AI is spreading through every industry
I have used AI nearly every day for the past three years. It helps me write code, research topics, and structure my thinking. I am not alone. Across Europe, companies are embedding AI models into their workflows: document processing, customer support, software development. The technology is real, it is improving, and it is not going away. The tools deliver more output per person, so adoption will continue. AI is already reshaping how we work. The question is who captures the value and what we do about it.
Much of the value lands outside Europe
The infrastructure behind AI, the cloud platforms, the systems that store and process data, the models themselves, is owned almost entirely by a small set of US companies. Microsoft, Amazon, and a few others provide the computing power. Databricks and Snowflake provide the platforms that store and organize data. OpenAI and Anthropic provide the models. When a European company adopts AI, it pays rent to these providers. The more AI is used, the larger the rent. Like computers and the internet before it, AI is moving to the core of how companies operate. Structural dependency on a handful of foreign providers is a vulnerability that grows with every new model deployment. European companies are paying rent to providers they cannot easily walk away from. If Microsoft hosts your AI models and those models become integral to how you operate, switching cloud providers is not something you do over a quarter. It threatens your ability to function.
Additionally, the rent is not taxed where the value is generated. These companies route European revenue through Irish subsidiaries and pay effective tax rates lower than in the companies they actually operate in. The hundreds of millions of Europeans whose data and purchasing power make these platforms valuable do not see enough of it flow back into the schools, roads, and hospitals of the countries they live in.
Enforce the rules we already have
Some have proposed an algorithm tax, a levy on automated decision-making systems. Target the technology directly, make the machines pay. But an algorithm tax treats AI as if it were a new category of economic activity, distinct from the software and infrastructure that it comes with. AI models run on servers. Even if models eventually refine themselves, in the very end, large language models are still steered and used by humans. I wrote about this distinction earlier: humans cannot outsource strategic thinking and understanding to AI. AI is used to generate economic value that will be captured via profits and cashflows, just like any other value-adding process of companies. Taxing the algorithm itself is like taxing the welding robot instead of the car that comes out of the factory.
Foreign AI companies enable us to use this new technology and therefore bring tremendous value. But it is a real problem that they operate under a different set of rules than the societies they serve. Instead of creating a new tax category, we should enforce the ones we already have. A global minimum corporate tax is the cleanest path. It removes the incentive to route profits through low-tax jurisdictions. If that proves politically impossible given current geopolitical tensions, a digital services tax is a fallback. It is protectionism, and it carries the risk of retaliation. But the alternative, letting the largest companies in the world extract value from an entire continent while paying near zero, is worse.
At the same time, Europe has to keep working on running AI services more independently. This means building more European infrastructure and using the data we have more effectively.
Let people own a piece of the future
AI amplifies a dynamic that predates it: for decades, returns on invested capital have grown faster than the broader economy. The owners of capital accumulate wealth at a rate that wage earners cannot match. AI accelerates this because it makes capital, in the form of computing power and AI models, more productive relative to labor. While the speed is new, the mechanism stays the same.
Capital already outpaces wages, so wages alone cannot fund retirement. Citizens need to own a slice of the capital that is growing. A broad basket of stocks from around the world, held through a government-supported pension structure, lets citizens participate in the returns of that accumulating capital. Germany is already moving in this direction, shifting away from its pay-as-you-go pension system toward one that includes broad stock market investments. It is the kind of structural adjustment that was overdue before AI arrived. The technology just made it even more important.
Fair taxation and broad capital participation are not new ideas. AI is a technological shift, not a rewrite of economic law. Companies are still valued on future cash flows. In the very end, humans carry the responsibility for what the machines do. The rules of the game have not changed; the clock just runs faster. We do not need a bespoke tax on algorithms. We need to do the structural work we already knew was necessary, and the path forward is the same as before.