NextFin News - The artificial intelligence boom is hitting its first serious test of faith. After two years of near-limitless capital flowing into chips, data centers and AI startups, the market is now asking the question it largely avoided during the euphoria: where is the revenue? Anthropic told investors its third-quarter revenue rose 45% to $16.8 billion — impressive in isolation, but slower than the blistering pace of early 2026 and below some expectations. OpenAI's annualized revenue sits at roughly $50 billion, not the $70 billion some outlets reported, after investors mistakenly applied a rival's accounting method. And the four hyperscalers funding this build-out are on track to spend more than $650 billion this year while their combined free cash flow turns negative for the first time in years. The AI revolution is not over. But the free-money phase is.
The Boom Has Outrun Its Receipts
The numbers tell the story of an investment cycle that has moved faster than the businesses underneath it. Anthropic, the Claude maker preparing a blockbuster November IPO, disclosed to investors that third-quarter revenue climbed to $16.8 billion from $11.5 billion in the second quarter. Growth of 45% would be enviable in almost any industry. In AI, where expectations were set by triple-digit expansion, it registered as a slowdown — and the market reaction was immediate. The company said it was profitable on an adjusted basis for the second straight quarter, but the message investors absorbed was that the pace set earlier in the year is moderating.
The OpenAI episode was more damaging because it struck at the reliability of the boom's headline figures. Reports last month put OpenAI's annualized revenue near $70 billion. The reality, according to people familiar with the matter, is roughly $50 billion as of September. The discrepancy was not a deterioration in the business; it was an accounting error by investors who calculated OpenAI's revenue using Anthropic's methodology, which counts sales through cloud providers differently. A $20 billion swing on a methodology mix-up is exactly the kind of fog that makes investors stop extrapolating and start auditing.
Meanwhile, the bill for the build-out keeps arriving. Microsoft, Alphabet, Amazon and Meta — the four hyperscalers that buy most of the world's AI chips and build most of its data centers — are on track to spend upward of $650 billion on AI investments in 2026. Amazon alone plans about $200 billion in capital expenditures; Alphabet guided to $175 billion to $185 billion; Meta to $115 billion to $135 billion; Microsoft's run-rate points to roughly $145 billion. At the low end, that is about $635 billion, a 67% jump from the $381 billion the four spent in 2025. Bank of America analysts now expect aggregate free cash flow for the group to finish 2026 in negative territory — a first in years, and a reversal of the cash-printing model that justified their valuations.
The market has begun to price this tension. The seven largest technology companies — Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta and Broadcom — are each down more than 10% from their 52-week highs, and as a group have surrendered more than 13% since peaking in mid-May. Nvidia, the AI chipmaker whose shares have been the boom's ticker symbol, fell nearly 2% in Monday trading after a filing showed that Thiel Macro, the hedge fund of Palantir co-founder Peter Thiel, sold its stake in the chipmaker. Broader indexes wobbled with it. The S&P 500 dropped 1.11% to 7718.94 and the Nasdaq Composite lost 1.88% to 26692.87 in a session driven by hot inflation data and AI doubts. As of this writing, Nvidia trades at $217.23, down 1.84% on the day, with a market capitalization of $5.24 trillion.
The bond market is sending its own signal. Investors have been dumping data-center bonds, treating AI borrowers as if they are entering what market participants call a "Darwin moment" — a natural-selection event in which only the fittest balance sheets survive. Treasuries rallied as investors sought shelter from AI turbulence, the classic flight-to-quality move that accompanies a risk reassessment.
And the physical build-out is not even keeping pace with the financial commitments. A JPMorgan analysis found that more than 60% of data-center capacity planned for completion in 2027 is not yet under construction, with another 7% delayed. Supply-chain backlogs, permitting fights and power availability are the bottlenecks. In other words, the industry has raised more money than it can physically spend — and the money is starting to cost more.
The Revenue Gap Is a Valuation Problem, Not Just a Growth Problem
The central tension is not that AI revenue is failing to grow. It is growing quickly. Anthropic's 45% quarterly jump and OpenAI's $50 billion run-rate would make almost any software company the darling of Wall Street. The problem is the denominator. When four cloud providers commit $650 billion in a single year to infrastructure whose revenue base is still a fraction of that, the return math requires assumptions that leave little room for error.
This is where the OpenAI methodology correction matters more than the number itself. A $20 billion revision in the perceived revenue of the industry's most famous private company reveals how little hard data the market has been working with. For most of the boom, investors extrapolated from announcements, waitlists and demo days. Now, with Anthropic heading for an IPO and OpenAI in fundraising talks at a $1.4 trillion valuation, the numbers are being forced into the light. The first figures to arrive under scrutiny have come in below the whisper numbers.
The valuation consequence is direct. If OpenAI is worth a multiple of $50 billion rather than $70 billion in annualized revenue, then the private-market marks across the AI stack — and the public-market comparables that shadow them — were priced on a revenue base that did not exist. That does not mean the companies are worthless. It means the multiple contracted before the earnings caught up.
Free Cash Flow Has Turned Negative: The Funding Model Is Changing
For a decade, the hyperscalers funded their expansion out of the enormous free cash flow generated by cloud and advertising. That model is flipping. Bank of America expects aggregate hyperscaler free cash flow to finish 2026 negative. When the companies that were supposed to be the ATM for the AI build-out become net consumers of capital, the funding has to come from somewhere else.
It already is. Alphabet is pursuing an $85 billion equity raise. Meta has explored fresh equity sales to fund its AI ambitions. OpenAI is in talks to raise $30 billion or more at a $1.4 trillion valuation, including discussions with funds from the United Arab Emirates. Data-center borrowers are finding the bond market less forgiving, with credit spreads widening as investors demand a higher premium for AI-linked debt.
This is the second-order effect that the first wave of AI bulls underplayed. The story was never just "AI demand is strong." The story was "AI demand is strong enough to justify spending hundreds of billions ahead of revenue, funded by cash flow that no longer exists." Remove the cash-flow assumption and the entire financing chain tightens at once: equity gets diluted, debt gets more expensive, and projects with marginal returns get deferred.
The physical bottlenecks compound the financial ones. JPMorgan's finding that more than 60% of planned 2027 data-center capacity is not yet under construction means capital is being committed faster than it can be deployed. Committed-but-undeployed capex earns nothing. It sits as work-in-progress on the balance sheet while the cost of capital rises. That is the worst of both worlds: the spending hits the cash-flow statement, but the revenue-generating capacity does not hit the income statement on schedule.
Cyclical Air Pocket or Structural Break?
This is the call that determines whether the selloff is a buying opportunity or the start of something longer. The answer is both, and they must be separated rather than blended.
The cyclical leg is a valuation air pocket. AI adoption is real and irreversible — enterprises are deploying models, consumer products are being rewritten around them, and compute demand is structurally higher than it was in 2023. But the stocks were priced for a straight-line extrapolation of early-2026 growth rates, with funding costs near historic lows and revenue figures that turned out to be fuzzy. A correction that compresses multiples while revenue catches up is a cyclical event. It reverts when the numbers align.
The structural leg is the end of indiscriminate capital. The regime that funded any project with "AI" in the pitch deck is over. What follows is a Darwinian sorting — the phrase data-center bond investors are using is apt — in which projects with clear power access, chip supply and paying customers survive, and the rest get stranded. That is not a reversion; it is a permanent change in the cost and discipline of capital. Goldman Sachs raised its estimate for combined hyperscaler capex from fiscal 2025 through fiscal 2030 to $5.3 trillion, up from $4.5 trillion before first-quarter earnings. The spending is not disappearing. It is being repriced.
The distinction matters because it tells investors which parts of the trade are broken and which are not. Nvidia's chips are still scarce. Data-center capacity is still short. The companies with real revenue — Anthropic on an adjusted basis, the hyperscalers' cloud arms — are still growing. What has broken is the assumption that capital was free and that every dollar of capex would convert to revenue on a straight line.
The Counter-Thesis: This Is the Monetization Phase Beginning
The bull case is serious and deserves its due. Wedbush analyst Dan Ives has argued that the industry is entering its monetization phase over the next six to twelve months — the period when the infrastructure spend finally shows up as earnings. Anthropic is already adjusted-profitable. The hyperscalers have survived capex cycles before; their free cash flow has turned negative during build-out phases and recovered as capacity came online. AI stocks still outperformed the broader market over the past year, and the Nasdaq 100 spent much of 2026 near record highs.
The strongest version of the bull argument is historical: every infrastructure boom — railroads, fiber optics, the first cloud build-out — looked irrational at the peak of spending, and most of the value ended up captured by the companies that owned the infrastructure, not the startups that rented it. If that pattern holds, today's selloff is the moment to own the picks and shovels, not the moment to leave the mine.
The counter to the counter-thesis is equally concrete. The railroad and fiber analogies are cautionary as much as encouraging: the companies that overbuilt destroyed shareholder value for a decade, even though the technology changed the world. The difference this time, bulls say, is that hyperscalers are spending cash they earn rather than money they borrow. But that cash pile is now turning negative, and the financing is shifting to equity raises and debt at wider spreads — precisely the conditions that turned previous build-outs into value traps.
I think what you're going to see happen in second quarter earnings that's going to surprise a lot of people is you're going to see one or more of these hyperscalers announce a reduction of capex commitments.
That forecast, from Thomas Hayes, chair of Great Hill Capital, captures the pivot the market is now pricing. If the largest buyers of AI infrastructure begin to pull back, the sellers of AI infrastructure — chipmakers, equipment suppliers, data-center developers — face an earnings revision that has not yet fully landed.
The falsifying signal is specific. If hyperscaler capital expenditures decelerate for two consecutive quarters while AI revenue per dollar of capex fails to rise, the bubble thesis is confirmed. Conversely, if OpenAI and Anthropic list at or above their current private valuations with accelerating revenue growth, and hyperscaler free cash flow returns to positive within four quarters, this selloff will be remembered as the cyclical dip before the next leg higher.
What Comes Next
The AI boom has not ended. It has entered its accounting phase.
In the short term, expect volatility to remain elevated. The market is repricing the entire AI complex off a revenue base that is smaller and less certain than the one embedded in current valuations. Chip stocks, which led the boom, remain the most exposed to any downward revision in capex expectations. Nvidia's $5.24 trillion market capitalization prices in years of uninterrupted demand growth; even a pause in the order rate would compress that multiple.
In the medium term, the sorting begins. Companies with secured power, chip supply and contracted revenue will continue to raise capital — at a higher price. Those without will face the Darwin moment the bond market is already describing. The equity raises now being discussed by Alphabet, Meta and OpenAI are the first visible transfers of risk from public-market balance sheets to new investors.
In the long term, the structural call stands: AI compute demand is a regime shift, not a cycle. The companies that emerge from this tightening with intact balance sheets and real monetization will define the next decade of technology. But the path there requires a repricing first, and repricings are messy.
What to watch, in order: hyperscaler capex guidance for 2027; the pricing and demand for the OpenAI and Anthropic IPOs; data-center bond spreads; and the next two quarters of AI revenue per capex dollar. The first of those to break will tell investors whether this is a cyclical air pocket or the start of a longer winter.
The AI boom was always going to meet reality. It has. The question now is whether reality grows to meet the boom — or whether the boom shrinks to meet reality.
Explore more exclusive insights at nextfin.ai.
