NextFin News - The world's most influential economists are admitting they are "driving in the fog" on artificial intelligence, yet their forecasts are being used to justify a buildout that the Bank for International Settlements says is already outrunning the balance sheets of the companies funding it. The profession faces an uncomfortable question: in making themselves indispensable to the AI boom, have economists stopped being useful as its referees?
The Numbers Behind the Boom
The scale of the commitment is without precedent in corporate history. The five largest hyperscalers are on pace to spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026 combined, a sum the BIS said in its 2026 Annual Economic Report is already outpacing their earnings and free cash flow, forcing some to issue debt to cover the gap. Goldman Sachs' May 2026 "Tracking Trillions" report puts the annual tab at $765 billion this year, rising to $1.6 trillion by 2031, for a cumulative six-year buildout of roughly $7.6 trillion.
Individual guidance illustrates the intensity. Amazon has projected about $200 billion in capital expenditure for 2026, up from $125 billion in 2025; Alphabet guided to between $175 billion and $185 billion, up from $91 billion; Meta to between $115 billion and $135 billion, up from $72 billion; and Microsoft to roughly $120 billion. Compiled from company earnings calls, these four companies alone plan to invest up to $630 billion in 2026, about a 62 percent increase on the prior year.
Against that spending, the measured payoff remains thin. The BIS acknowledged that task-level studies consistently report large efficiency gains, often of 20 percent to 50 percent in time savings, but warned that "the scale and pace of the current AI investment boom, accompanied by expectations of large productivity payoffs, bear resemblance to these precedents" - earlier investment surges that "ended with an eventual reversal in investment, inducing economy-wide recessions."
The Profession's Own Confession
It is against this backdrop that more than 200 economists, including 16 Nobel laureates and the chief economists of OpenAI and Anthropic, released a statement in July 2026 titled "We Must Act Now." The document does not offer predictions. It issues a confession. "We are driving in the fog, and it is extraordinarily difficult to anticipate what will happen next," said Anton Korinek, a University of Virginia economics professor and one of the statement's organizers.
"We are driving in the fog, and it is extraordinarily difficult to anticipate what will happen next."
The statement, organized by economists Erik Brynjolfsson, Ajay Agrawal, Anton Korinek and Tom Cunningham, warns that AI may become radically more powerful over the next decade, that the economic transformation could exceed the Industrial Revolution while unfolding over a vastly shorter time frame, and that economists, policymakers and technology leaders must act now to build the institutions to steer the technology. Michael Spence, a Nobel laureate at NYU, called for an "all hands on deck" approach given the "high degree of uncertainty about how big and when the impact will arrive."
The admission carries a particular irony. A profession whose social license rests on its claim to forecast the consequences of policy and investment is publicly conceding that it cannot see what is coming - even as its members are more in demand than ever to produce exactly those forecasts.
The Forecast Divide: Economists Versus Technologists
The gap between what economists expect and what AI practitioners expect is the widest in modern memory. A compilation of forecasts by economist Tom Cunningham found that most economists expect AI to add 0.1 to 1.5 percentage points a year to economic growth, while most AI insiders expect 3 to 30 percent a year. Each side has notable exceptions - Brynjolfsson and Korinek are more optimistic than their peers; Andrej Karpathy is more skeptical than his - but the chasm is systematic.
The Forecasting Research Institute's survey of economists found median expectations for annual labor productivity growth of 2.0 percent by 2030 in the unconditional scenario, barely above the 1.94 percent baseline in 2025. AI experts are significantly more optimistic, particularly in the rapid-progress scenario. The survey authors noted an "apparent tension": economists assign a 61.4 percent probability to moderate or rapid AI progress by 2030, yet their unconditional GDP and total factor productivity forecasts do not substantially depart from recent baselines. Their written rationales point to time-lagged diffusion - historically, transformative technologies from electrification to personal computers took decades to show up in the productivity statistics.
At the skeptical extreme stands Daron Acemoglu, the MIT Nobel laureate, who estimates generative AI will deliver a "nontrivial but modest" boost: about 0.7 percent to total factor productivity and 1.1 percent to the level of GDP over the next decade. Researchers at the American Enterprise Institute set his estimate against their own baseline of a 15 percent increase in US labor productivity and GDP. The difference is not a rounding error; it is a disagreement about whether the current wave of spending will ever earn its keep.
Why Economists Have Become Useful to the Boom
Here lies the uncomfortable mechanism. A $7.6 trillion buildout cannot be sold to boards, bond markets and regulators on engineering enthusiasm alone. It requires the vocabulary of economics: net present value, productivity multipliers, total factor productivity, diffusion curves, equilibrium effects. Every forecast - whether Acemoglu's modest 1.1 percent or a consultancy's trillions in unlocked value - performs the same function: it converts an act of faith into a number that can be modeled, debated and approved.
Consultancy forecasts routinely put the economic value of AI in the trillions. Goldman Sachs Research has projected that generative AI could drive a 7 percent, or almost $7 trillion, increase in global GDP and lift productivity growth by 1.5 percentage points over a 10-year period. McKinsey has estimated that generative AI could add $2.6 trillion to $4.4 trillion annually to the global economy. These numbers have influenced everything from stock valuations to government policy.
The demand for economists inside technology companies has grown accordingly. Amazon now has more economists working full-time than even the largest academic economics department, according to Harvard Business School research, and companies including Microsoft, Netflix, Uber and Meta have built in-house economics teams. These economists work on pricing, testing, forecasting and policy - functions that directly shape investment decisions. The Washington Post reported in July 2026 that AI firms are actively poaching top academic economists even as public anxiety about job losses rises, with the companies arguing their research can provide answers.
The conflict is not necessarily one of conscious bias. Most economists in industry produce genuinely rigorous work. The problem is structural: the profession's most lucrative opportunities now lie on the side of the buildout, not on the side of the skepticism. A career spent asking whether the spending is justified is less rewarded than a career spent modeling how the spending will pay off.
The Track Record Problem
There is also the matter of history. Economists have a poor record of forecasting the economic impact of general-purpose technologies at the moment of their arrival. Robert Solow's famous 1987 observation - "You can see the computer age everywhere but in the productivity statistics" - captured the productivity paradox of the first information age. Gains arrived, but with a lag of a decade or more, and distributed very differently from what contemporaries expected.
Today's evidence is mixed in ways that should humble both sides. A record 337 S&P 500 companies cited "AI" on their first-quarter 2026 earnings calls, 68 percent of all calls in the period and the highest number in a decade of data, according to FactSet. Yet the aggregate data have not confirmed the promised transformation. The San Francisco Federal Reserve summarized the state of play: "Most macro-studies of productivity growth find limited evidence of a significant AI effect. Even firms that say it's useful find little evidence of transformative gains." San Francisco Fed President Mary Daly delivered the line in prepared remarks in February 2026.
Yet Erik Brynjolfsson, director of Stanford's Digital Economy Lab, pointed to the latest Bureau of Labor Statistics revisions as evidence that the "fog may finally be lifting." With payroll growth for 2025 revised down to just 181,000 jobs while fourth-quarter real GDP tracked up 3.7 percent, he argued his own analysis indicates US productivity jumped roughly 2.7 percent in 2025 - nearly double the 1.4 percent annual average of the past decade. The same data can support two opposite readings, which is precisely the economist's dilemma.
The Second-Order Danger: When Forecasts Become Fuel
The deeper risk is not that economists will be wrong. They often are, and the profession has mechanisms - peer review, replication, the slow verdict of data - for correcting itself. The risk is that their forecasts, whatever their accuracy, become inputs to the boom they are supposed to be judging.
Consider the transmission chain. A hyperscaler announces a $200 billion capex plan. Its chief economist produces a model showing the spending will earn a competitive return if AI adoption follows a particular diffusion curve. Bond rating agencies and lenders cite the model. Equity analysts incorporate it. Policymakers, told that AI investment is supporting GDP growth - AllianceBernstein estimates AI-related capex could contribute about 1.5 percentage points to US GDP growth in 2026-27 - have an incentive to avoid disrupting the cycle. The economist's forecast, designed to describe the world, has helped construct it.
This is the sense in which economists are making themselves "too useful." Usefulness to the boom and usefulness to the public are not the same thing. A profession that is genuinely useful to the public would be willing to say, as the "We Must Act Now" signatories effectively did, that the numbers cannot be known - and that decisions made under that uncertainty should carry a risk premium, not a seal of approval.
The Counter-Thesis: Usefulness Is the Point
The strongest case for the profession runs as follows. Economists are not selling out; they are doing their job under difficult conditions. Forecasting under deep uncertainty is not cowardice - it is intellectual honesty. The "We Must Act Now" statement is not a surrender but a call to build better measurement infrastructure, including tools like Brynjolfsson's Canaries Dashboard with ADP Research, which tracks 4.6 million workers across more than 730 occupations in near-real time.
Nor is the boom solely dependent on economist approval. The spending is being driven by a strategic race among a handful of firms that believe dominance in AI infrastructure is winner-take-most; they would build even if every economist on earth issued a negative forecast. In that view, economists are epiphenomenal to the boom, not its enablers, and blaming them confuses the map with the territory.
There is force in this. But it understates how much a $7.6 trillion commitment, much of it financed with debt that outruns free cash flow, depends on a continuous supply of credible narratives about future returns. When the chief economists of OpenAI and Anthropic - the very firms doing the spending - are among those signing a statement saying the field is flying blind, the line between independent assessor and interested party has become hard to locate.
What to Watch
Several signals will test whether the profession's usefulness has tipped into capture. First, the productivity data: if US labor productivity growth does not sustainably exceed 2.5 percent annually by 2028 while AI capex remains at record levels, the optimistic forecasts will have failed their first real test. Second, the debt markets: the BIS flagged that hyperscaler investment is outrunning earnings and free cash flow - a widening gap between capex and internally generated cash would signal that the boom is increasingly dependent on external financing rather than realized returns. Third, the forecasters themselves: watch whether economists embedded in AI firms continue to publish estimates that diverge sharply from academic counterparts, and whether those estimates are disclosed with their funding sources.
The falsifying signal for the "capture" thesis is specific: if independent academic forecasts and in-house tech-company forecasts converge on modest numbers while the buildout continues anyway, then economists are not driving the boom - they are merely describing it, and the criticism loses its force.
The Bottom Line
The AI boom is not waiting for economists to certify it. But it is using them - their models, their vocabulary, their authority - to make an unprecedented gamble legible to the rest of the economy. That creates a choice the profession cannot avoid. It can remain useful to the boom by supplying the forecasts that keep capital flowing, or it can remain useful to the public by saying, clearly and repeatedly, that the fog has not lifted and that the largest corporate investment surge in history is being undertaken without a map.
It cannot be both. And on the day more than 200 economists admitted they were driving blind, the claim that they are also indispensable navigators should be treated as the first thing to question.
Explore more exclusive insights at nextfin.ai.
