
Externalities
Externalities are costs or benefits that fall on uninvolved third parties, without anyone paying or being paid for them. In the tech debate, they show up wherever data centers, data collection, or automated systems have consequences for people who are neither customers nor providers.
When two parties make a deal, the outcome often affects third parties as well. A factory sells cement to a builder, and the smoke from the chimney lands on the neighbors. Those neighbors bought nothing and sold nothing. Yet they bear part of the cost. Economists call exactly this kind of side effect on uninvolved parties an externality. It can also be positive: someone who plants a beautiful front garden improves the view for the whole street without getting paid for it.
Why the price then misses the truth
In a market economy, price governs how much of something gets produced. If something is expensive, it is used sparingly. If it is cheap, people help themselves generously. This system only works if the price contains all the costs. In the case of externalities, that is precisely not the case.
The result is called market failure. The cement in the example is too cheap because the health damage suffered by the neighbors appears nowhere in the price. As a result, more of it gets bought than would make sense for society as a whole. With positive externalities, the opposite happens: basic research benefits many people, but the researcher only earns a fraction of that benefit. So too little research gets done.
For evaluating technology companies, this is a central point. A company can be highly profitable and still, on balance, harm society. The profits appear in the balance sheet, the costs fall on others. Anyone who only looks at the balance sheet does not see this shift.
How to bring the costs back into the calculation
The standard recipe comes from the economist Arthur Pigou and is called internalization. The state imposes a levy on the harmful activity, roughly equal to the amount of the damage. Then the damage shows up in the price, and the company factors it in on its own. Germany’s CO2 levy is built exactly this way.
A second method works with tradable rights. The state sets an upper limit, for instance on emissions, and distributes a corresponding number of certificates. Anyone who emits less can sell their remaining rights. This creates a price where none existed before. The European emissions trading system works on this principle.
Third, there are simple bans and limit values. They are cruder than a levy but easier to enforce. In all three approaches, the real difficulty lies in measurement. How much is a lost year of life expectancy worth? How does one put a figure on the damage caused by a piece of misinformation that spreads millions of times over? Such numbers are disputed for decades.
Externalities in the debate about AI and data centers
Data centers for artificial intelligence consume a great deal of electricity and cooling water. The operator pays its electricity bill, but not the consequences that running the power plant has for the climate. Water consumption in dry regions likewise affects local residents who get nothing from the servers. Such figures now regularly appear in the sustainability reports of major providers, because investors ask about them.
A second field is social networks. A recommendation feature optimized for time spent can reward conflict and outrage, because that keeps users engaged longer. The advertising revenue goes to the operator. The costs of heated debates are borne by society. Creators make a similar argument regarding training data: their texts and images improve a model without their being involved.
A common mistake is to confuse externalities with costs in general. Server prices or salaries are ordinary costs, because they are laid out in a contract and get paid. An externality only exists once no one bears the cost of the effect. Regulation such as the EU’s AI Act can be read as an attempt to assign precisely these unpaid effects to someone responsible.