Artificial intelligence is expected to bring important changes to the American economy. Many believe it will make businesses more efficient, increase productivity, and create new avenues for growth.

Financial markets have responded with great enthusiasm. But if we look at economic history, we find that the benefits of big technological changes usually take much longer to show up in the wider economy than investors might hope.

Take electricity, computers, and the internet as examples. Just having these technologies available did not immediately lead to big gains for the whole economy. Businesses had to invest in new infrastructure, train their workers in new skills, and change how they operated to truly benefit. We should expect AI to follow a similar path. Building data centers and expanding electricity infrastructure are important, but they are only part of what is needed for AI to have a broad impact.

It is also important to remember that AI is arriving in an economy that is increasingly K-shaped. There are those at the top branch of the K (upper income) and those at the bottom of the K (lower income). Americans have very different levels of income, wealth, education, and access to opportunities. These gaps matter because they shape who can invest in new technologies, who can learn to use them, and how quickly the benefits of AI can reach different parts of the population.

Large firms with access to capital and a skilled workforce may adopt AI technologies more rapidly. In contrast, small businesses may recognize the potential of AI but lack the financial resources, personnel, or time required to experiment with and implement these tools. Simply acquiring AI software does not guarantee increased productivity. Companies must reconsider existing processes, invest in employee training, and carefully assess when it is appropriate to rely on automated outputs.

This challenge is not new in economics. In 1987, economist Robert Solow observed that computers appeared to be ubiquitous except in productivity statistics. Economist Paul David later demonstrated that the economic payoff from electricity depended on redesigning factories and production processes. The lesson is clear: Invention is distinct from widespread, productive adoption.

The educational landscape in the U.S. is a key factor in this transition. Over one-third of Americans aged 25 and older have attained a high school diploma or less, while fewer than 40% have completed a bachelor's degree or higher. Notably, having less formal education does not preclude individuals from benefiting from AI, as many AI tools are designed to be user-friendly. However, effective use of these technologies in the workplace often still requires digital literacy, access to training, reliable internet connectivity, and sufficient time to acquire new skills.

For lower-income households, these are real issues. The cost of equipment and training can be as much as any other household expense. People who are working multiple jobs will have very limited time available to actually go and get the training. Even when AI tools are low-cost or free, the ability to use them effectively is not evenly distributed across the population.

The accompanying graph from the Federal Reserve puts this issue in perspective. Household wealth has increased substantially since 2011, but the largest dollar gains have accrued to households near the top of the wealth distribution. The bottom half of households holds only a small fraction of total wealth. This is a graph of wealth, not income, and it does not suggest AI created these differences. Rather, it shows the financial starting point from which the next technological transition is taking place.

Why does this matter? At first, AI's financial benefits will likely accrue to businesses and investors, often reflected in higher company valuations, even before the technology boosts productivity across the economy. Wealthier households can take part in these gains right away. Those with less wealth may have to wait longer, possibly seeing benefits later through better wages, new jobs, or lower prices, if those gains happen.

It would be a mistake to assume that educational attainment or income level alone will determine who benefits or loses from AI. Some occupations that require high levels of education may be significantly impacted by automation, while many jobs that require less formal education involve tasks that are not easily automated. The key question is whether workers and employers have the resources and flexibility to adapt.

For these reasons, I am cautious about expecting an immediate surge in AI-driven productivity. While financial markets can price in potential future outcomes today, the real economy must go through a slower process involving investment, training, experimentation, and adjustment. If the adoption of AI remains concentrated among firms and households that are already well-positioned, it may take even longer for the broader economic benefits to reach the wider population.

AI might eventually meet today’s high hopes. But its effect on long-term economic growth will depend not just on what the technology can do, but also on how widely Americans can actually use it. In this way, the AI revolution could be like a random walk, full of surprises, uneven gains, and considerable uncertainty about when the full benefits will materialize.

Dr. Vange Hochheimer is a professor of economics and finance at Whitworth University and CEO of Grand Fir Analytics LLC.