Everything About AI
NextFin News - Silicon Valley is promising a post-scarcity future in which artificial intelligence makes goods and services so cheap that money itself becomes less relevant. The bond market is pricing the opposite. Between 2023 and 2025, long-dated government bonds, inflation-linked debt and corporate credit all rose in price - yields fell - after every major AI breakthrough, a reaction that persisted rather than fading. If AI were about to unleash the growth surge the optimists describe, yields should have climbed. They did not. That gap between the abundance narrative and the price of risk is the most important signal in markets today.
The disconnect has a dollar figure. The four largest cloud builders - Amazon, Microsoft, Alphabet and Meta - plan to spend roughly $725 billion on capital expenditure in 2026, up about 77% from the roughly $410 billion spent in 2025, with the overwhelming majority going into AI infrastructure. Yet the revenue that infrastructure can plausibly generate this year sits in the tens of billions, not the hundreds of billions required to justify the spend. The delusion is not that AI is transformative. It is that abundance arrives before scarcity has been paid for.
The Capex Curve Is Running Away From the Revenue Curve
The scale of the buildout is no longer a matter of analyst speculation; it is company guidance disclosed in earnings calls and filings. Amazon is committing about $200 billion in 2026, Microsoft about $190 billion, Alphabet between $175 billion and $205 billion after raising its ceiling at its second-quarter earnings, and Meta between $125 billion and $145 billion. Combined, that is roughly three-quarters of a trillion dollars in a single year, nearly double what the group spent in 2025, which was itself nearly double 2024. Evercore and Bank of America now model combined 2027 capital spending above $1 trillion.
Global AI investment is forecast to exceed $1 trillion in 2026, according to research that adjusts the narrower hyperscaler figures to include the wider ecosystem. That puts AI capital expenditure at about 1.8% of US GDP this year, rising to 2.5% in 2027 and 2.8% in 2028 - levels the research notes are consistent with the 2% to 5% of GDP peak investment impulses seen in prior general-purpose technology buildouts. Cumulative spending across compute, data centers and power is projected at roughly $7.6 trillion between 2026 and 2031.
Against that stands the monetization math. JPMorgan estimates the industry needs to generate about $650 billion in new annual revenue just to earn a 10% return on the infrastructure being built. Current AI-attributable revenue, even under generous assumptions that credit all incremental cloud growth to AI, sits between $50 billion and $150 billion a year. That is a gap of between four and thirteen times. Bain & Company's projection is starker: about $2 trillion in annual AI revenue would be required by 2030.
The visible leaders are growing fast, but from a small base. Microsoft said its AI business passed a $37 billion annual revenue run rate, up 123% year over year. Google Cloud grew 63% in the first quarter of 2026, Azure 40% and AWS 28%. The three major cloud providers together captured only about $20 billion of AI-related revenue in 2025, driven primarily by model hosting and inference, according to an industry tracker. Growth of 100% on a small base is not the same as a business that can service a trillion-dollar capital stock.
"The good news for the capex outlook is that all leading indicators rank near the top end of their range since 2022. This pattern suggests a robust near-term growth outlook."
That is the investment cycle speaking, not the return profile. The mechanism behind the gap is straightforward: hyperscalers are buying the same finite inputs - GPUs, high-bandwidth memory, data-center capacity, power - and bidding up their prices. Microsoft's chief financial officer, Amy Hood, told investors that rising prices for memory chips and other components accounted for $25 billion of the company's 2026 budget. The industry is therefore paying more for capacity that does not yet produce revenue, compressing the return on each dollar before a single line of AI revenue has been written.
The Bond Market's Quiet Verdict on Abundance
If the abundance thesis were correct, the macro signature would be unmistakable: explosive productivity growth, rising potential output and, eventually, higher real yields as capital demand outstrips saving. Bond markets are refusing to price that future. Economists at the Massachusetts Institute of Technology have said they were "very, very surprised" by the reaction: between 2023 and 2025, bonds rallied on every AI breakthrough, and the effect was lasting rather than transitory.
The interpretation matters more than the observation. One reading, advanced around the World Economic Forum, is that AI will deepen inequality rather than deliver broad abundance. Productivity and profitability will concentrate in a narrow set of AI leaders - Amazon, Alphabet, Meta and, to a lesser extent, Microsoft - while a significant remainder of the economy faces layoffs and weaker consumption. The result would be "bad disinflation" from damaged demand outweighing the "good deflation" from falling unit costs. The net effect is disinflationary, but not in the extreme form envisioned by the post-scarcity camp.
That reading is a direct challenge to the most extreme version of the abundance narrative. The chief executive of Tesla and SpaceX has predicted that AI and robotics will create an economic explosion "truly beyond all precedent," and has argued that money will stop mattering by 2036 because robots and AI will provide more goods and services than any person could consume. Bond traders, who have no incentive to be bearish or bullish and simply price duration risk, are effectively betting against that outcome.
The debt that is funding the buildout makes the bond signal harder to dismiss. AI-related investment-grade issuance is centered on about $300 billion for 2026, with as much as $360 billion of new 10-year-equivalent duration supply hitting markets this year alone. Debt financing tied to the AI build-out is forecast at about $4.1 trillion through 2030. Those liabilities are fixed in nominal terms. If the abundance future fails to materialize and "bad disinflation" takes hold, the real burden of that debt rises while the revenues meant to service it disappoint.
The Productivity Evidence: Solow's Paradox Returns
The strongest empirical test of the abundance claim is whether AI is measurably raising productivity today. The answer so far is ambiguous at best. A National Bureau of Economic Research study published in February 2026, surveying nearly 6,000 chief executives, chief financial officers and senior managers across the United States, Britain, Germany and Australia, found that 89% of respondents saw no measurable impact from AI on their firm's productivity - defined as sales per employee - over the previous three years. More than 90% reported no impact on employment. Among the minority that did report effects, the average productivity boost was a mere 0.29%.
The same executives predicted AI would lift their firm's productivity by only 1.4% and output by 0.8% over the next three years, while reducing employment by 0.7%. Those are the expectations of the people closest to implementation, and they are far removed from the transformation being priced into infrastructure spending.
There is a counter-reading. Some economists point to US labor productivity growth of about 2.7% in 2025, nearly double the 1.4% annual average of the prior decade, and frame it through a "productivity J-curve": investment in a general-purpose technology runs ahead of realized value during the installation phase, with the payoff arriving only after complementary reorganization. Adoption is undeniably spreading - the share of businesses using AI rose from 61% in early 2025 to 71% by early 2026. The World Economic Forum's chief economists, surveyed in January 2026, expect meaningful productivity gains within three to five years, not the immediate revolution that technology vendors have marketed.
The J-curve defense is coherent, but it is a defense of patience, not of abundance. It says the payoff is coming later - which is precisely an admission that it is not here now. The capital being deployed today is being priced as if the payoff were already visible.
The Strongest Case for the Bulls - and What Would Prove It Wrong
The bull case is not without merit. General-purpose technologies - the railway, electricity, the internet - have all required decades of overbuilding before their economic value became measurable. The railway mania of the 1840s left Britain with far more track than traffic justified, yet the railway ultimately transformed the economy. Nvidia's results illustrate the demand side: fourth-quarter fiscal 2026 revenue of $68.13 billion, up 73% year over year, with data-center revenue of $62.31 billion and gross margin of 71%. The chief executive described the moment as "the agentic AI inflection point." On that evidence, the infrastructure buyers are not bluffing; they genuinely need the capacity.
The bull case also rests on circular but real demand: the hyperscalers are each other's customers, and their capital budgets are the revenue of the semiconductor supply chain. As long as balance sheets hold and capital markets stay open, the cycle can extend further than fundamentals alone would justify. Alphabet's stock rose 5% after Google Cloud's 63% growth, validating the full-stack thesis that owning the model, the cloud and the silicon captures the margin.
But the circularity is also the fragility. The revenue that justifies the next round of capex is, in part, the capex of the customer. If enterprise adoption does not convert into external demand - if the roughly $20 billion of 2025 AI cloud revenue does not scale toward the $650 billion required for a 10% return - the cycle terminates in a balance-sheet recession for the builders, not in abundance for everyone else.
The falsifying signal is concrete. The delusion thesis is wrong if hyperscaler AI-attributable revenue reaches roughly $650 billion a year by 2028 - the level JPMorgan's model requires for a 10% return - or if measured labor productivity sustains growth above 3% a year for three consecutive years alongside rising real wages for the median worker. Either outcome would mean the revenue curve has caught the capex curve, and the bond market is wrong.
Who Benefits, Who Is Exposed, and What to Watch
The implications split cleanly by time horizon. In the short term, the cycle still has momentum: leading indicators for AI capital expenditure rank near the top of their range, and the suppliers of the buildout - semiconductors, memory, power equipment, data-center real estate - continue to see order books that justify further spending. The risk is a rotation, not an immediate collapse. Meta's shares fell about 10% in a single session after the company raised its 2026 capital-expenditure guidance, the first visible investor rebellion against the spending curve. Analysis of company filings showed Amazon's free cash flow declined 95% on a trailing basis to about $1.2 billion, illustrating how quickly the buildout consumes cash even at a company with deep resources.
Over the medium term, the winners and losers diverge. The exposed are the holders of long-duration assets priced for abundance - growth equities whose valuations assume uninterrupted AI-driven earnings growth, and the credit that funds data-center construction at fixed rates. If "bad disinflation" from labor displacement outweighs "good deflation" from falling unit costs, the debt overhang becomes the transmission channel: fixed nominal liabilities meet weaker nominal revenues. The beneficiaries, paradoxically, are the very bonds that have been rallying - long-duration government debt acts as the hedge against the growth shortfall the abundance narrative denies.
In the long run, the structural question is whether AI is a true general-purpose technology that lifts the productivity frontier, or a concentrated profit engine that displaces labor faster than it creates demand. The evidence today supports the second description more than the first. An updated study from Stanford's Digital Economy Lab, published in August 2026 and analyzing millions of payroll records, found that employment for workers aged 22 to 25 in the most AI-exposed occupations has fallen 19% relative to more AI-resistant occupations since late 2022, while older workers in the same roles have held stable or grown. That is not economy-wide abundance; it is a redistribution of opportunity toward those who own the capital.
The base case is a cyclical capex correction within a still-valuable technology, not a return to the pre-AI world. The upside case is the J-curve realises: enterprise adoption converts, revenue compounds toward the hundreds of billions, and the bond market's caution looks like a missed opportunity. The downside case is that external demand fails to materialize, the circular capex loop breaks, and the $4.1 trillion of debt financing the buildout becomes a drag on credit markets rather than a bridge to abundance.
The central judgment is this: the abundance delusion is not a claim that AI lacks value. It is the claim that the value arrives on the timetable the market has priced, and that it is broad-based rather than concentrated. The bond market, the productivity data and the revenue gap all point the same way. Scarcity has not been abolished; it has merely been refinanced.
Explore more exclusive insights at nextfin.ai.
