Artificial intelligence is advancing so quickly that some observers see it as more than a new technology. They see it as a solution to one of economics’ oldest problems. Markets fail for many reasons, but a common thread runs through most of them. Information is costly to acquire. Finding the right trading partner takes time. Negotiating agreements is expensive. Monitoring performance is difficult. Economists refer to these frictions as transaction costs.

AI optimists believe the technology could dramatically reduce those costs. In textbook economics, buyers and sellers find each other, know what they are buying, bargain freely and enforce promises at little cost. In the real world, however, people search, verify, compare, negotiate, monitor and sue. Those activities consume time and money. When they are expensive, gains from trade go unrealized. That is the basic insight of transaction cost economics . The boundary between markets that work and markets that fail lies in the cost of making a bargain work.

AI can help because much of market exchange is really an information problem. A buyer wants to know what a contract says, whether a product is reliable, whether a supplier can be trusted and whether a better option exists elsewhere. AI can answer many of those questions faster and cheaper than before. It can make hidden information easier to find and complex information easier to understand.

So far so good. Better information and lower bargaining costs should make markets work better. But it does not follow that market failure is heading for extinction. Every familiar category of market failure has an AI version, and in many cases the technology reduces one friction while amplifying another.

Externalities do not go away

Externalities arise when private decisions impose costs or benefits on third parties outside the transaction. AI may help internalize some of these spillovers by making them easier to detect, price and account for. It will also create new ones.

Data centers are the clearest example. The International Energy Agency reported that electricity demand from data centers rose 17 percent in 2025, with AI-focused centers growing even faster. Total data center electricity use is expected to double by 2030, while power use from AI-focused centers is poised to triple. Those costs do not fall neatly on the purchaser of an AI subscription. They show up in local grid investment, water pressure, land disputes, backup power generation, and emissions.

The point is not that AI is bad for the environment. The same technology can optimize power grids and make factories more efficient. The point is that where a user sees a cheap query, a community may see higher peak electricity demand or a water system under stress.

Better information can mean better deception

Asymmetric information arises when one side of a transaction knows more than the other. AI can narrow that gap. It can read the fine print, compare complex products and explain risks in plain English. It can also widen the gap by making deception cheaper.

The same tools that help consumers evaluate a product can help bad actors make deceptive products look more legitimate. Fake testimonials become easier to write. Images become easier to fabricate. Invoices and credentials become easier to mimic. FTC fraud data show consumers reported losing more than $12.5 billion to fraud in 2024, up 25 percent from the year before. An FBI warning about generative AI notes that criminals can use it to make fraud more believable while reducing the time and effort needed to deceive targets.

AI also introduces a new information asymmetry. The AI system itself may know more about the buyer, the seller, and the transaction than either human party does. When the tool is aligned with both buyer and seller, that knowledge can make markets work better. When it is aligned with one party over the other, it can make manipulation easier. And when it is aligned with neither, it may pursue its own goals, benefiting neither party.

Public goods still need patrons

Some knowledge is valuable precisely because many people can use it at once. A disease warning, for example, becomes more useful when it reaches more people. But if everyone can benefit from the warning whether or not they helped pay for it, private incentives to produce the information may be weak. The result is often too little investment in creating and maintaining the knowledge society relies on.

AI will help spread useful knowledge, which is a genuine benefit. It can make research, education and expertise easier to access. But making knowledge easier to use is not the same as giving people stronger incentives to produce it. The public health study still requires a clinical trial. Poor school districts still need better testing. And cybersecurity still requires firms to invest in protections whose benefits often extend far beyond the firms paying for them.

In addition to public goods, AI will also create public bads. False information can spread like pollution. One person’s exposure to a fake study, synthetic image or false claim does not prevent millions of others from seeing it too. When AI makes misinformation cheaper to produce and easier to personalize, the damage can spread widely before anyone has a strong private incentive to correct it.

Markets may concentrate while property rights weaken

Monopoly arises when a small number of firms gain enough market power to limit competition. AI has several features that could push markets in that direction. The most advanced systems require enormous amounts of computing power, specialized talent, data and distribution. Those inputs are not evenly spread across the economy. They are often concentrated among a small number of large firms.

This matters because lower costs for users can coexist with higher barriers for producers. A consumer may get a cheap AI assistant. A new competitor may face the much harder task of securing chips, cloud capacity, training data and access to customers. The small startup may have a clever model design, but the incumbent may control the infrastructure needed to reach scale. An FTC staff report raised concerns about partnerships between large cloud providers and AI developers. The worry is that these arrangements could shape who gets access to key inputs and make it harder for customers to switch providers.

AI also puts pressure on property rights in creative industries. When imitation becomes cheaper, the market value of original work can fall. The music industry offers a warning. After file sharing took off, U.S. recorded music revenues fell from $14.6 billion in 1999 to $6.97 billion in 2014, according to RIAA figures . Streaming eventually helped rebuild the industry, but the rewards remained highly concentrated. A small number of artists capture a large share of listening, revenue and cultural attention. Spotify data for 2025 show 80 artists generating more than $10 million from the platform, while 303,200 generated more than $1,000, out of roughly 13 million people who have uploaded at least one track. At the very top, Forbes puts Taylor Swift’s net worth at $2 billion.

The broader lesson is that digital markets can weaken property rights while concentrating rewards among a few winners. AI may push writing, graphic design, software development and other knowledge industries in a similar direction. It may democratize production in some areas while centralizing distribution, giving more people the tools to create but leaving the largest platforms with much of the power to decide what gets seen and monetized.

The crowd can get smarter and still be mad

A final category of market failure is behavioral. Markets are made of real people, and people do not always behave like the rational actors in economics textbooks. They follow stories, extrapolate recent trends unrealistically, and take comfort in doing what everyone else is doing.

AI may sometimes discipline those instincts. A model can force an investor to look at the numbers, compare risks or pause before making a bad decision. But it can also make herding behavior easier. If millions of investors rely on similar models trained on similar data, their judgments may begin to converge. Portfolio managers may outsource decisions to systems that reproduce the same consensus. Firms may adopt similar pricing strategies because they rely on the same tools to recommend them. Consumers may crowd into the same products because recommendation engines all direct attention in the same way.

Financial regulators already see the risk. The Financial Stability Board has warned that AI could amplify vulnerabilities when many firms use similar models and react to the same signals. The European Central Bank has raised a similar concern, warning that widespread AI use may increase herding behavior and correlated returns while strengthening winner take all dynamics in finance. If a small number of models come to dominate investment decisions, advantages in data, computing power and distribution could become self-reinforcing. The result may be a financial system that is more concentrated, more correlated, and less diverse in its judgments.

The danger is that AI could make investing faster and easier without making it more grounded in fundamentals. When many market participants buy because the same tools point in the same direction, plausible narratives can turn into asset bubbles. Faster information processing does not guarantee better prices if everyone is processing the information in the same way.

AI could also affect saving decisions. Personalized financial advice may help some households plan better and build more wealth. If people believe AI will make them more productive and wealthier in the future, they may choose to save less today. If they fear AI will make jobs less secure, they may save more as a precaution. Either response is plausible, so there is no reason to assume AI will automatically push savings and investment toward socially optimal levels.

Greater efficiency is likely. Perfection is not.

None of this means AI will be bad for markets on net. The opposite is more likely. By making information easier to find and understand, AI should make more exchanges possible. It can help buyers and sellers discover one another, reduce the cost of earning trust, and bring expert assistance within reach of people and firms that once could not afford it.

But markets have never been frictionless. They depend on property rights, courts, reputation, norms and sometimes also regulation. AI will enter that imperfect world. It will reduce some transaction costs while creating others. It will solve some information problems while also generating new ones. It will give more people the tools to create, while perhaps leaving a smaller number of platforms with the power to decide what gets seen and monetized. It will give people better information, but also new reasons to follow the crowd.

AI may well make markets more efficient. What it will not do is make them perfect. Markets have always been imperfect, and they will remain so.