NextFin News - Technology funds have been pulling in money at roughly 13 times the rate of the rest of the market, a flow regime that has powered the artificial-intelligence rally for more than a year. Now the rally's own architects are raising the first serious doubt, and the question investors face is whether the flows that built the trade can hold when the people building the models say to slow down.
Over the weekend of September 12-13, Anthropic chief executive Dario Amodei published an essay calling for the industry to pace the development of frontier AI models, a position endorsed within hours by OpenAI's Sam Altman and Elon Musk. Global AI-related shares fell on Monday, September 14, with chipmakers and AI suppliers bearing the brunt of the selling. The move is small compared with the gains of the past two years, but it lands on a market that has become unusually dependent on one kind of money: technology and AI-themed funds that, according to the fund-flow data behind the 13-to-1 figure, have been outpacing every other corner of the market by a wide margin.
The stakes are higher than a single sector rotation. When one theme commands 13 times the flow rate of everything else, the trade stops being about stock picking and becomes about market structure. The scare that began with an essay is now a test of whether the money that built the AI rally is patient capital or momentum capital — and those two types of money behave very differently when the story changes.
The Flow Regime That Built the AI Rally
The backdrop is a concentration of investor money that is hard to overstate. Global technology equity funds attracted about $195 billion in net inflows over the 12 months through late August 2026, the most of any major sector and more than the next nine sectors combined, according to data compiled by Deutsche Bank Asset Management from EPFR and Haver Analytics. Those inflows doubled over the four months leading into the end of August. For comparison, industrials and materials funds, ranked second and third, drew about $55 billion and $45 billion respectively.
The concentration is not limited to global sector funds. In the United States, technology sector funds took a record $19 billion of the $25 billion that flowed into all sector funds through July, State Street Global Advisors reported, even as the technology sector returned roughly minus 10 percent over the same period. In June alone, $13 billion of $17 billion in sector inflows went to technology, a 78 percent share well above the sector's 45 percent share of sector assets. AI-themed exchange-traded funds added more than $20 billion in net inflows during 2026, according to fund-industry roundups.
Put differently, investors have been paying up for exposure to a trade that has not been cheap, and they have been doing it with fresh money rather than recycled gains. That is the definition of a flow-driven rally: demand for the shares exceeds the supply of sellers because new capital is arriving faster than fundamentals are changing. It works until it doesn't.
The mechanics matter. A fund that receives net inflows must deploy them, and a thematically mandated fund has limited places to put the money. It buys the same names every day — the chip designers, the memory suppliers, the data-center builders — regardless of price. That mechanical buying lifts every holder's mark-to-market, which shows up as performance, which attracts more inflows, which forces more buying. The loop is circular, self-validating, and invisible while it is working. Investors see rising prices and call it insight; in reality it is plumbing.
What the AI Scare Actually Says
The trigger was not an earnings miss, a cancelled order, or a downgrade. It was an argument. Amodei's 3,800-word essay, titled "We Must Pace the Frontier," called for a global slowdown in the pace at which frontier models gain new capabilities, alongside additional safeguards including independent third-party evaluations. Altman backed the proposal; Musk wrote, "Dario is right."
The market read the subtext quickly: if the companies buying the chips want to slow the rate of model development, the growth curve for computing demand could flatten sooner than the stock prices assume. SK Hynix contracts slid about 2.5 percent on a blockchain-based perpetual-futures venue over the weekend, the first price to move. On Monday, losses spread across chipmakers and AI suppliers in Asia, Europe, and the United States.
But a weekend survey of market strategists reached a more measured conclusion: the slowdown calls are likely to weigh on chipmaker and supply-chain stocks in the near term, but will probably have limited long-term impact as long as spending on computing infrastructure remains strong. The report noted that no company had cut a chip order, cancelled a data center, or revised its capital-spending plan as of the weekend. The selling was about expectations, not orders.
"For a long time, the market treated AI spending as unquestionably positive," Nigel Green, chief executive of the financial consultancy deVere Group, said in June as doubts first surfaced. "Investors are now becoming more demanding. They want evidence that unprecedented spending will translate into unprecedented profits."
That sentence captures the pivot. For two years the market priced AI capital expenditure as inherently value-creating. The question has shifted to whether the revenue on the other end will justify it.
Why 13-to-1 Flows Are More Fragile Than They Look
Here is the mechanism that makes the flow ratio dangerous. When money arrives at 13 times the rate of the rest of the market, it does three things at once. It lifts valuations faster than earnings can grow. It draws more issuers into the trade, as companies rebrand themselves as AI beneficiaries. And it creates a reflexive loop in which rising share prices validate more capital expenditure, which validates more inflows, which lifts prices again.
The loop is self-reinforcing on the way up and self-reversing on the way down. That is why a 2.5 percent weekend move in one memory chip on a thin venue can matter: it is not the size of the move that matters, it is what the move implies about the next marginal buyer. In a flow-driven market, the marginal buyer is a fund with a mandate to own AI, not a valuation investor waiting for a discount. If that fund pauses, there is no natural buyer underneath.
The financing side of the trade shows how much leverage the rally has taken on. The five largest hyperscalers — Amazon, Microsoft, Alphabet, Meta, and Oracle — collectively issued approximately $200 billion of investment-grade debt in the first half of 2026, almost double the issuance during all of 2025, according to iShares' Fall 2026 Investment Directions. Technology sector borrowing has accounted for about 20 percent of all new U.S. investment-grade debt issued. When the AI buildout is funded partly by debt, the discount rate matters twice: it sets the hurdle rate for the projects and the cost of the capital that pays for them.
The historical analog is instructive. In June and July 2026, South Korea's KOSPI fell as much as 44 percent from its record intraday high over 40 calendar days, erasing an estimated $2 trillion to $2.18 trillion in market value, driven almost entirely by Samsung Electronics and SK Hynix. The trigger was the same question investors are asking now: whether Big Tech's AI infrastructure spending would deliver short-term returns, and whether demand for high-bandwidth memory would hold. The Korean market was a concentrated, flow-heavy expression of the AI trade, and when the narrative turned, the exit was crowded.
Concentration is the second fragility. Technology stocks make up nearly half the market capitalization of U.S. equities, according to iShares' Fall 2026 Investment Directions. A pullback in AI is therefore not a sector rotation; it is a market event. The 13-to-1 flow ratio describes not just a sector but the marginal source of demand for the entire index.
The Counter-Argument: The Capex Cycle Has Not Broken
The bull case is substantial, and it rests on physical commitments rather than sentiment. Taiwan Semiconductor Manufacturing has announced a capital-spending plan of $52 billion to $56 billion for 2026, an increase of at least a quarter from 2025. The six largest U.S. hyperscalers are projected to invest more than $500 billion in AI infrastructure. The four largest hyperscalers are on track to spend about $650 billion in combined capital expenditure in 2026. Goldman Sachs Research projects global AI-related investment will reach $1 trillion by the end of 2026, with AI capital expenditure rising from 0.9 percent of global GDP this year to 1.3 percent in 2027 and 1.4 percent by 2028.
Three strategists surveyed over the weekend — Gary Tan of Allspring Global Investments in Singapore, Billy Leung of Global X Management in Sydney, and Charu Chanana of Saxo Markets in Singapore — each expected near-term pressure on chip stocks but saw the multi-year picture as intact. The reason is simple supply and demand: demand for chips, energy, and computing power continues to outstrip supply. An essay about pacing model development does not switch off data centers that are already being built.
Nor is the valuation argument one-sided. Nvidia, the bellwether, trades at roughly 22 times forward earnings, near the low end of the 18-to-25 range it has occupied through 2026, after a pullback from its May highs. A lower multiple is not a catalyst on its own, but it provides a cushion: the valuation compression that drove the post-earnings declines of 2024 and 2025 has already happened. The stock no longer needs multiple expansion to work; it needs the guidance to hold.
The bull case also has a supply-side logic that the scare overlooks. If frontier model development slows, the value of existing compute capacity rises, because scarcity is what determines pricing power in a capacity-constrained market. The companies that already own the chips, the data centers, and the power contracts would benefit from a slower pace of new supply. In that reading, the Amodei essay is not bearish for the hardware trade; it is a cartel-like argument for restraining capacity growth.
Cyclical Scare, Structural Buildout — and the Gap Between Them
This is the judgment the market has to make, and it is being asked to make both halves at once. The flow concentration is cyclical: it is positioning, it is reversible, and it has reversed before. The 13-to-1 ratio is a measure of how crowded the trade has become, and crowded trades do not correct gradually — they correct when the marginal buyer steps away.
The buildout is structural: it is concrete, power contracts, chip fabs, and multi-year capital-spending commitments that will not be undone by an essay. Demand for computing continues to outstrip supply, and the companies spending the money have the cash flows to keep spending. This half of the trade does not revert on its own.
The risk is the gap between the two. If flows stall before the capex cycle matures, the market will reprice AI earnings on a higher discount rate even though nothing has changed in the physical buildout. That is the second-order effect that the 13-to-1 ratio makes possible: the stock prices can fall further than the fundamentals justify, because the prices were set by flow momentum, not by discounted cash flows.
Goldman Sachs Research has estimated the present discounted value of potential AI-related capital revenues to U.S. companies at roughly $9 trillion. AI-related companies have gained approximately $27 trillion in market value since November 2022, up from about $19 trillion seven months earlier. The gap between those two numbers is the valuation risk in one line: the market has priced more than the baseline case delivers.
What to Watch: The Signal That Would Prove the Thesis Wrong
The near-term path depends on two things, only one of which is about AI. The Federal Reserve meets on September 16, and several strategists noted the rate decision may matter as much to chip stocks as the Amodei essay. Rising Treasury yields and oil prices near triple digits raise the discount rate on long-duration growth assets; a dovish surprise would cushion the AI complex, a hawkish one would compound the pressure. The Fed's decision is the swing factor because it sets the denominator for every long-duration cash-flow valuation in the market, AI or otherwise.
On the AI side, the falsifying signal is specific. If technology and AI-themed funds post three or more consecutive weeks of net redemptions, the flow regime has shifted and the 13-to-1 rally is over. If, instead, inflows continue through the scare, the dip is noise and the trade remains intact. A second signal would be a cut or delay in hyperscaler capital-spending guidance; that would move the argument from sentiment to fundamentals.
The base case is that the essay produces a near-term de-risking but not a capex reversal: chip stocks wobble, the leaders consolidate, and the flow ratio compresses without breaking. The downside case is that the Fed stays hawkish, yields rise, and the flow reflex turns negative, producing a Korean-style air pocket in the most concentrated names. The upside case is that the slowdown call is read as a supply constraint on frontier models, which would raise the value of existing compute and support the hardware suppliers.
Across time horizons, the picture splits. In the short term, sentiment and positioning dominate, and the 13-to-1 flow ratio makes the complex vulnerable to a reflexive unwind. Over the medium term, earnings and guidance will decide whether the capex cycle is justified. Over the long term, the structural question is whether AI productivity gains justify the $27 trillion in market value that has already been claimed. Those three horizons can point in opposite directions, and they are all being priced at once.
The 13-to-1 flow ratio was the engine of the AI rally. It can also be the first thing to fail. The market is about to find out whether the money that built the trade believes in the models enough to hold through the moment when the model builders themselves ask for patience.
Explore more exclusive insights at nextfin.ai.
