A new note from Goldman Sachs is pushing back against the prevailing narrative that artificial intelligence will soon reshape the global economy. The investment bank’s economists argue that the technology, while impressive in demonstrations, remains far too expensive and unreliable to drive the kind of productivity boom that many tech optimists are forecasting. The report presents a sobering reality check for an industry that has poured hundreds of billions of dollars into AI infrastructure with the expectation of a massive payoff.
Goldman Sachs chief economist Jan Hatzius and his team examined the historical trajectory of major technological shifts, from the electric motor to the personal computer. Their conclusion is blunt. AI today does not resemble those earlier breakthroughs in terms of cost efficiency or proven return on investment. The economists point out that for an AI system to be adopted at scale, it needs to solve a genuine business problem more cheaply and reliably than existing methods. By that measure, current generative AI tools often fall short.
High costs and reliability issues
The central problem, according to the Goldman analysis, is that running large language models and other AI systems remains extraordinarily expensive. Training a single advanced model can cost tens of millions of dollars, and inference, the process of actually using the model to generate answers, adds significant ongoing operational costs. For many businesses, the cost of deploying AI to automate a task still outweighs the savings from the labor it replaces.
Reliability is a second major hurdle. The report notes that AI models frequently hallucinate, generate incorrect information, or struggle with tasks that require consistent logic. In a business setting, these errors can be costly and damaging to customer relationships. Companies cannot simply trust an AI to handle customer service, accounting, or legal document review without extensive human oversight, which erases much of the efficiency gain. The economists suggest that until these reliability problems are solved, enterprise adoption will remain limited to narrow, low-stakes applications.
The Goldman team also questions the timeline for AI to meaningfully boost productivity. Historically, major technologies took decades to show up in national productivity statistics. The personal computer revolution, for instance, did not produce a measurable productivity acceleration until years after widespread adoption. The economists argue that AI will likely follow a similar slow path, and that the current hype cycle has grossly compressed expectations.
Wall Street bets big despite the skepticism
Despite the cautious outlook from their own economics department, Goldman Sachs itself has been a major player in the AI investment frenzy. The bank has advised on huge funding rounds for AI companies and has been bullish on the stock market potential of firms like Nvidia. This tension between the economics team’s analysis and the investment banking division’s enthusiasm highlights a broader conflict in the financial world. Many investors are betting that AI will be the next transformative platform, even as careful analysts point to the lack of evidence.
The report specifically challenges the idea that AI will eliminate millions of jobs in the near future. Instead, the economists see AI augmenting existing roles in the short term, much like software tools augmented clerical work in the 1990s. Jobs may change, but mass displacement is unlikely without a radical improvement in the cost and capability of AI systems. This is a direct rebuttal to the more alarmist predictions from some tech executives and futurists.
What this means for the AI industry is a potential recalibration of expectations. If the Goldman analysis is correct, the enormous capital expenditure on data centers and GPU clusters may not yield the rapid returns that investors are currently pricing in. Some companies could face a harsh reality check if AI fails to deliver the promised efficiency gains. The hype may not be a bubble in the sense of a crash, but it could certainly deflate as the timeline for real-world impact stretches further into the future.
The message from Goldman is not that AI is unimportant or that it will never matter. It is that the technology is currently being oversold as a near-term economic force. The true transformation, if it comes, will likely be slower, more expensive, and more difficult than the current boosterism suggests. For founders and investors building in this space, the challenge is to stay realistic about what the technology can actually do today, while still preparing for the long arc of progress. For more context on how these shifts affect the startup ecosystem, read our analysis on Mylistingo.
