HSE researchers analyzed how effectively the global generative artificial intelligence market turns investments into real income, and came to the conclusion that AI is developing faster than it pays off today.
The results are published in the journal Foresight and STI Governance, according to representatives of the Higher School of Economics.
In recent years, generative artificial intelligence (GenAI) has become one of the main areas of technological investment. Companies are investing billions of dollars in chips, servers, and data center infrastructure, hoping for a rapid economic impact from large language models.
However, market expectations may be overestimated. Yaroslav Kuzminov, Research Supervisor at the Higher School of Economics, and Ekaterina Kruchinskaya, Associate Professor at the Faculty of Social Sciences and Senior Lecturer at the Department of Higher Mathematics at the Higher School of Economics, decided to assess how balanced the generative AI market is and whether there is a gap between investments in infrastructure and revenues from artificial intelligence technologies.
The authors applied the DEA method, a model that is used to analyze the effectiveness of complex economic systems based on a variety of input and output parameters. In this case, the "input" was the income of manufacturers of AI hardware (chips, servers, semiconductors, and data center infrastructure). Among them are companies such as AMD, Intel, NVIDIA, etc. The "output" is the revenue of companies developing and monetizing AI solutions, including Sony, OpenAI, Google DeepMind, Amazon, Apple, and others. This model is literally an imitation of the AI market at the "entrance" and at the "exit" with the assumption that these players set the main agenda.
The analysis covered the period from 2016 to 2024. It is important to note that years were, in fact, units of analysis, although usually these are companies — this is the tradition of the method. This was done deliberately: the authors wanted to test the effectiveness of AI in each specific year as a whole, and not in a single company. To verify the sustainability of the results, calculations were carried out both in absolute terms and adjusted for global GDP. This approach allowed us to assess the relative effectiveness of the generative AI market in different years.
The analysis showed that the development of the GenAI market is nonlinear. As generative models were introduced and initially commercialized, efficiency increased from 2016 to 2021. However, starting in 2021, the trend is changing: performance indicators are declining, despite a sharp increase in investment. After a short-term spike in 2023, efficiency dropped back to 2022 levels.
"Purely methodologically, the results suggest that the AI solutions market is developing according to a catch-up model: revenues from software products have not yet compensated for large-scale investments in hardware infrastructure. The increased demand for chips and computing power is stimulated by the development of large language models, but their commercial returns remain limited and do not cover the cost of hard technologies and further investments in them," Ekaterina Kruchinskaya, Associate Professor at the Faculty of Social Sciences, Senior Lecturer at the Department of Higher Mathematics at the National Research University Higher School of Economics.
According to the researchers, the current development model strengthens the position of hardware manufacturers, but without a return to the economy, because computing power exists as an end in itself. The market for such AI solutions and applications that can influence social processes (for example, increase productivity in the labor market), not only faces limitations — the high cost of hardware and runs, a shortage of qualified personnel and technological limits of models — but also is not a profitable market, especially in comparison with its costs.
"AI is really changing not only the economy and business models of companies, but also public life. Each of us notices this on a daily basis. At the same time, his influence is carried out and spreads, but not as fast as it seems, and not as productively as we would like. Many people are talking about an AI bubble, which is not a new process for the global economy. It is worth saying carefully that there are risks of a bubble. Our model opens an instrumental discussion in this direction. It is important to have not only a tool, but also an application plan, and it is simple. Without increasing the efficiency of applied solutions, their implementation and more balanced investment planning, we will not move further in a positive direction," Yaroslav Kuzminov, Research Director at the National Research University Higher School of Economics.
The authors emphasize that such research is important not only for the scientific world, but also for business, investors and the formation of a balanced scientific and technological policy in the field of artificial intelligence.
