Goldman Sachs Reveals Sluggish AI Adoption Among US Firms Despite Promising Productivity Gains
A Goldman Sachs report found that despite massive AI investment, actual adoption among US firms remains sluggish. Only 5.4% of businesses reported using AI in production, raising questions about whether the trillion-dollar capex cycle is justified by near-term returns.
Goldman Sachs published a report on December 16 that complicates the dominant narrative of an AI revolution sweeping through American business. The investment bank found that only 5.4% of US firms reported using artificial intelligence in the production of goods and services, a figure essentially unchanged from the previous quarter. Despite an estimated $200 billion in AI-related capital expenditure in 2024, the technology’s penetration into the broader economy remains, by Goldman’s own admission, surprisingly shallow.
The report is the latest in a series from the bank that has taken a more skeptical view of AI’s near-term economic impact than the technology industry itself promotes. In June, Goldman’s head of global equity research, Jim Covello, published a widely circulated note arguing that the AI infrastructure buildout “has not yet found its killer application” and warning that the $1 trillion expected to flow into data centers, chips, and power generation over the coming years would be difficult to justify without measurable productivity gains.
The 5.4% figure deserves context. It comes from the Census Bureau’s Business Trends and Outlook Survey, which asks firms whether they use AI in production. It does not capture AI used for internal tools, code generation, or back-office automation—uses that are widespread in the technology sector but less visible in official statistics. Even accounting for those omissions, the number suggests a gap between the boardroom rhetoric and the factory floor reality.
Adoption rates are highest in information technology (18%), professional services (12%), and finance (10%). Manufacturing and construction are both below 4%.
The finding matters because it calls into question the underlying assumption of the AI capex cycle. Nvidia, Microsoft, Amazon, and Google are spending as if every company in America will need AI compute within the next three years. If actual adoption follows the current trajectory, a significant portion of that infrastructure will be underutilized. Goldman’s own equity research team has begun describing the situation as a “field of dreams” problem: the infrastructure is being built, but it is not yet clear that the customers will come.
The counterargument, which Goldman’s technology bankers are simultaneously pitching to clients, is that adoption historically follows an S-curve. The 5.4% measured today represents early adopters; the inflection point comes when the technology reaches roughly 15% penetration, which could happen within two years if the current pace of enterprise AI product development continues. Both narratives are plausible. The only thing the data makes clear is that, as of December 2024, the bet has not yet paid off.
