Everything About AI
NextFin News - Thirty-three trillion dollars of stock-market gains now rest on the future of artificial intelligence, and the question investors can no longer dodge is whether AI will ever make enough money to justify them. The figure being put on the table for the value added to equities since the buildout began in late 2022 is $33 trillion, and an independent research series from Goldman Sachs points in the same direction: it estimated the gain at about $27 trillion as of mid-year, up from $19 trillion just seven months earlier. The market is effectively betting that AI will capture a far larger share of corporate profits than it does today - a bet that is winning in the short run but is running into a hard arithmetic problem.
The $33 Trillion Bet Meets the $9 Trillion Profit Pool
The scale of the wager is easiest to grasp in two numbers. AI-related companies have gained approximately $33 trillion in market value since the buildout began, while the same research team at Goldman Sachs estimates the present discounted value of the additional profits AI could create for the U.S. economy at roughly $9 trillion in its baseline case. The gap between the two is the story.
The rally is also the most concentrated in a generation. Of the 85.6% gain in the broad U.S. market index since May 2023, 22.76 percentage points came from semiconductor stocks alone - and 12.2 points of that came from Nvidia by itself. The semiconductor industry's market capitalization has grown more than fourfold, to $9.4 trillion in April 2026 from $2.2 trillion in May 2023. Nvidia, which became the first company to reach a $4 trillion valuation in July 2025 and passed $5 trillion in April 2026, has risen more than 14-fold since the end of 2022, adding more than $4.5 trillion in market value in roughly three years.
This is not a broad-based boom. Broadcom's accelerator revenue jumped 840.2% between the March 2023 and March 2026 quarters, and its accelerator sales are projected to reach $18.3 billion by the end of 2026, a 210% increase. But the software-application industry tells the opposite story: its market capitalization fell to $1.3 trillion in April 2026 from $1.8 trillion in October 2025, a 26% decline in six months. The AI trade has been a hardware trade, and a narrow one at that.
The fuel behind it is a capital-spending wave without a modern peer. Global AI investment is forecast to exceed $1 trillion in 2026, with U.S. spending at about $581 billion. Goldman Sachs Research projects AI capital expenditure will rise from 1.8% of U.S. GDP in 2026 to 2.5% in 2027 and 2.8% in 2028. Spending plans from the largest cloud and computing companies for 2026 are nearly 50% higher than they were roughly six months earlier. As Dominic Wilson and Vickie Chang of Goldman Sachs Research put it: "The AI investment boom is neither as broad-based nor as long-lived as the 1990s tech boom yet, but it is matching its scale."
Investors are being paid to believe that the money will come back. An analysis of hyperscaler filings puts the group's capital spending at between 31% and 83% of revenue, up from 10% to 15% in 2020 - ratios once seen only in utilities and telecoms. Amazon's 2026 capital-spending guidance of roughly $200 billion alone exceeds the combined annual capex of the entire publicly traded U.S. energy sector. Morgan Stanley and J.P. Morgan estimate the technology sector will need to issue roughly $1.5 trillion in new debt over 2025 to 2027 to fund the buildout. And industry analysis anchored to Bain and Company's framework suggests that sustaining the current investment trajectory requires roughly $500 billion of annual spending to generate about $2 trillion of revenue - a four-times revenue multiple on capital that has not yet been demonstrated at scale.
The tension, then, is precise: a $33 trillion repricing of equities, a $9 trillion baseline estimate of the profit pool AI can create, and a spending wave that is doubling faster than the revenue to pay for it. The rest of this piece asks whether that gap can close without a painful reset.
How a Capital-Spending Wave Becomes a Stock-Market Wave
The transmission channel is mechanical, and it is working exactly as a capex supercycle should. When a handful of giant buyers commit to spending hundreds of billions on chips, servers, data centers, and networking gear, the revenue shows up first at the suppliers. Nvidia's graphics processors, Broadcom's custom accelerators, memory makers, and the contractors building the data centers book the orders immediately. Those orders become earnings beats, the earnings beats become raised guidance, and the raised guidance becomes multiple expansion - the market pays more for each dollar of profit because it expects the growth to persist.
That is why the semiconductor industry's market value has quadrupled while the software-application industry has shrunk by more than a quarter. The money is physically flowing through the hardware layer. Broadcom's 840% accelerator-revenue growth over three years is not a sentiment story; it is an order-book story. Nvidia's $4.5 trillion of added market value is the capitalized value of a multi-year backlog that buyers have already committed to.
But a capex cycle has a built-in asymmetry that investors often miss until it is too late. The supplier's revenue is the buyer's cost. Every dollar of accelerator revenue for Nvidia is a dollar of capital expenditure for Microsoft, Meta, Alphabet, Amazon, or the neoclouds - and those buyers are spending it in the hope of selling AI services at a profit later. The rally prices the supplier's certain revenue as if the buyer's uncertain future profit were equally certain. That is the first crack in the chain: the market has capitalized the buildout as though the payoff were already proven.
Cyclical Wave or Structural Shift - Deciding Which One This Is
This is the judgment the whole thesis turns on, and the honest answer is that both forces are at work - but they operate on different time horizons, and confusing them is how investors lose money.
The structural case is genuinely strong, and it is stronger than the dot-com comparison suggests. Goldman Sachs Research identified four conditions that signaled a bubble in the late 1990s: eroding profit margins, surging wage growth and unit labor costs, a widening current-account deficit, and valuations running far ahead of earnings. As of late 2025, three of those four were absent from the AI boom. Corporate profit margins have climbed rather than eroded. Wage growth and unit labor costs are rising more slowly than they did in the 1990s. The U.S. current-account deficit has been shrinking, not growing. And, critically, earnings and earnings expectations have risen rapidly alongside prices.
That is a meaningful distinction. In 2000, Cisco's record market value rested on profits that never arrived. Today, Nvidia's $5 trillion valuation rests on profits that are arriving on schedule and then some. Artificial intelligence is also a general-purpose technology in the way the internet was: it reshapes productivity across the whole economy, not just inside the technology sector. The 2026 outlook from J.P. Morgan states the bull case plainly: the firm sees no bubble about to burst and expects the technology to disrupt labor markets and boost productivity globally.
The cyclical case, however, is equally real - and it concerns the financing leg of the boom, not the technology. Capital expenditure as a share of revenue cannot stay at 31% to 83% forever. No industry in history has sustained utility-like capex intensity indefinitely without either a sharp drop in spending or a collapse in returns. The estimate that the technology sector will need to issue roughly $1.5 trillion in new debt over the next three years is the tell: when the buyers start financing their spending with bonds rather than free cash flow, the cycle has moved from self-funding to credit-dependent. Credit-dependent cycles revert.
So the cleanest way to separate the two: the technology is structural, the capex supercycle is cyclical. AI will still be transformative in 2030. But the period in which hyperscalers double their spending every year, finance it with debt, and reward suppliers with 800% revenue growth is a wave, and waves break. The market's $33 trillion gain has priced elements of both as if they were one permanent regime.
The Expectation Gap: $33 Trillion Priced Against $9 Trillion of Profits
Here is the second-order question that most of the commentary skips. Even if AI is a genuine revolution, and even if the hyperscalers eventually monetize it, the stock market has moved ahead of the most credible estimate of what that monetization is worth.
Goldman Sachs Research's baseline estimate of the present discounted value of potential AI-related capital revenues to U.S. companies is about $9 trillion. AI-related companies have gained approximately $27 trillion in market value since November 2022 - and the $33 trillion figure implies the gap has widened further. The researchers are explicit about what must happen to close it: closing the gap "needs increasingly optimistic assumptions."
The most credible stories for substantial stock gains depend on the assumption that AI companies earn a bigger slice of total profits than the average company - in other words, that AI suppliers and model owners capture a disproportionate share of the economy's profit pool, not just a slice of a growing pie. And the market is not merely assuming this will happen; it is assuming it will persist. "The underlying assumption that the market is making is that this process will continue," Wilson and Chang write. That single sentence captures the entire risk. If AI's profit share rises and then stabilizes, today's prices are defensible. If it rises and then mean-reverts as competition commoditizes the models - as the software-application industry's 26% decline already hints it might - the $33 trillion contains a valuation layer that has no earnings underneath it.
This is where the concentration data bites. Technology stocks' share of the U.S. stock market reached 37.5% as of May 31, 2026, according to index data that excludes Alphabet and Meta, classified as communication services, and Amazon, classified as consumer cyclical - all of which are heavily invested in AI. That concentration surpasses levels seen during the late-1990s internet bubble. A market that depends on a handful of names to deliver a profit-share revolution is fragile in a way a broad market is not. One miss, one guidance cut, one delay in the AI revenue ramp, and the index has nowhere to hide.
There is also a geographic reordering underway that most investors have not priced. The AI supply chain has lifted the economies at its center: Taiwan has overtaken Canada to become the world's sixth-largest stock market, while South Korea has leapfrogged the United Kingdom into eighth place, according to HSBC data tracking global equity-market capitalization rankings. The boom is redrawing the hierarchy of global markets, not just the composition of the S&P 500.
The Strongest Case Against This View - and What Would Prove It Wrong
The bull case is not a strawman, and it deserves its weight. The counter-thesis runs like this: the monetization is already arriving, just not where the skeptics are looking. AI revenue is ramping inside the hyperscalers' cloud divisions, in enterprise software priced per seat, in advertising systems optimized by models, and in the productivity gains that do not show up as AI revenue at all but do show up as higher margins. The finding that recent gains have been more earnings-driven than valuation-driven supports this. J.P. Morgan's 2026 outlook sees no bubble. And history offers a precedent for a long lag between infrastructure buildout and payoff: the railroad and internet buildouts both required a decade of overinvestment before the productivity surge became visible in the data.
There is force in this. Declaring a bubble every time investment runs ahead of near-term revenue has been a losing trade for three years, and the earnings data so far side with the bulls.
But the counter-thesis has one vulnerability that can be tested. The sustainability math is unforgiving: Bain's framework implies that roughly $500 billion of annual AI cloud investment needs about $2 trillion of annual revenue - a four-times revenue multiple - to be self-sustaining. If AI-related revenue growth does not accelerate to exceed capital-spending growth within the next four quarters, and if the hyperscalers' capex-to-revenue ratio does not compress from the current 31%-83% range back toward the roughly 25% capex-intensity threshold that the framework treats as sustainable, then the structural-upside thesis is wrong. The $33 trillion repricing will have outrun the cash AI can generate, and the reset will come through multiple compression in the suppliers, not through a failure of the technology itself.
That is the falsifying signal. Watch the quarterly capital-spending disclosures of the five largest hyperscalers against their AI-attributable revenue growth. If spending keeps compounding faster than the revenue for two consecutive quarters, the wave is still cresting - and the distance to shore is shorter than the chart suggests.
Who Wins, Who Is Exposed, and the Three Paths Ahead
Cashing in the mechanism rather than retelling it, the impact splits cleanly by position in the chain. The beneficiaries of the next leg are the suppliers with contracted backlogs - the chip designers, the memory makers, the power and cooling infrastructure providers - because their revenue is already committed. The exposed are the equity holders of the hyperscalers if the capex-to-revenue ratio fails to compress, and the broader index because of its heavy technology weight. The debt markets are the silent stakeholder: roughly $1.5 trillion of new technology issuance over 2025-2027 means a credit event, not an earnings event, could be the trigger that forces a reassessment.
The forward look should be split by horizon, because the same facts point in opposite directions depending on the timeframe:
- Short term (sentiment and liquidity, 0-6 months): The trend is still up. Earnings are rising, guidance is strong, and the hyperscalers remain supply-constrained rather than demand-constrained. Momentum favors the hardware layer.
- Medium term (fundamentals, 6-24 months): This is the danger zone. The capex-to-revenue ratio must begin compressing, or the debt-funded buildout narrative will attract a risk premium. This is where the falsifying signal lives.
- Long term (structural, 2-10 years): AI remains a genuine general-purpose technology. Even a painful cyclical reset in the suppliers would not invalidate the productivity transformation - it would just redistribute who captures the value, likely away from the infrastructure layer and toward the application layer that has underperformed so far.
Three scenarios frame the range:
- Base case: AI revenue accelerates but not fast enough to fully close the $33 trillion-versus-$9 trillion gap immediately. The market digests the mismatch through a 15%-25% correction in the semiconductor complex while the broader index grinds higher on non-AI earnings. Trigger: hyperscaler capex growth slows to match revenue growth within four quarters.
- Upside case: Enterprise AI adoption inflects sooner than expected, the capex-to-revenue ratio compresses smoothly, and the profit-share shift materializes. The $33 trillion proves conservative. Trigger: two consecutive quarters of AI revenue growth exceeding capex growth across the five largest hyperscalers.
- Downside case: Spending continues to outrun revenue, debt issuance stretches credit spreads, and a single large guidance cut from a hyperscaler forces a repricing. The semiconductor complex gives back half its gains. Trigger: the capex-to-revenue ratio fails to compress for two consecutive quarters while credit spreads widen by more than 50 basis points.
All data in this article are as of September 21, 2026.
This is not a bet on whether AI is real. It is a bet on who captures the profit, and the market has already paid the suppliers as if the answer were settled. It is not.
Explore more exclusive insights at nextfin.ai.
