's resignation has further deepened his concerns about the entire AI industrial chain.
01
Specifically, several reports basically point to several hard constraints faced by the current AI narrative: capital expenditure, high interest rates, Democratic regulatory constraints, and trust concerns in the industrial chain.
These factors are actually the sources of recent market concerns.
The Goldman Sachs report states that the proportion of incremental AI investment in GDP will peak in early 2027, but it does not mean that capital expenditure will turn downward, only that the growth rate of investment will slow down.
Stock market pricing originally pays more attention to marginal changes. The past bull market in the upstream computing power sector relied on the continuously rising growth rate of capital expenditure. Once this momentum stops, the core support for high valuations will loosen.
According to Goldman Sachs' calculations, the pull curve of cloud vendors' capital expenditure on S&P 500 EPS has an obvious corresponding relationship: it contributed 2 percentage points in 2024, 5 percentage points in 2025, peaked at 12 percentage points in 2026, fell back to 9 percentage points in 2027, and directly became a drag on earnings by 5 percentage points by 2028.
2026 is basically the last window period for capital expenditure to drive earnings.
The step-down in earnings growth also reflects this change simultaneously. The relevant earnings expectation of the S&P has dropped from 36% to 11%, and the European, Japanese and Asian markets have weakened synchronously.
Objectively speaking, a growth rate of 11% is not bad in history. What makes people uncomfortable is the decline process from 36% to 11%.
In the past, we enjoyed the Davis double-click where both earnings and valuations rose together. In the future, there will most likely only be slow earnings support and continuous pressure on valuations, and it will be difficult to replicate the market trend of easy profits in the previous two years.
The second unavoidable practical constraint is the high interest rate market environment.
The suppression of the high interest rate environment on the AI track is two-way.
AI is a typical long-duration asset, most of whose returns are in the long term. The rise of long-term interest rates directly increases the discount rate, and the valuation is naturally suppressed.
Cloud giants such as Microsoft and Google have strong foundations and abundant cash flow, so the impact on them is limited. The most affected are the group of intermediate GPU cloud service providers, such as CoreWeave and Nebius named by Burry. They are big customers of NVIDIA, expanded rapidly through large-scale leasing and procurement of GPUs, and are highly dependent on external financing.
Once interest rates rise, their borrowing costs will increase directly, forcing expansion to shrink, while the original valuation was given according to an extremely high growth rate, and the expectation gap is very likely to widen.
The circular trading risk mentioned by the Federal Reserve is also worthy of attention.
Technology companies and start-ups invest in each other, purchase from each other, and endorse each other's revenues. When funds are loose, they can beautify their accounts and push up valuations. After financing tightens, they become a series of linked risk exposures. Once one company has a cash flow problem, the risk will spread along the cooperation chain. This is not a problem of valuation level, but a hidden danger at the credit level.
The biggest negative trigger recently is the Jensen Huang incident, which has shaken the pricing anchor of the entire chain.
It is still difficult to draw a conclusion about how much long-term impact this incident will have on NVIDIA's own products and orders.
In the past few years, the order expectations and valuation logic of many companies in chips, OEM manufacturing, storage, and downstream cloud vendors have been built on NVIDIA's strong position. Once this anchor shakes, everyone will re-evaluate the assets in their hands.
The big short seller Burry obviously seized this window, but his thinking has also changed.
In the past, he preferred to bet directly on individual stocks such as NVIDIA and Palantir, but now the focus has shifted to indices and semiconductor ETFs. He doubled his Nasdaq 100 put options while buying back short positions in the VanEck Semiconductor ETF, and TSMC and NVIDIA are exactly the heaviest positions in this ETF.
He compared the high point of the Nasdaq in early October to the peak of the bubble in 2000, and called out "bull market trap", focusing on whether NVIDIA's influence will gradually weaken.
Of course, his strategy may not be correct. After all, the difference between this analogy is very large.
During the 2000 bubble, a large number of companies had no cash flow and a mess of liabilities; now the balance sheets of leading US tech giants are quite healthy, and the valuation premium of the five major tech stocks relative to other stocks has converged to a very low level, far from the madness of that year.
Moreover, the threshold for this options bet is very high, and it can only make a profit if the index falls by more than 20% before expiration. The time, range and direction must all be correct.
There is another easily overlooked pressure coming from the shift of regulation.
After the Democratic Party takes office, the United States' attitude towards AI infrastructure will most likely gradually shift to control, and more constraints will be imposed on the energy consumption, land and power supporting facilities of computing power clusters and large data centers.
This is not to strangle AI, but it will disrupt the rhythm of unrestrained computing power stacking, prolong the expansion cycle, and increase construction costs.
Looking back at the capital expenditure curve, this variable will further lower the expansion slope. Originally, it was expected that computing power would continue to grow explosively, but now there are more constraints on approval, energy consumption and compliance, the high point comes earlier, and the realization rhythm is slower.
With the four layers of pressure superimposed, it is logical that the valuation bubble blown up by optimistic narratives is squeezed.
02
But we need to be clear that the bursting of the AI valuation bubble does not mean that the AI industry is going downhill, and the underlying demand supporting the long-term logic of AI has not disappeared.
There is also a very interesting signal that Goldman Sachs has mentioned many times.
The forward PE of leading US cloud giants is only 19.7 times, which is lower than the lower edge of their 20-year historical range, and even lower than the recognized defensive sector of consumer staples.
Many people regard them as high-elasticity growth stocks, but their valuations have fallen below those of defensive stocks, which shows that a large part of pessimistic expectations have been priced in, and not everyone is blindly optimistic.
In the long run, tightening regulation is not all bad. The disordered computing power arms race stops, forcing everyone to calculate an account: whether the invested computing power can bring stable revenue and profits.
Only when the business model is proven feasible and does not rely on endless money burning to expand capacity, can the AI story be more solid. It is a short-term pain, but in the long run, it is a process of eliminating the false and retaining the true.
After the bubble is squeezed out, the opportunity structure of the market will also follow a new set of logic.
The simplest strategy in the past two years was to go all-in on the upstream "shovel sellers", including chips, optical modules and storage. The closer they are to computing power, the better they perform. This logic is built on the continuous acceleration of capital expenditure, and it will be difficult to replicate in the future.
The market's assessment criteria will shift from "who is expanding production" to "who can earn real money".
The first category worth tracking are cloud giants whose valuations have fully corrected. The slowdown in capital expenditure growth means less investment and improved free cash flow for them. With solid balance sheets and diversified revenue structures, they are gradually transforming from high-volatility growth stocks to defensive varieties with growth attributes, and their cost-effectiveness will gradually stand out in the volatile market.
The second category are AI application enterprises that have achieved real implementation. In the past, the market focused all its attention on computing power, and the value of the application side was covered.
In the next stage, the market will keep testing whether AI can help customers save costs and increase revenue, and whether there is stable payment and renewal. Only companies that obtain real productivity dividends and have clear ROI calculations can take over as the new main line.
This is also the "succession echelon" mentioned by Goldman Sachs. Whether the transition is smooth determines whether this round of adjustment will achieve a soft landing or continue to fall sharply.
The third category are cyclical sectors that have priced in a large number of pessimistic expectations, such as Korean memory chips. The stock price has implied the expectation that earnings will be sharply cut by 70% in 2027, but the order schedule has been sold to 2029, and the PB has been suppressed to a historical low. The odds seem high, but the risks are not small, and the volatility is extremely intense.
Goldman Sachs is not blindly bullish on Asia either. The core concern is that the concentration is too high, with several targets such as TSMC, Samsung, and SK Hynix accounting for too large a proportion. Once the logic reverses, the pullback will be very damaging.
Conversely, there are many targets that need to be actively avoided. Small and medium-sized GPU cloud vendors that are highly bound to NVIDIA's orders, have no stable cash flow and rely on financing for blood transfusion, and pure money-burning start-ups without implemented products are the most vulnerable groups in this round of stress test.
03
Overall, the current AI narrative is indeed facing great pressure. Under this market background, if people still hold the idea of previous years to heavily allocate upstream computing power and continue to pay for "faster and faster growth", this is an inertial behavior that is very easy to lose money.
A more prudent approach is to reduce concentrated bets on a single track and appropriately diversify positions.
The market has now shifted from a unilateral bull market to two-way volatility, and the value of hedging and multi-asset allocation will be higher than in previous years.
This round of test for the AI narrative is not testing whether AI can change the world, but whether these investments can be converted into stable returns. The process of the bubble dissipating is destined to be difficult, and the volatility will last for a long time. But after the market clears, assets that can withstand the test and have solid cash flow will usher in the next stage of market trend that can go further.
During this period, just stabilize your mindset, withstand the pressure, and wait slowly for the dawn.
This article is from the WeChat Official Account "Gelong", written by Du Xingxia, and published with authorization from 36Kr.