AI真实增长 vs 美银70%熊市信号：美股到底该不该跑？
TL;DR
- Bank of America warns of too many risk signals in the US stock market, advising investors to take profits and control risks.
- AI leaders still have revenue and capital expenditure support, but the market has already priced in a lot of future growth.
- Related tickers: SPY, QQQ, NVDA, MSFT, GOOGL, AMZN, META, AVGO, AMD, SOXX.
US stock investors now face a situation that goes beyond simply being bullish or bearish.
On one side is the Bank of America US Equity and Quantitative Strategy team. Led by Savita Subramanian, the team published a client report on June 5 titled "Too many red flags. Take profits." According to an Axios report on June 9, the report believes there are already too many risk signals in the US stock market and offers a more direct portfolio suggestion: take profits.
On the other side is the still-strong AI fundamentals. Microsoft, Google, Amazon, and Meta continue to increase AI and data center capital expenditures. Nvidia's data center demand remains the core anchor of the semiconductor cycle. Unlike the 2000 internet bubble, the current leaders are giants with cash flow, profits, cloud revenue, and chip orders.
So the real question has shifted from "Is AI a bubble?" or "Is Bank of America calling a top?" to another, harder question: When historical top signals and real AI growth coexist, how should investors understand the current risks in the US stock market?
The answer might be more uncomfortable than simple bearishness: The AI bull market may not be over, but it has moved from a phase of "buying growth" to a phase of "testing the speed of growth realization."
Bank of America's Warning is About Deteriorating Odds
The value of the Bank of America report lies in placing the current market within a historical risk framework, rather than providing an exact top timing.
According to multiple financial media reports citing the Bank of America report, approximately 70% of the 10 bear market warning indicators they track have been triggered. This ratio is close to the average level seen before the S&P 500 topped multiple times since 1990. The bank's framework also shows that 17 out of 20 valuation metrics for the S&P 500 indicate statistical overvaluation, with 8 of them higher than the peak of the 2000 internet bubble. The CAPE (Cyclically Adjusted Price-to-Earnings ratio) or P/E10 is around 40, in a historically extremely high range.
These numbers can each be refuted individually. High valuations don't mean a drop tomorrow. Historical signals have been accurate, but not every time. And the stronger profits of AI companies truly differentiate today from 2000. However, when valuation, market breadth, style divergence, and momentum all show extreme readings simultaneously, Bank of America's key point is closer to: The market could still be held, but the odds have worsened.
Market breadth is the key factor here. Indices are still at highs, but the rally is increasingly dependent on a few AI and tech leaders. The current US stock market exhibits a narrow leadership characteristic similar to historical topping phases: A few stocks contribute the majority of index gains, the percentage of S&P 500 components above key moving averages has declined, and many individual stocks are not close to their own highs. The strength at the index level masks the declining internal participation.
Style divergence reinforces the same signal. Bank of America notes that the median return difference between the best and worst quintile stocks in the tech sector is about 120 percentage points, the highest since February 2000 and close to the 130 percentage points seen just before the March 2000 top. This looks more like capital concentrating bets on a few perceived certainties, rather than the broad-based diffusion seen in a normal bull market.
For investors holding SPY, QQQ, NVDA, or SOXX, the most dangerous aspect of this structure is the lower margin for error. Indices could certainly continue to rise, but when gains are increasingly determined by a few stocks, any deviation in the earnings, guidance, capital expenditure returns, or valuation assumptions of a single leader can be amplified into a drawdown for the entire portfolio.
AI This Time Cannot Simply Be Compared to 2000
If one only looks at Bank of America's valuation and breadth signals, it's easy to directly compare the current market to 2000. But this comparison is only half right.
A typical characteristic of the 2000 internet bubble was that many companies lacked mature business models, and investors were primarily trading on the vision of "the internet changing the world." The AI leaders today are different. The cloud and AI businesses of Microsoft, Google, Amazon, and Meta are already reflected in real revenue, capital expenditure plans, and data center demand. Nvidia is not just a narrative center; it's also a chip supplier highly concentrated in profits and cash flow.
Nvidia's latest earnings report provided the strongest support for the bulls. The company's FY2027 Q1 earnings report, released in May 2026, showed quarterly revenue of $81.6 billion, with data center revenue of $75.2 billion, up 92% year-over-year. In the face of these numbers, simply calling the AI rally a "bubble without fundamentals" is not very convincing.
Based on this, AI optimists, including management teams of large tech companies and growth investors, refute the bubble theory. They argue that this rally is more like an infrastructure cycle: training and inference demands drive GPU, networking, storage, power, and data center construction; cloud providers trade higher CapEx for future AI service revenue; and enterprises further integrate AI into software, advertising, search, office suites, and development processes.
This framework has a factual basis. Over the past few earnings seasons, large cloud providers have consistently emphasized strong AI demand and sustained cloud business growth. Nvidia's data center revenue has become a key pillar of the US stock market's earnings growth narrative. Broadcom, AMD, data center, and power infrastructure companies are also part of this same investment chain. The market's willingness to give these companies higher valuations is not just due to a compelling story; orders, revenue, and profits are indeed materializing.
This is also why Bank of America's signals cannot be crudely interpreted as "the AI bull market is over." If the underlying fundamentals are still improving, a high-valuation rally can last longer than historical precedents suggest. Especially in a market where passive flows, index weights, and institutional allocations reinforce the dominance of leaders, the strong getting stronger is itself a part of the capital flow mechanism.
But AI being real does not equal valuations being safe. A common misunderstanding can occur here: as long as the technological revolution is real, the price is not expensive. Historically, many bubbles were built precisely on real technology being priced in too early and too fully. The internet truly changed the world, but investors who bought many internet stocks in 2000 still experienced a long period of valuation compression.
The core divergence in the current AI rally is shifting from "Is AI useful?" to "How many years of future growth has the market already priced in?" The reason Bank of America's historical signals are important is precisely that they remind investors: even if the fundamentals are real, risk still rises when prices already reflect too much future good news.
Pressure Shifts to Revenue and Cash Flow
The AI bull market enters its most difficult phase, not because demand suddenly disappears. The real change is that the market is starting to demand more proof.
Over the past two years, investors were willing to pay high valuations for AI leaders because the growth path seemed clear: cloud providers increase CapEx, chip companies sell more high-end GPUs, data center and networking equipment companies receive orders, and future enterprise applications unlock even greater revenue. Entering 2026, the market needs to see not just continued investment, but whether that investment can translate into sufficiently high revenue, profit margins, and free cash flow.
Capital expenditure is the focal point of this issue. Microsoft, Google, Amazon, and Meta are consistently increasing their AI and data center investments; the direction is basically clear. However, estimates from different institutions and media regarding the specific scale vary significantly. More importantly, investors are starting to worry about the pressure higher CapEx puts on free cash flow and return on invested capital. It cannot be simply stated as "AI investment is unrecoverable," but as the investment curve steepens, the market's requirements for the return curve will also increase.
For Microsoft, Google, Amazon, and Meta, continuing to increase AI investment is a strategic necessity. Whoever stops risks falling behind in cloud, search, advertising, office suites, models, and developer ecosystems. However, from a shareholder's perspective, the higher the CapEx, the more future earnings reports need to demonstrate that these investments can generate revenue growth, stable margins, and resilient cash flow.
For the semiconductor chain represented by Nvidia, Broadcom, AMD, and SOXX, the logic is slightly different. They are direct beneficiaries of the AI investment cycle, with orders and profits materializing earlier. But precisely because the market already views them as core winners of the AI infrastructure cycle, any sign of slowing CapEx, delayed procurement, or an emphasis on investment discipline from downstream cloud providers will cause semiconductor valuations to react first.
This creates a more fragile feedback loop. Cloud giants increase CapEx, supporting chip company revenue. Chip company high growth supports index gains. Index gains and earnings upgrades, in turn, reinforce market confidence in the long AI cycle. If any link in this chain slows down, the market may not necessarily face the "end of AI," but rather a need for valuations to re-align with the speed of realization.
H2 Earnings Must Prove Risk Can Be Covered by Growth
Bank of America's 70% bear market signal won't automatically turn into a market top, and the strong earnings of AI leaders won't automatically eliminate valuation risk. What truly needs to be verified next is whether sustained growth can cover these risk signals emanating from valuations and market structure.
The most direct observation window is the second half of 2026 earnings season. Investors need to see continued AI revenue growth from large tech companies, alongside profit margins not being significantly eroded by CapEx and depreciation pressures. While cloud providers continue to invest, they must also demonstrate sufficiently strong customer demand. The orders and guidance from semiconductor companies like Nvidia, Broadcom, and AMD will reflect whether the pace of downstream investment is slowing.
Another variable is market breadth. If the S&P 500 and Nasdaq continue to hit new highs, but fewer stocks participate in the rally, and high P/E stocks systematically outperform low P/E stocks, the historical topping structure that Bank of America points to will become harder to ignore. Conversely, if earnings strength spreads to more sectors and indices become less dependent on just a few AI leaders, the risk signals might have a chance to be gradually absorbed by time and performance.
For the average investor, the current environment is more suitable for a portfolio and concentration check. Simply using "bullish AI" or "bearish US stocks" won't solve the problem. AI may still be the most important investment theme for years to come, but when valuations, breadth, and CapEx pressure all rise simultaneously, continuing to hold it has transformed from an early trend discovery into a bet on the speed of realization.
