美国正将经济押注于“扩展”人工智能：当人们需要时，智能在哪里？（英）

Servaas Storm* Working Paper No. 244 December 1st, 2025 ABSTRACT The AI industry is betting that ‘scaling’,i.e., adding more and moredata, GPUs, computeinfrastructure and dollars, will lead to machine superintelligence or Artificial General Intelligence(AGI)—which in turn will lead to exponential growth of output, productivity and profits for theindustry and the larger American economy. Focusing on AGI and generic LLMs, the point of thispaper is plain: AI’s ‘scaling’ strategy must fail and the AI data-centerinvestment bubble will pop.The paper identifies four bottlenecks: (1) the planned $5 trillion investment in data centerinfrastructure (during 2026-2030) is not going to pay off; AI revenues will not increase enoughand AI inference cost continue to rise faster than revenues; (2) AI firms will have to resort tohyper- fast enough, because upstream suppliers—producing everything from copper wire to turbines totransformers and switchgear—will run into labor shortages, long waiting times for power gridconnections, material bottlenecks and regulatory blowback; and (4)the strategic bet of frontier AIfirms that AGI can be achieved by building ever more data centers and using ever more chips isalready going bad; AI products will continue to be untrustworthy for high-stake usage. As a result,the magical projections ofexponential growth, which defy economic and financial logic and fatally https://doi.org/10.36687/inetwp244 Keywords: Artificial intelligence; AGI; AI bubble; LLMs; circular financing; revenues; price-earnings ratio; leverage; scaling; inference cost; hallucinations; Chinese competition; AI- “If there must be madness, something may be saidofhaving iton a heroic scale” “The propensities to swindle and to be swindled run parallel tothe propensity to speculate during a boom.”―Robert Z. Aliber, Charles P. Kindleberger & Robert N. This paper reviewsfour major factors that will (rathersooner than later)popthe irrationalenthusiasm around the Artificial Intelligence (AI) infrastructure spending spree in the U.S.1 The AI industry is betting the U.S. economy on its strategy of ‘scaling’,i.e., adding more andmore data, GPUs, compute infrastructure and dollars, in the belief that this will lead to machine superintelligenceor Artificial General Intelligence (AGI)—and, of course, that thecapital expenditures can be recouped, with a good rate of return, because the superintelligent My main point is simple: the ‘scaling’ strategyto achieve AGIis already showing significantdiminishing returns, because more data and more GPUs cannot fundamentally improve theperformance of LLMs, after a point. The algorithms are not constructed on proper and robustworld models, but instead are built to autocomplete, based on sophisticated pattern-matching.As a result,AGI will remain a dream andthe magical projections of exponential growth, However, the generic LLMs are different fromspecialized,domain-specific AI toolsthatarealready being usedto great effectin many scientific disciplines, such as protein science,3code generation,and pharmaceuticalresearch.Targeted machine-learning systems can manage loadon the electricity grid more effectivelyandhelp reducecarbonemissions in trucking,shipping, steelmaking and mining industries.Each use case should bescrutinizedfor itsutility, thebiases and risks it might introduce, and itspropensity to destroy jobs that depend The fact that the AI industry is not only theprincipalsource of growth in an otherwisesclerotic U.S. economy (Storm 2025), but is also driven by a concentrated set of hyper-scalersengaging in ‘circular’ transactionsbased onaggressively optimistic long-term cash flow-generating potential should be cause for worry (seeArun 2025). We now laugh at the utter foolishness of some of the richest merchants in Amsterdam who at the height of Tulip Mania To be clear, cracks are showing and worries of an AI bubble are intensifying (e.g., Schmidtand Xu 2025; Thornhill 2025), and rapidly so. In recent weeks, several hedge fundstrimmedtheir stakesin some of the largest seven tech firms (the so-called ‘Magnificent Seven’) (Sen 2025), especially afterwarnings from the CEOsof Goldman Sachs6and Morgan Stanley(Saini and Nishant 2025).More than 50 per cent of the fund managerssurveyed by Bank ofAmerica (during November 2025), who between them manage around $500 billion infinancial assets, said that AI stocks were already in a bubble; a net 20 per cent of these fundmanagers stated that corporations were spending too much ontheir datacentercenter The AI bubble will eventually pop because: 1.There is no world in which the enormous spending in datacentercenterinfrastructure(more than $5 trillion in the next five years) is going to pay off; the AI-revenue projectionsare pie-in-the-sky, as customers are unlikely to pay (enough) for the rather modest servicesoffered by the LLMs and given the eventual oversupply of LLM services.At the same 2.There is no way in which the AI industry can fund