NextFin News - Wells Fargo & Co. has become the second major bank to warn on US stocks this week, as its equity strategists flag "broad cautiousness" heading into September amid rising anxiety over whether the artificial-intelligence investment boom can keep paying for itself. The shift, led by chief equity strategist Ohsung Kwon, lands as the market enters the calendar year's weakest month for equities, faces a midterm election in November, and digests an oil spike that has revived inflation fears.
The caution is not a call for a crash. Wells Fargo's own sentiment indicator remains more positive than negative after the mid-August selloff, and the strategists expect fears around AI capital expenditure to peak rather than spiral. But the fact that two of Wall Street's largest banks have turned defensive within days of each other signals that the burden of proof has shifted onto the AI trade: investors are no longer asking how much more can be spent, but whether the spending will generate returns.
The Warning and the Backdrop
Wells Fargo's analysts, led by Kwon, cited three converging pressures. First, investors are growing increasingly anxious about the sustainability of the AI-investment boom. Second, September is historically the worst month on the calendar for equities. Third, the upcoming midterm elections add a layer of policy uncertainty around the AI trade. The bank joins JPMorgan Chase & Co., whose technical strategist Jason Hunter told clients earlier in the week to consider trimming stock positions as technical warning signs mounted, even while acknowledging that major indexes still show an upward trend.
Hunter's note put the concern in precise terms: "In isolation, the broad indexes are still bullishly trending and have not yet given any clear warning signs that a downturn is imminent. However, a deeper look at market internals, shifts in leadership, and the technical setups across a number of markets that typically lead, raise concern heading into late-summer/early-fall bearish seasonality." He also flagged that AI stocks are showing similarities to the moves that preceded the 1999-2000 technology peak, when hardware and communications-equipment spending ran years ahead of the revenue that was supposed to justify it.
The timing is not incidental. Since 1928, the S&P 500 has produced an average return of negative 1.17% in September, and the month has been negative in 56% of years - the only month with a negative long-term return track record, according to Bank of America data. Midterm election years compound the pattern: since 1931, the S&P 500 has averaged a 4.7% return in midterm years, roughly half the 9.5% average for non-midterm years, though still positive, Capital Group data shows. Historically, midterm years deliver below-average returns and higher volatility for most of the twelve months before recovering toward year-end once results are known.
Then there is oil. Brent crude jumped 3.5% to $91.30 a barrel and West Texas Intermediate climbed 3.7% to $86.49 after US forces struck Iranian missile launchers near the Strait of Hormuz, a chokepoint through which about one-fifth of the world's oil typically flows. An energy shock does not need to become a full-scale supply crisis to matter for equities: it needs only to nudge the inflation path higher and remind investors that the Federal Reserve's rate path is not a one-way street. As of August 31, interest-rate derivatives implied a 57.5% probability of a 25-basis-point rate increase at the September policy meeting.
Why AI Capital Expenditure Is Now the Fault Line
The central question is not whether AI is real. It is whether the spending required to build it is outpacing the revenue it can produce - and what happens to equity multiples when that gap becomes the market's focus.
The scale of the build-out is unprecedented. The largest US technology firms now plan to spend as much as $725 billion on capital expenditure this year, primarily on AI data-center equipment, with Alphabet Inc. and Meta Platforms Inc. both raising full-year guidance and Microsoft Corp. setting a first estimate of $190 billion through December, matching Alphabet. Amazon.com Inc. kept its guidance unchanged at $200 billion. One research estimate puts combined hyperscaler spending at roughly 102% of cloud revenue for 2026, meaning Amazon, Alphabet, and Microsoft are recycling essentially all of their cloud income back into AI infrastructure.
That math is the transmission mechanism behind the caution. When capital expenditure consumes nearly all of the cash generated by the divisions funding it, three things follow. First, free cash flow growth stalls even as reported earnings rise, because depreciation on the new data centers begins to weigh. Second, the market's valuation of the AI trade rests increasingly on revenue that has not yet been monetized - the promise that today's infrastructure spend becomes tomorrow's recurring software and services income. Third, any sign that monetization is lagging forces a double compression: investors cut both the earnings estimates and the multiple they are willing to pay for them.
Wells Fargo expects capital expenditure fears around the AI build-out to peak. That framing is important. A peak in fears is not the same as a peak in spending - spending can keep rising even as investor anxiety about it tops out, if companies demonstrate that the infrastructure is being absorbed by demand. The strategists' view implies the pivot point is not the dollar figure but the evidence of return on invested capital.
The strategists wrote that they see "broad cautiousness" heading into September. The phrase captures the breadth of the concern rather than a single-name problem: this is not a warning about one company's guidance, but about the market's dependence on one narrative holding together.
Cyclical Weakness or Structural Repricing?
The right read of this moment determines whether investors should treat it as a buying opportunity or a regime change. The answer is both - and that is what makes it dangerous to oversimplify.
The cyclical leg is real and familiar. September seasonality, midterm-election uncertainty, and an oil-driven inflation scare are all mean-reverting forces. September has been negative in only 56% of years since 1928, which means it has been positive 44% of the time. Midterm years have still delivered positive returns on average. Oil shocks that do not escalate into sustained supply disruptions tend to fade as the geopolitical premium drains out of the price. These are the conditions under which "buy the dip" has worked for a decade.
But a structural leg runs underneath the cyclical noise, and it is the one Wells Fargo and JPMorgan are really flagging. The AI capex cycle is structurally different from the cloud build-out of the 2010s because the payoff horizon is longer, the capital intensity is higher, and the revenue model is less proven. Cloud computing monetized quickly because it replaced a known line item - the corporate data center - with a monthly subscription. Generative AI is still searching for its equivalent killer application at consumer and enterprise scale. When a market values infrastructure spending as though the returns are already visible, the eventual recognition that they are not is a structural repricing, not a cyclical dip. It does not revert on its own; it requires either the revenue to arrive or the multiple to fall.
History offers a cautionary analog. The fiber-optic build-out of the late 1990s was not wrong about the internet - demand did eventually arrive, in volumes the builders imagined and beyond. But the capital was deployed years ahead of the revenue, and the equity holders who funded the build-out at peak multiples waited more than a decade to recover. The lesson is not that AI is a bubble; it is that being right about the technology and being right about the stock are two different bets.
The Second-Order Trade the Market Has Not Fully Priced
The first-order consequence of cautious bank notes is obvious: some investors trim exposure, and volatility rises. The second-order consequence is more consequential and less discussed.
If AI capex fears peak while oil-driven inflation fears rise at the same time, the two forces converge on the discount rate from opposite directions. Higher inflation expectations push the risk-free rate up; uncertainty about AI earnings quality pushes the equity risk premium up. A stock's value is the present value of its future cash flows, and both inputs are moving against it simultaneously. This is why a market that looks merely range-bound can still be doing significant damage beneath the surface: the index can hold up on the weight of a few names while the implied cost of capital for the whole market quietly rises.
There is also a cross-asset channel. When long-duration growth stocks become less attractive, capital does not simply leave the market - it rotates. Energy, defense, and cash-rich value names tend to benefit from an oil shock and a higher-rate environment, while the same conditions hurt the long-duration tech complex that has led this cycle. The "tale of two markets" framing that has appeared in recent analysis - haves and have-nots within the same index - is the visible symptom of that rotation. An investor who is long the index but underestimates the dispersion can be right about direction and wrong about every position.
The Strongest Case Against the Caution
The bull case deserves its full weight, because it is not weak. Earnings growth remains intact, and the market is being asked to discount a profit cycle that has not broken. The S&P 500 was up roughly 13% year-to-date at recent highs, and breadth has been healthier than in the narrow leadership of prior years: nearly 70% of S&P 500 components have traded above their 200-day moving average, evidence of an advance that is broad rather than carried by a handful of megacaps.
More importantly, the caution itself is measured. Wells Fargo's sentiment indicator remains net positive. The strategists expect capex fears to peak, not spending to collapse. JPMorgan's technical team acknowledged the indexes still trend upward. This is risk management, not capitulation - and risk management in an uptrend usually produces pullbacks, not bear markets.
The counter-argument, then, is that the market is being asked to price in a problem that has not yet shown up in the data. Until AI revenue fails to materialize, until free cash flow actually contracts, or until the central bank is forced to tighten into a weakening economy, the cautious notes are a hedge rather than a verdict. And hedges cost money in a market that keeps grinding higher.
The answer to that case is specific. The risk is not that the AI thesis is wrong today; it is that the equity market has moved ahead of the earnings reality by a margin that leaves no room for the thesis to be merely on time. When spending is running at 102% of the revenue funding it, the market is pricing flawless execution for years. Flawless execution is a low-probability path in any capital-intensive cycle. The cautious banks are not betting against AI; they are betting against perfection.
What to Watch and What Would Prove It Wrong
The falsifying signal is quantifiable: if the mega-cap technology complex reports that AI-driven revenue growth is tracking at or above the pace of capital expenditure growth for two consecutive quarters - that is, capex intensity stops rising as a share of revenue - then the structural repricing thesis is wrong, and this caution will look like a missed rally. Conversely, if capex continues to accelerate while AI revenue growth decelerates, the gap widens and the multiple has further to fall.
By time horizon, the picture splits cleanly:
- Short term (weeks): sentiment and liquidity dominate. September seasonality and the midterm headline risk favor choppy, lower-volatility-adjusted returns. The oil premium can extend the pressure if it pushes inflation expectations above recent highs.
- Medium term (three to six months): fundamentals take over. The next two quarters of hyperscaler earnings will show whether the capex is being absorbed. This is the decision window.
- Long term (years): the structural question resolves. Either AI monetizes at scale and the current spend looks cheap in hindsight, or the return on invested capital disappoints and the market rerates the complex toward a lower terminal multiple.
Three scenarios frame the path:
- Base case: capex growth moderates toward revenue growth, oil stabilizes, and the market grinds sideways with elevated dispersion between sectors. The index gives back part of its year-to-date gain but avoids a bear market.
- Upside case: AI revenue inflects faster than spending, the rate path turns easier, and the midterm outcome removes policy uncertainty. The caution becomes the bottom for the next leg higher.
- Downside case: capex keeps accelerating while monetization lags, oil holds above $90 and feeds into inflation, and the rate path stays restrictive. The multiple compresses even if earnings hold - a 2022-style rerating without an earnings recession.
The beneficiaries of this setup are the parts of the market that do not depend on falling discount rates: energy producers, defense contractors, and cash-generative value names with pricing power. The exposed are the long-duration growth complex and any company whose valuation assumes AI revenue arriving on a schedule it has not yet demonstrated.
The central judgment is this: the market is not being asked to abandon the AI trade, but to stop paying for it as though the returns are already certain. September's seasonality and the midterm fog are the occasion for the caution, not the cause. The cause is a capital cycle that has run ahead of its own receipts - and capital cycles, eventually, collect.
Explore more exclusive insights at nextfin.ai.
