更关键的是，GDP帐面可能低估AI的真实产出，因为同样的金额如今能买到更多AI算力与推理结果。Peterson Institute的Anton Korinek与Patrick McKelvey指出，传统品质调整难以跟上AI品质快速提升；他们估计，美国AI计算支出从2023年的370亿美元增至2025年的2,190亿美元，而在控制品质后，AI输出价格（推理）在同期下跌94%。在他们的上限估计下，若能正确反映AI价格下跌，2025年实质GDP可能高出4个百分点，但这一数字仍具高度争议，且AI「slop」究竟有多少生产性用途尚未明朗。
晶片产业方面，收入与成长明显在2024年后急升，但主要集中于少数环节。费城半导体指数成分公司的收入增长显示，Nvidia的GPU与Micron等记忆体厂商贡献了最强劲的上行；若将这两家公司剔除，Sox指数的整体增速更接近一般景气循环高点，而非全面性的超级循环。作者因此认为，当前更像是广泛晶片景气循环叠加GPU与记忆体子行业的集中式超级循环。
AI is increasingly visible in US real GDP, but the measured effect is still too small to overstate. Computer equipment and software investment rose sharply at the start of 2025 and now stands more than 1 percentage point above the long-run trend as a share of real GDP; yet first-quarter US real GDP grew only 1.6% annualized. Because many data-center components are imported, the investment surge will be offset by negative net-import effects, and personal consumption, while sometimes smaller in a single quarter, is still about 10 times larger than AI-related investment in the economy overall.
The bigger issue is that GDP may be missing much of AI’s real output because each dollar buys more AI every day. Anton Korinek and Patrick McKelvey argue that conventional hedonic adjustments cannot keep up with the pace of quality improvement. Using chip production, GPU rental rates, and inference prices, they estimate US AI compute spending rose from $37bn in 2023 to $219bn in 2025, while quality-adjusted inference prices fell 94% over the same period. In their upper-bound scenario, properly adjusting for lower AI prices could have made 2025 real GDP 4 percentage points higher, though the productivity value of AI-generated “slop” remains unresolved.
In chips, the pattern looks less like a broad mania than a normal cyclical peak plus a concentrated surge in a few subsectors. Philadelphia Semiconductor (“sox”) index revenue growth accelerated sharply from 2024, but the move is driven mainly by Nvidia’s GPU business and, more recently, Micron’s memory business. Excluding those two names leaves a much more ordinary cycle for most chip firms. The broader conclusion is that the industry is in a standard chip boom, but with a supercycle-like burst concentrated in GPUs and memory rather than across the whole sector.