NextFin News - The artificial-intelligence trade that has powered the S&P 500 to a roughly 10.6% gain this year is now flashing its clearest warning signs, as traders weigh record capital spending by the largest technology companies against a funding gap that is projected to open as soon as the third quarter of 2026. The 10-year Treasury yield, at 4.94% on September 17, has become the market's tripwire: a move above 5% makes the valuation math behind the AI buildout much harder.
The tension is stark, and it sits inside the same investor base. In Bank of America's latest Global Fund Manager Survey of institutional investors, 38% named AI hyperscaler capital expenditure as the most likely source of a systemic credit event - the top-ranked risk for a second consecutive month, ahead of private credit at 23%. Yet 71% of those same managers said they do not expect any hyperscaler to cut spending in 2026, up sharply from 61% a month earlier. The AI trade is simultaneously the market's biggest fear and its strongest conviction.
The Funding Gap No One Wanted to Model
The core of the worry is arithmetic, not sentiment. The five largest hyperscalers - Microsoft, Amazon, Alphabet, Meta and Oracle - are raising cash capital expenditures faster than the cash their operations generate. Aggregate cash capex across the group is on track to overtake operating cash flow around the third quarter of 2026, the point at which their combined free cash flow reaches zero, according to Epoch AI, which extracts data directly from SEC EDGAR filings.
The divergence is widening. Operating cash flow across the five companies is growing at roughly 23% a year, while cash capex is compounding at about 70% a year. On Epoch's quarterly data, aggregate cash capex rose from $36.6 billion in the first quarter of 2022 to $162.6 billion in the second quarter of 2026, while operating cash flow moved from $65.6 billion to $177.5 billion over the same stretch. The trend lines cross in the September quarter of 2026.
FactSet's numbers point the same way. Aggregate cash capex for the five hyperscalers is expected to exceed $690 billion across their fiscal-2026 periods, an increase of more than 80% year over year and the largest annual jump of the cycle. Calendar-2026 capex guidance sits close to $800 billion when finance leases and customer prepayments are included, and spending is projected to top $900 billion by fiscal 2028. Alphabet has guided to $195 billion to $205 billion in 2026 capital expenditures; Meta to $130 billion to $145 billion; Amazon's plan sits near $220 billion. Microsoft alone spent $41 billion on capex in its most recent quarter, with roughly two-thirds directed at shorter-lived assets such as CPUs and GPUs.
That is the first-order fact. The second-order question is how it gets paid for. When capital spending exceeds the cash a business generates, the difference must come from somewhere else: drawing down cash balances, issuing debt, selling equity, or signing leasing arrangements that push payments into the future. Several of the hyperscalers have already moved. Oracle announced an equity-and-debt financing plan for calendar 2026. Amazon launched an 11-part U.S. high-grade bond offering in March. Alphabet proposed an $80 billion equity capital raise to expand AI infrastructure and compute. Each step is rational for a company with a multiyear AI roadmap - and each is a departure from the years when the AI buildout was funded out of free cash flow.
Why the Bond Market Is the Real Referee
The reason traders are wary is that AI valuations are long-duration assets priced off the risk-free rate. When the 10-year Treasury yield rises, the present value of earnings expected years from now falls - mechanically, with no judgment required. That is why Ruchir Sharma, founder of Breakout Capital, said in a July interview that the AI bubble could burst if the 10-year Treasury yield moves above 5%. With the yield at 4.94% on September 17, the market is roughly 6 basis points from that threshold.
"There are plenty of signs that we are now in the late stages of a bubble in AI," said John Higgins, chief economic adviser for financial markets at Capital Economics. The firm forecasts the S&P 500 will reach 8,250 in 2026 before falling to 6,500 by the end of 2027.
Sharma's framework rests on what he calls the four O's: overinvestment, overvaluation, over-ownership and over-leverage. On his read, technology capital expenditure as a share of GDP is now comparable to the 2000 dot-com bubble; the share of American financial wealth held in equities is higher than it was in 2000; and while the early AI builders funded expansion with cash, companies such as Meta and Amazon are now issuing debt for the arms race. The trigger, in his view, is not earnings disappointment but the cost of capital.
The transmission channel runs through three doors. First, higher yields compress the multiples that growth stocks trade at. Second, they raise the actual financing cost for the hyperscalers as they tap debt markets, turning today's growth plan into tomorrow's interest expense. Third, they hit the leveraged players on the other side of the trade - the quant funds and momentum managers who have crowded into semiconductors. A monthly institutional-positioning survey found long global semiconductors to be the most crowded trade among 53% of managers. When a crowded trade meets a rising discount rate, the unwind is fast and indiscriminate.
The benchmark index closed at 7,551.81 on September 16, down from 7,718.60 on September 4 - a pullback of more than 2% in under two weeks. Nvidia, the bellwether of the AI trade, fell 3.4% to $210.96 on September 15 after concerns that calls from AI leaders to slow development would hurt chip and data-center hardware firms, and was changing hands near $214 in the following session. The stock's 52-week range of $164.28 to $236.54 shows how much of the recent move was multiple, not earnings.
Cyclical Squeeze or Structural Shift? Both, and That Is the Problem
Is this a cyclical pullback that will revert, or a structural break that will not? The answer is both - and confusing the two is how investors lose money.
The structural leg is real. AI is not a fad, and the infrastructure being built has genuine demand behind it. Microsoft's commercial remaining performance obligation - its contracted backlog - reached $678 billion in the fiscal fourth quarter reported in late July, up 84% year over year, with management expecting roughly 30% of it to convert into revenue within 12 months. At Alphabet, Cloud, Gemini, TPU systems and enterprise AI demand are outpacing available capacity. Nvidia CEO Jensen Huang, writing on X on September 8, described the company's chips as "a productive, revenue-generating asset" and said "NVIDIA compute is fungible, durable and highly rentable." The capex is front-loaded; the returns arrive over a longer horizon. That is the pattern of a platform shift, not a bubble in the dot-com sense, where companies spent on fiber-optic dark capacity that never generated a dollar.
But the cyclical leg is equally real, and it is the one that moves markets over the next six to twelve months. Capex growing at 70% a year against operating cash flow growing at 23% cannot persist without external financing, and external financing has a price that moves with the bond market. This is a funding-cycle squeeze: the gap between cash out and cash in widens fastest exactly when the cost of bridging it is rising. History offers a template. In the late 1990s, telecom capital expenditure as a share of GDP peaked before the bubble burst; the spending was real, the fiber was real, but the financing structure broke first. The same sequence - real demand, unrealistic financing assumptions, a rate-driven repricing - is what traders are now pricing in.
Three historical cycles show the pattern. First, the 1999-2001 telecom buildout, where capex peaked as a share of GDP before revenues caught up. Second, the 2021-2022 pandemic-era supply-chain capex, where companies over-ordered chips and containers, then worked down inventory through 2023. Third, the 2014-2016 energy capex cycle, where oil companies cut spending within quarters of the price peak because debt markets closed. The common thread: capex cycles overshoot on the way up and correct faster than the underlying demand justifies, because the financing window slams shut before the revenue window does.
The evidence for the structural call is simpler: the rules of the game have changed. Hyperscalers are no longer buying boxes; they are building a new compute utility. Once a model is trained on a customer's data and embedded in their workflow, switching costs are high. That is a durable revenue stream, and it is why the hyperscalers are willing to run free cash flow to zero rather than slow down.
So the base case is a cyclical squeeze layered on a structural uptrend. The stocks can fall 20% to 30% on a funding scare without the AI thesis being wrong. That is exactly the environment the September survey describes: managers are underweight bonds, underweight cash, and still do not expect capex to fall.
The Counter-Thesis: This Time the Balance Sheets Are Strong
The strongest case against the bearish read is also the simplest: this is not 2000. The dot-com companies had no earnings and no cash flow. The hyperscalers have both, in staggering amounts. Microsoft, Amazon, Alphabet, Meta and Oracle remain profitable and increasingly so - Epoch's own analysis notes that while capex is paid upfront, it counts against profits gradually through depreciation. A company that can fund $800 billion of capex out of a combination of cash flow and investment-grade debt is not a dot-com shell. Major Wall Street banks, including Goldman Sachs and J.P. Morgan, have argued that the AI growth is fundamentally justified, and Vanguard's 2026 economic outlook frames the risk as one-sided to the downside for credit rather than an imminent equity collapse.
There is force to this argument. The hyperscalers are not borrowing at junk rates; they are issuing investment-grade paper at yields that, while higher than 2021, are still manageable against their earnings power. Oracle's net debt has accelerated, but from a small base relative to the group's cash generation. And the demand side is not imaginary: enterprises are signing multiyear AI contracts, and the commercial backlogs are measurable.
But the counter-thesis answers only the solvency question, not the valuation question. A company can be solvent and still see its stock halve when the discount rate rises. The dot-com bust did not kill the internet; it killed the valuations attached to it. The same distinction applies now. If the 10-year yield holds below 5% and capex growth moderates toward 30% a year, the structural bulls win and the pullback is a buying opportunity. If the 10-year breaks 5% and stays there while capex keeps compounding at 70%, the funding gap becomes a refinancing problem, and the cyclical bears win.
The falsifying signal is specific. If the 10-year Treasury yield prints at or above 5% for five consecutive trading sessions while hyperscaler capex guidance for 2027 comes in above $900 billion, the bearish funding-gap thesis is confirmed. Conversely, if the 10-year falls back below 4.5% and at least two hyperscalers guide to slower capex growth in their next earnings reports, the risk-off read is wrong and the dip is an entry point for the structural trade.
What Comes Next: Three Horizons
Short term (weeks to a quarter): sentiment and liquidity dominate. The September pullback can extend if the 10-year yield retests 5% and risk appetite thins. Semiconductors, as the most crowded trade, are the first to give back gains. Watch the yield, not the headlines.
Medium term (six to eighteen months): fundamentals take over. The key data points are quarterly capex guidance from the hyperscalers and the monetization of their AI services - Azure growth, Google Cloud margins, AWS AI revenue. If capex growth slows toward 30% while AI revenue accelerates, free cash flow troughs and the group rerates higher. If capex stays above 50% growth with no offsetting revenue, the funding gap widens and credit spreads do the talking.
Long term (three to five years): the structural question resolves. Either AI becomes a durable profit pool that justifies the buildout, or it becomes a utility with commoditized margins. The winners will be the companies that converted capex into recurring revenue, not the ones that simply spent the most.
Base case, on this analysis: a choppy, less unified AI trade, with the S&P 500 range-bound between 7,200 and 8,000 through year-end as the market digests the funding gap. Upside case: yields fall back below 4.5%, capex discipline returns, and the index retests its recent highs near 7,750. Downside case: the 10-year breaks 5% and holds - triggering the threshold Sharma set - and the S&P 500 tests 6,800 as the crowded semiconductor trade unwinds. Capital Economics, by contrast, sees the index reaching 8,250 in 2026 before falling to 6,500 by the end of 2027.
The market is not betting against artificial intelligence. It is betting that the bill for building it will come due faster than the revenue arrives - and that when it does, the cost of paying it will be higher than anyone planned for.
Data as of September 17, 2026.
Explore more exclusive insights at nextfin.ai.
