What connects divergent political debates on Artificial Intelligence (AI) is a persistent question: who controls the systems that society increasingly depends on, and what does that mean for society? That question was at the heart of a recent panel I joined at Pakhuis de Zwijger, organized by Professor Daniel Mügge. Below you can find some reflections on the conversation from the perspective of Leitmotiv’s work.

The premise we need to question

There is a strong consensus, particularly in European policy circles, that AI is a driver of economic growth. But before we rush to discuss how best to harness it, we need to interrogate the premise itself.

AI is not a neutral tool. It is virtually impossible to use AI without the hyperscaler clouds: Amazon, Google, Microsoft. So when a government, a hospital, or a newsroom adopts AI, they are not simply upgrading their systems. They are exporting critical functions, like handling patient data, developing software to service citizens, to provide access to accurate, independent news, to a handful of global corporations, deepening a dependency that is very hard to reverse.

I have seen this play out in worrying ways. I have spoken to diplomats from the Baltics who privately acknowledge that their governments are too reliant on the hyperscalers, but who will not say so publicly because they fear consequences from these companies or from Washington. I have also seen it in the media sector, where I worked with Agustin Ferrari Braun, a PhD researcher at the UvA, on a paper called “Clouded Judgements”[^1] about the dangers of outsourcing European media infrastructure to cloud computing platforms.

We talk about AI as if it were electricity: a neutral input that makes everything more efficient. But there is a better metaphor, one I draw from the work of the Programmable Infrastructures Project[^2] and computing historian Nathan Ensmenger.[^3] Adopting AI means moving your entire functioning into ‘someone else’s factory’. They own the machinery, they set the production line, they decide the terms. And once you have moved in, you cannot easily move out.

Nobody would hand over a country’s physical manufacturing base to a handful of tech corporations without serious debate. Imagine saying: we will hand the production of electricity, the running of our water management, or our food production to Amazon, Google, and Microsoft. It sounds incredulous. But that is exactly what is happening with ‘AI’ and it is being sold as mere efficiency.

The losses we are not accounting for

The most important thing we are losing is the ability to question the trade-off itself. We are being told, by the Draghis[^4] and the Wenninks[^5] of this world, that it is either AI or be damned. Either we embrace this model of productivity or we fall behind. But that framing is a little too self-serving towards their aims. Why should we accept that the choice is between productivity and decline, when we all know that progress in society relies on far more than productivity? It relies on democratic accountability, solidarity, and care. None of which can be achieved through AI as such.

And then there is the magical thinking. For a sector that prides itself on data-driven analysis, the claims around AI as a driver of growth are remarkably shaky.

At Leitmotiv, we have done research on the Dutch case that directly challenges claims made in the Wennink report[^5] regarding a lack of data center capacity. Our analysis[^6] of data reported under the EU Energy Efficiency Directive (EED) showed that data centers in the Netherlands use only one-third of their available space for IT equipment, and two-thirds of their available electrical connection capacity remains unused. On top of that, only 104 of the approximately 160 data centers above 500 kW have reported to the RVO, despite a legal obligation to do so.

When that is our baseline, we certainly do not need to start by building more.

Yet, a lot of the current political debate in Europe remains based on supply-driven wishful thinking. The European Commission’s proposed Cloud and AI Development Act (CAIDA) is built on the assumption that Europe needs more data center capacity to stay competitive in AI, but the evidence base[^7] for that scarcity claim is thin. Similarly, in the Netherlands, Wennink’s major advisory report cites an ING bank blog as its source for data center scarcity, while independent research, including ours at Leitmotiv,[^8] points in a very different direction. In our most recent work,[^9] we estimate that Microsoft’s data centers in the Netherlands currently claim approximately 548 MW of grid capacity, equivalent to the grid capacity claimed by over 1.9 million Dutch households, and nearly 1.4% of all electricity use in the Netherlands in 2024.

The real loss is this: we are making enormous public policy commitments based on claims that need to be much better substantiated at best, and do not survive scrutiny at worst. And in the process, we are not just adopting a new technology. We are exporting the critical functioning of our governments, universities, to Big Tech hyperscalers.

Here is a simple exercise I would encourage everyone to try. Next time you hear someone say “we are bringing AI to” some sector, organization, or process, flip it around in your mind: we are handing over education to Big Tech. We are handing over the justice system to Big Tech. We are handing over the government to Big Tech. Then tell me if that sounds like a step forward.

Who gets to define acceptable risk?

Ideally, the level of acceptable risk should be defined by those most directly affected by these technologies, and preferably not by those most invested in their success: venture capitalists, shareholders, and tech CEOs. That sounds obvious, but it is the opposite of how these conversations are usually structured.

Just earlier this month, Politico leaked[^10] that the European Commission plans to open a “dialogue” with Washington on tech rules, raising concerns about a back door for the Trump administration into the EU’s flagship digital laws. Guess who is not involved in those conversations: we are. The people doing critical research, the people raising public interest concerns, the people with real data on the environmental impact of the tech sector.

But I want to add one more dimension, drawing on the earlier mentioned research: we need to treat the dependency that AI creates on hyperscaler cloud companies as a risk in itself. Every conversation about the risks of AI that does not include the risk of making our institutions beholden to the computing environments of Amazon, Google, and Microsoft is an incomplete conversation. Going forward, whenever we talk about AI risk, we should make sure that conversation includes the questions: who controls the infrastructure it all runs on?

What needs to change

We need to see beyond the hype. AI is a symptom, not the cause. What we are entering is a new era in the computing industry where power is shifting again, away from social media companies and towards those who build digital infrastructure: the factory of the digitized world.

Civil society, especially the digital rights movement I am part of, is working to catch up to these changes. For good reasons, we have spent years focused on the impact of consumer-facing technology on individuals, like social media apps. And we have had real wins. But we are now contending with something different: the capture of governments, media, healthcare, and universities by big cloud computing companies. Or in other words, we need to bring transparency to the fastest growing side of the tech industry: the provision of business-to-business services and its impact on the environment we all live in.

Here is the uncomfortable question we need to sit with: what do we overlook when as individuals, we have perfect privacy through open-source apps like Signal or Mastodon, yet such privacy still runs on Google or AWS cloud environments? Especially if these can be turned off at the flick of a switch by these companies or by the governments of the countries they reside in.

What a different ‘leitmotiv’ for a just technological future could look like

One in which we actually dare to explore alternative futures. And that means rethinking not just individual technologies like AI, but the larger systems they are building around all of us.

The good news is that there is incredibly rich work to draw on for alternatives. Octavia Butler’s science fiction imagined building new societies from collapse in The Parable of the Sower. Ingrid Robeyns makes the case for limitarianism, capping extreme wealth and power. David Graeber showed us that radically different ways of organising society are not hypothetical but historical. And feminists have long argued that care, not productivity or competitiveness, should be the organizing principle of society.

Our focus at Leitmotiv is on the environmental and democratic impacts of digital infrastructure, from data centers to cloud computing to AI systems—and how those could be done alternatively. Practically, this means we are working with local activists resisting data centers as well as with EU policymakers to build transparency into the cloud provision market. In doing so, we work to make these infrastructures visible, accountable, and subject to democratic deliberation. Not because we think technology is inherently bad, but because we believe that choices about how our digital systems are built, who controls them, and whose interests they serve are political choices that deserve political scrutiny.

The ideas are there. The question is whether we have the political will to take them seriously as a basis for technology policy, rather than treating them as a nice afterthought to the next industrial strategy.

The questions we should keep asking

Not “what can AI do?” but: what deeper shifts in the tech industry lie behind the AI hype? What new reliance on hyperscaler cloud computing are we sleepwalking into? What are the long term environmental consequences of doing so? Those are the shifts that will define who holds power in society for decades to come, and that is also where the leverage for change lies. If we only ever talk about AI, we are looking at the surface. The real action is in the infrastructure underneath. We try to bring these debates out of the server room and into the public square, where they belong.

References