AI must not only cut costs. It must create revenues
After a 2025 marked by continuous increases in the prices of stocks most exposed to Artificial Intelligence – an enthusiasm that also lifted the valuation of less exposed companies – 2026 began in a very different way. Within a few days, the share prices of firms producing software for services fell by more than 20%, influenced in part by a report from analysts at Citrini Research and by a blog post from Matt Shumer, CEO and co-founder of Otherside AI. It is easy to be confused by such price movements, especially for those who question investors’ ability to rationally price listed companies based on fundamentals. But market volatility is only the surface of the issue. If we step back and focus on what might be called the “economy of Artificial Intelligence,” what do we really know? Have blog posts revealed data and results that academic research has so far overlooked?
Technological innovation and productivity
In the 1990s, Nobel laureate Robert Solow famously observed that one could see computers everywhere in the United States – except in productivity statistics. The so-called “Solow paradox” taught a fundamental lesson: technological innovation does not automatically translate into aggregate growth. It was a surprising outcome. Why had the diffusion of personal computers and the adoption of systems such as Windows – which “democratized computing” and allowed even those without programming skills to use machines – failed to boost productivity in a largely service-based economy? For some, the answer is straightforward: productivity does not rise simply because everyone owns a computer, but because people equipped with computers build networks for exchanging information.
The debate has resurfaced in recent years. AI is a new technology that has been affecting the U.S. economy for about a decade, and once again there is no consensus about its impact. If ChatGPT is the “window of the new millennium,” the tool that democratizes computer coding, how long will it take before we see tangible results? On one side are the enthusiasts, such as Stanford economist Erik Brynjolfsson, who has for years documented the impact of predictive analytics and AI on revenues and costs. On the other are the chronic pessimists, such as Nobel laureate Daron Acemoglu, who has argued that AI will have only a minimal effect on productivity. Both may be right. Aggregate productivity does not depend on the adoption of a single tool, but on the reorganization of processes, incentives, and business models. It is not the technology that makes the difference – it is the system that incorporates it. The first key lesson from forty years of debate on innovation and productivity is therefore simple: there is no linear relationship between the two variables at the level of the overall economic system. This does not mean that productivity effects will never materialize, only that they must be carefully observed and measured – over long periods.
AI, the labor market, and income
If productivity divides opinion, the labor market polarizes it even more. Optimists and pessimists trade places when assessing AI’s ultimate impact on employment. Those who believe AI may have a short-term impact on productivity more readily imagine a future in which millions lose their jobs, replaced by machines. Once again, economic analysis helps clarify what might happen, as the issue concerns the degree of substitution or complementarity between human labor and machines. Jobs in which human labor is easily replaceable (call centers? coding?) are likely to see a reduction in labor and an increase in capital. Where AI acts as an augmenting tool, however, one can even imagine an increase in both labor and physical capital – especially for those best able to use the new tools. The second lesson is equally simple and important: the labor market will change, and new opportunities and jobs will emerge. But if AI functions as an augmentation tool, we are likely to see a marked shift in income distribution in favor of physical capital—and thus of company owners, whether publicly listed or not—unless the public sector intervenes to rebalance the outcome.
The Real Market Issue: Revenues
Academic papers provide important evidence on AI’s impact on firms in terms of revenues and costs, even if they are not read in the context of short blog posts that can move billions of dollars in market capitalization. Investors’ concerns in 2026 had been anticipated at least a year earlier by a paper from OpenAI researchers, who quantified (using ChatGPT, of course) the percentage exposure of different economic sectors to AI, finding very high levels (45% of jobs have exposure of at least 40%). But what are investors pricing in? Other studies show that, so far, investors and analysts have revised the valuations of listed companies by incorporating scenarios of cost reduction, but not revenue expansion. And this is the real challenge for firms across all sectors: to demonstrate that AI can generate revenues, not merely reduce costs.
Conclusions
If AI is used only to produce the same goods and services with less labor, the macroeconomic result will be income redistribution – not necessarily higher growth. One firm’s costs are another firm’s revenues. Cutting costs at scale without creating new markets risks compressing the system as a whole. The challenge for companies is not to prove that AI increases EBITDA. It is to prove that it expands the revenue frontier. This is not a technological question. The decisive question is not whether AI works. It does. The question is whether it will be used to:
• substitute labor or increase overall productivity;
• compress costs or create new markets;
• concentrate income or generate inclusive growth.
Financial markets are beginning to grapple with this transition. Efficiency is no longer enough. What is required is real growth. AI will not be judged by how much it cuts, but by how much it creates. And it is on this ground that it will be decided whether the revolution will be expansionary – or merely redistributive.
Andrea Beltratti is Full Professor in the Department of Finance of the Bocconi University, where he teaches Economics of the Real Estate Market and Equity Portfolio Management, and Academic Director of the Executive Master in Finance (EMF) at the SDA Bocconi School of Management.
Alessia Bezzecchi is an Associate Professor of Practice in Corporate Finance & Real Estate at the SDA Bocconi School of Management, where she is the Program Director of the Executive Master in Finance (EMF) and of the Executive Program in Real Estate Finance and Real Estate (EPFIRE).
Photo iStock / akinbostanci
