NextFin News - The Federal Reserve is set to conclude its September policy meeting on Wednesday with markets pricing more than a 90% chance of a quarter-point rate increase, even as the biggest debate in equities - whether the artificial-intelligence boom is a platform shift or the largest bubble in U.S. history - flares back into the open. The two stories are usually covered separately. They should not be: both are ultimately about the same thing, the price of money, and both are now pointing in the same direction at the same time.
On one side of the trade, the Fed's own chairman has said financial conditions "do not appear restrictive enough." On his standard for acting, he said:
"We must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed."
Policymakers "have work to do" if they cannot see inflation returning clearly to the 2% target. On the other side, the people who built the AI boom are warning about it: Meta's chief executive says a bubble collapse is "definitely a possibility," Anthropic's chief executive published a 3,800-word call to slow frontier development, and the investor who called both the dot-com crash and the 2008 crisis calls this the biggest investment bubble in American history. The S&P 500 has fallen in six of the past seven sessions into the meeting, the 10-year Treasury yield touched 5.04% - its highest level since 2007 - and oil climbed toward $108 a barrel after a drone strike shut a Saudi pipeline that carries close to one barrel in twenty used worldwide. Investors are being asked to underwrite a $725 billion AI capital-spending plan for 2026 while the central bank that sets their discount rate prepares to make borrowing more expensive.
The Setup: One Meeting, Two Repricings
The immediate question for the Federal Open Market Committee is mechanical. The benchmark federal funds rate now sits in a 3.50%-3.75% range, held there for a second straight meeting after the July decision, when three regional bank presidents - Beth Hammack of Cleveland, Neel Kashkari of Minneapolis, and Lorie Logan of Dallas - dissented in favor of a quarter-point increase. Since then, the pressure to act has only grown. The consumer price index rose 3.4% in August on a trailing twelve-month basis, matching what economists expected and doing little to ease the Fed's problem. On the Fed's preferred gauge, the personal consumption expenditures index, headline inflation was 3.7% year over year as of July and core PCE was 3.3% - still well above the 2% objective.
Rate futures have repriced accordingly. On the day of Fed Chairman Kevin Warsh's Jackson Hole speech, August 28, CME Group's FedWatch tool showed a 57.4% probability of a 25-basis-point hike at the September meeting. By August 31 that was 66%. By mid-September, after the August inflation print, it had crossed 90%, with the prediction-market platform Kalshi at roughly 83%. A move to 3.75%-4.00% would be the first increase in the current tightening cycle, and it would mark a stark reversal from the cut path that looked like a given earlier in the year.
The equity side of the story is just as crowded. Nvidia, the most valuable company in the world with a market capitalization above $5 trillion, has seen its forward earnings multiple compress to about 24x from near 40x a year earlier - not because earnings collapsed, but because the spending wave behind them grew faster than the stock. Hyperscalers - Amazon, Microsoft, Alphabet, and Meta - have lifted their combined 2026 AI capital-expenditure guidance to $725 billion, up 77% from 2025's $410 billion. That is a lot of money to justify, and the people being asked to justify it are starting to hedge.
Why the AI Debate Is Not Just Noise
The bubble argument has moved from the fringe to the people who profit most from the boom. Deutsche Bank analysts Adrian Cox and Stefan Abrudan called September "the summer AI turned ugly," pointing to a gap between the infrastructure being built and the revenue it can realistically earn. Their evidence is circumstantial but broad: web searches for AI run ten times higher than crypto ever did at its peak, and S&P 500 companies mentioned "AI" more than 3,300 times in a single earnings quarter. Morgan Stanley's Mike Shalett has put the concentration in even starker terms - the AI complex accounts for roughly 75% of the S&P 500's returns since the rally began, about 80% of its profits, and close to 90% of its capital spending. When one theme carries that much of the market, a change in the discount rate does not just trim a few valuations. It rewrites the whole model.
Mark Zuckerberg, whose company is committing as much as $145 billion a year to AI infrastructure, nonetheless told interviewers that a collapse is "definitely a possibility." His exact words:
"But I do think there's definitely a possibility, at least empirically, based on past large infrastructure buildouts and how they led to bubbles, that something like that would happen here."
He cited the railroad buildout of the nineteenth century and the dot-com boom as the templates. Sam Altman of OpenAI made the same point more compactly: "When bubbles happen, smart people get overexcited about a kernel of truth." AI is the kernel. The exuberance is the variable.
The skeptics are not arguing that AI is fake. They are arguing that the financing is circular. Infrastructure providers such as Microsoft and Nvidia have taken stakes in, or extended credit to, the very AI developers who are their biggest customers. The developers use that capital to buy more capacity from the same providers, and the resulting transactions are booked as revenue growth. If that loop closes on itself, the industry is not discovering demand so much as funding it. The risk, as one analysis of these arrangements put it, is that "circular" deals create skewed incentives that magnify losses if demand fails to match expectations.
Against that, the bull case is simple and powerful: utilization is real. Unlike the dark fiber of the dot-com era, today's GPUs are fully booked, and cloud revenue at the hyperscalers is growing at a pace not seen in years. Google Cloud's backlog of signed but undelivered contracts roughly doubled in a single quarter to about $462 billion. Amazon Web Services grew 28% in its latest quarter, its fastest rate in roughly four years. Jensen Huang has called this the largest infrastructure buildout in human history, and Goldman Sachs estimates AI infrastructure spending reaching $3 trillion to $4 trillion by 2030. The bears see a capex echo chamber. The bulls see a platform shift with a waiting list.
The Mechanism: Why These Two Stories Multiply Rather Than Add
The obvious read is additive: higher rates hurt tech stocks, and bubble fears hurt tech stocks, so tech stocks face two headwinds. That understates the interaction. The transmission runs through the discount rate, the hurdle rate for new capital, and the balance sheets funding the buildout - and it compounds.
First, the discount rate. A long-duration asset is valued by the present value of cash flows that arrive far in the future. Raise the risk-free rate from a 3.5% policy floor toward 4%, and the denominator of every AI valuation model rises at the same moment the numerator - those distant AI profits - becomes harder to defend. The multiple compression Nvidia has already endured, from 40x forward earnings to 24x on a still-soaring earnings base, is the market doing this arithmetic in real time.
Second, the hurdle rate. Hyperscalers are not spending out of idle cash. Meta's advertising engine generates roughly $55 billion a quarter; its AI machine is committing as much as $145 billion a year. That gap has to be financed, and the cost of financing just rose. Every project that cleared a 3.5% hurdle does not automatically clear a 4% hurdle. The marginal data center is the one that gets delayed, and delayed capex becomes slower revenue for the chip suppliers.
Third, the balance sheet. If the circular-financing critique has any force, the weakest link is the AI developer that depends on continued funding to keep buying chips. Tighter policy does not just raise the cost of that funding; it narrows the set of investors willing to provide it. That is why Zuckerberg singled out OpenAI and Anthropic when he noted that their ability to keep raising money depends on "all of these macroeconomic factors that are out of their control." A capex boom funded by cheap capital meets a central bank whose job is to make capital less cheap. The two forces are not independent.
There is also the energy channel, which is where the inflation story and the AI story physically collide. Training and running frontier models requires electricity, and electricity in 2026 runs on hydrocarbons at the margin. Oil climbed toward $108 a barrel after a drone strike from southern Iraq hit Saudi Arabia's East-West pipeline, which can move four to five million barrels a day - close to one barrel in twenty worldwide - and after attacks on tankers in the Strait of Hormuz raised the risk of a wider supply disruption. Energy prices feed the PCE index. The PCE index guides the Fed. The Fed sets the rate that discounts AI earnings. The loop is closed, and it runs through the gas pump.
The Cyclical Call: Overbuild, Not Regime Change
Is this a structural shift or a cyclical overbuild? The answer matters because it determines what happens after the meeting. My judgment: the AI infrastructure buildout is a cyclical wave layered on top of a structural technology shift, and the cyclical leg is the one that will hurt investors first. A structural claim requires evidence of a permanent change in the rules of the game. AI clears that bar - the technology is real, utilization is high, and the productivity gains are measurable. But a cyclical claim requires only three things: a short-term driver, a history of mean reversion, and a demonstrated pattern of overbuild. The AI capex cycle has all three.
History is unkind to infrastructure manias even when the underlying technology wins. The railroads transformed the nineteenth-century economy and still left most railroad investors ruined. The fiber-optic networks of the 1990s built the backbone of the modern internet and still produced a 78% decline in the Nasdaq. The pattern is consistent: the technology survives, the overbuilders do not, and the returns accrue to the users of the capacity rather than the owners of it. What is different this time is not the pattern but the speed. Hyperscaler capex guidance rose 77% in a single year. Cycles that compress that much tend to overshoot on the way down as well.
The monetary-policy side is a choice, not a regime change, and choices can be reversed. Warsh's Fed is reacting to inflation data that has run above target for five years, to financial conditions that are not restrictive, and to an energy shock that could push prices higher still. If core inflation prints soft for two or three months, the hike path pauses. If oil breaks the other way, it pauses faster. Policy is cyclical by design. The AI buildout is cyclical by history. The technology is structural. Investors who conflate the three will get the timing wrong even if they get the thesis right.
The Counter-Thesis: What the Bears Are Missing
The strongest case against this reading is that it mistakes a valuation reset for a fundamental break. The forward multiple on the market's central AI beneficiary has already fallen from 40x to 24x while earnings kept rising. That is not a bubble popping; it is a bubble deflating in slow motion while the business underneath keeps growing. Cloud growth at the hyperscalers is accelerating, not rolling over. The capex is not vanity - it is pre-selling capacity that customers are already waiting for. And a 25-basis-point move in the policy rate is small relative to the earnings growth being priced. By this account, the market has already done the work: the fear is in the multiple, the fundamentals are intact, and the hike is a non-event.
There is force in that argument, and it is backed by the most important fact in the whole debate: AI is not the dot-com era's dark fiber. The GPUs are being used. But the counter-thesis leans on one assumption that the next three months will test - that capex growth can keep outrunning the cost of capital indefinitely. It cannot. The falsifying signal for my own view is specific: if core PCE prints below 0.2% month over month for two consecutive months while hyperscaler capex guidance holds and AI-driven cloud revenue growth stays above 25% year over year, then the double squeeze I am describing is not happening, and the "valuation reset, fundamentals intact" read wins. Until then, the burden of proof sits with the investors writing $725 billion checks as the price of money rises.
What Comes Next: Three Horizons, Three Scenarios
In the short term - the meeting itself and the days after - the market reaction depends less on the 25 basis points than on the language around them. A hike accompanied by a statement that financial conditions remain insufficiently restrictive would push the 10-year yield back toward 5% and pressure the Nasdaq further. A hike paired with a pause in the rhetoric would be read as a one-and-done inflation insurance move and could spark a relief rally. The watch item is not the decision; it is the sentence that follows it.
Over the medium term, the earnings seasons of late 2026 and early 2027 will separate the two theses. The key metric is not AI revenue at the chipmakers - that number will look impressive either way. It is the return on the hyperscalers' capital: whether each dollar of capex produces more than a dollar of cloud and advertising revenue within a reasonable payback window. If the circular-financing critique is right, that return will disappoint. If the platform-shift thesis is right, it will hold.
Over the long term, the structural verdict is almost certainly that AI is real and transformative. The railroad analogy cuts both ways: railroad investors lost money, but the economy that rode the rails grew richer. The beneficiaries of a structural AI shift may not be the companies building the data centers. They may be the firms that use cheap intelligence to restructure their own cost base - and they will be buying that capacity after the overbuilders have been shaken out.
Three scenarios frame the path. The base case: a 25-basis-point hike, a pause while inflation data confirms or denies the need for more, and a 15%-25% digestion in the most extended AI names as capex growth normalizes. The upside case: inflation rolls over quickly, the Fed signals the September hike is the last for this cycle, and AI revenue growth validates the spending - in which case the multiple compression reverses as fast as it came. The downside case: oil stays above $100, core inflation prints hot again, the Fed hikes twice more, and the circular-financing critique lands - a combination that would test the 24x multiple from below rather than above.
The central bank sets the price of money. The market decides what that price is worth. For the AI buildout, those two judgments are about to collide, and the investors who survive will be the ones who can tell a cyclical overbuild from a structural shift - and a real return from a circular transaction.
Explore more exclusive insights at nextfin.ai.
