The bank’s latest research note shows that AI-linked annualised revenue has ballooned by more than $340 billion since 2022 and is on track to swell by nearly $580 billion by the end of 2025. Yet, AI-related real investment recorded in national accounts has only climbed by $42 billion.
This means that most of the cash flowing into AI does not show up in official U.S. economic growth figures.
Goldman’s analysis essentially blames surging semiconductor prices and cross-border sales—not factors that prop up domestic economic output .
The investment bank found that higher profit margins (some $30 billion) and foreign revenue (another $130 billion) explain roughly half of this apparent mismatch.
Meanwhile, the Bureau of Economic Analysis excludes semiconductors and cloud computing from official GDP tallies because they are considered “intermediate inputs.”
Goldman argues that this methodology could understate the real GDP bump from AI-related spending by up to $100 billion, given that semiconductor supply has risen significantly while cloud services remain out of the GDP spotlight.
Looking ahead, Goldman expects AI investment to branch into data centres, servers, networking hardware, and utilities—categories likely to be counted toward real US GDP.
That shift, they say, could deliver a mild economic lift in 2025 as AI’s tentacles spread beyond the often overlooked domains of microchips and cloud services.
However, the bank cautions that, without changes to how national accounts are calculated, the most enormous slices of AI-driven spending may remain largely invisible in official economic measurements.
Even after explaining away most of the discrepancy, there is still $50 billion in unaccounted-for AI investment, pointing to a gap between what Wall Street sees on corporate ledgers and what the government chalks up in its books.
The data highlights that AI’s grand promises of massive growth and productivity gains are still tangled up in technical classifications that leave much of the money out of the official GDP narrative.
