For the better part of two years, the artificial intelligence trade has been the most powerful force in American capital markets. Hundreds of billions of dollars have poured into data centers, semiconductor fabrication, and AI software startups. Nvidia alone has seen its market capitalization swell past $3 trillion. Yet behind the euphoria, two of Wall Street’s most influential banks have arrived at a striking conclusion: AI’s actual contribution to U.S. GDP growth may be, for all practical purposes, zero.
Goldman Sachs and Morgan Stanley, in separate analyses published in recent weeks, have each attempted to quantify the macroeconomic impact of artificial intelligence on the American economy in the near and medium term. Their findings, while differing in methodology, converge on a remarkably modest estimate — one that stands in sharp contrast to the breathless narratives emanating from Silicon Valley and the technology sector’s investor relations departments.
Goldman Sachs Pegs AI’s Growth Contribution at a Fraction of a Percent
Goldman Sachs economists, led by Jan Hatzius, have modeled the potential GDP uplift from AI adoption across U.S. industries. According to reporting by Slashdot, the bank’s analysis found that AI’s contribution to U.S. economic growth is currently negligible and is likely to remain so for several years. The Goldman team’s model suggests that even under optimistic adoption scenarios, AI might add only a small fraction of a percentage point to annual GDP growth over the next decade — a figure that rounds to essentially zero in the context of an economy that typically grows between 2% and 3% per year.
Morgan Stanley’s research unit reached a broadly similar conclusion through its own framework. The bank examined the rate at which businesses are actually deploying AI tools in ways that measurably improve productivity, and found that the gap between AI investment and AI-driven output remains enormous. Firms are spending heavily on the technology, but the productivity gains — the mechanism through which AI would translate into GDP growth — have yet to materialize at scale.
The Productivity Paradox Returns With a Vengeance
Economists will recognize this pattern. It echoes what Nobel laureate Robert Solow famously observed about computers in 1987: “You can see the computer age everywhere but in the productivity statistics.” The so-called Solow Paradox persisted for roughly a decade before the internet era finally produced measurable productivity gains in the late 1990s. The question now is whether AI will follow a similar trajectory — and if so, how long the lag will last.
The Goldman Sachs and Morgan Stanley analyses suggest that the current moment bears more resemblance to the early 1990s than to the late 1990s productivity boom. Businesses are experimenting with large language models and generative AI tools, but most deployments remain in pilot phases or are confined to narrow use cases such as customer service chatbots, code assistance, and document summarization. These applications, while useful, do not yet represent the kind of broad-based, economy-wide transformation that would register in national output statistics.
Capital Expenditure Is Soaring, but Returns Are Lagging
What makes the current situation particularly notable is the sheer scale of capital being deployed. Microsoft, Alphabet, Amazon, and Meta have collectively committed well over $200 billion in AI-related capital expenditure for 2025 alone. Nvidia continues to report record quarterly revenues. The construction of massive data centers has become one of the fastest-growing segments of the U.S. construction industry, straining electrical grids and driving up demand for natural gas.
Yet as both Goldman Sachs and Morgan Stanley point out, capital expenditure is not the same as economic output. The spending itself does contribute to GDP through the investment channel, but the return on that investment — in the form of higher productivity, new products, and expanded economic capacity — is what ultimately determines whether AI will be a lasting contributor to growth or an expensive detour. So far, the evidence for transformative returns is thin.
Survey Data Tells a Story of Cautious Adoption
Recent survey data supports the banks’ skepticism. The U.S. Census Bureau’s Business Trends and Outlook Survey has consistently shown that only a small minority of American businesses report using AI in their operations. Among those that do, the most common applications are modest in scope. A February 2025 survey by the Federal Reserve Bank of Atlanta found that while business leaders express enthusiasm about AI’s potential, most describe their organizations as being in early experimental stages rather than full deployment.
This pattern holds across industries. In financial services, where AI adoption is among the most advanced, banks and asset managers have deployed machine learning models for fraud detection and risk assessment for years. But generative AI — the technology driving the current investment boom — is being adopted more cautiously. Compliance concerns, hallucination risks, and the difficulty of integrating large language models into regulated workflows have all slowed deployment.
The Historical Record Offers Both Caution and Hope
Proponents of AI’s economic potential argue that the Goldman Sachs and Morgan Stanley analyses are measuring the wrong time horizon. Erik Brynjolfsson of Stanford University, one of the foremost scholars of technology and productivity, has argued that general-purpose technologies typically take 15 to 30 years to fully reshape an economy. Electricity, the internal combustion engine, and the personal computer all followed this pattern. By this logic, expecting AI to show up in GDP statistics in 2025 is akin to judging the automobile’s economic impact in 1910.
There is merit to this argument. The productivity gains from electricity did not fully materialize until factories were redesigned around electric motors rather than simply substituting them for steam engines. Similarly, the personal computer’s productivity impact was muted until the development of the internet, enterprise software, and organizational changes that allowed businesses to take full advantage of digital tools. AI may require analogous complementary innovations and institutional adaptations before its economic potential is realized.
Wall Street Faces an Uncomfortable Tension
The banks’ findings create an awkward dynamic for Wall Street itself. Goldman Sachs and Morgan Stanley are among the most active underwriters and advisers in the AI sector. Their investment banking divisions have profited handsomely from AI-related IPOs, secondary offerings, and M&A activity. Their wealth management arms have steered client capital into AI-themed funds and individual technology stocks. To publish research suggesting that AI’s economic impact is negligible — even as their transaction businesses benefit from AI enthusiasm — reflects either admirable intellectual honesty or an institutional disconnect, depending on one’s perspective.
The tension is not lost on market participants. Some investors have begun to question whether the AI trade has become disconnected from economic fundamentals. The massive run-up in technology stocks, particularly those associated with AI infrastructure, has been driven largely by expectations of future earnings growth. If those earnings depend on widespread AI adoption that the banks’ own economists say is not happening, the valuation case becomes considerably harder to defend.
What the Data Will Need to Show
For AI to begin registering in growth statistics, several things would need to happen simultaneously. First, adoption would need to move beyond experimentation and into core business processes at a significant number of firms. Second, the labor market would need to show signs of AI-driven productivity gains — either through higher output per worker or through the reallocation of labor from routine tasks to higher-value activities. Third, new products and services enabled by AI would need to generate measurable economic value.
None of these conditions are currently being met at scale, according to the data examined by both banks. Labor productivity growth in the United States has been running at roughly 1.5% to 2% annually in recent quarters — respectable by historical standards, but not materially different from pre-AI trends. There is no statistical evidence yet that AI is bending the productivity curve upward.
A Trillion-Dollar Bet Awaiting Validation
The implications of the Goldman Sachs and Morgan Stanley research extend well beyond academic economics. If AI’s contribution to growth remains near zero for the next several years, the current pace of capital expenditure by technology companies may prove unsustainable. Investors who have priced in rapid AI-driven earnings growth across the technology sector may face disappointment. And policymakers who have pointed to AI as a potential solution to slowing productivity growth and demographic headwinds may need to look elsewhere for answers.
None of this means that AI will never deliver on its economic promise. The technology is advancing rapidly, costs are falling, and the range of potential applications continues to expand. But the gap between potential and reality remains vast. Wall Street’s own analysts have now put numbers on that gap, and the numbers are sobering. The AI revolution, if it comes, has not yet begun to show up where it matters most: in the actual performance of the American economy. For now, the biggest beneficiaries of the AI boom remain the companies selling the tools — not the economy that is supposed to be transformed by them.