Who governs the algorithms of the green transition? The questions that need to be asked

Artificial intelligence has entered green industrial policy. It entered through energy system planning, critical mineral logistics, carbon traceability models, smart grid management, and many other optimisations. It entered fast, without much noise, and almost always through the back door: not as an explicit political decision, but as a consequence of adopting the available tools, of accepting the platforms that already exist, of using systems that someone else built for another purpose.

And yet, nobody seems to be asking the most obvious question. Not whether AI is useful for the green transition — that is taken for granted, sometimes too quickly. The missing question is a different one: who builds these systems? Who owns them? Who decides what they do and how they evolve? In whose interest do they operate, and at whose expense?

These are not technical questions. They are political questions. And the fact that they are almost absent from the debate on energy transition and green industrial policy is not an innocent omission. It is a symptom of the assumptions that organise that debate: that technology is neutral, that it is available, that it is a global public good accessible to any state with the will and resources. The problem is that none of those three assumptions are true. Critics of tech have been questioning these ideas for decades, yet they seem absent from many debates today.

Ha-Joon Chang (2002) built one of the most precise metaphors1 in the political economy of development: the ladder that gets kicked away. Countries that are rich today industrialised their economies with state protection, active subsidies, and deliberate industrial policy. Once at the top, from that position of advantage, they built the rules of the international trading system that prohibited others from doing exactly what they had done. Free markets for those who already won. Interventionism banned for everyone else.

In the digital economy, that ladder gets kicked away twice.

The first kick is intellectual property (IP) rules. The most powerful AI models, the ones now entering energy planning systems, grid management, industrial optimisation models, are built in the Global North and protected by an IP system designed to keep them that way. It is not that countries of the South lack the capacity to develop technology. It is that the rules of the game are built to ensure that capacity does not translate into market power, technological independence, or real alternatives. The gap is not technical. It is political.

The second kick is free trade agreements. These agreements  (bilateral, plurilateral, and even within the WTO and WIPO) consolidate intellectual property restrictions and pile on new ones. They prohibit tariffs on digital flows, limit data localisation requirements, prohibit algorithmic transfer and access, and enforce regulatory neutrality that only benefits those who arrived first and built global platforms before regulation existed2. These are treaties written, in large part, by and for the interests of Northern technology industry.

We need to seriously ask what digital and technological capacities states need to implement in terms of green industrial policy. A state that wants to design active industrial policy in the twenty-first century needs data about its productive structure, trade flows, value chains, energy consumption, and emissions, among other things. It needs models to process that data and project scenarios. It needs systems to monitor policy compliance and adjust them in real time. It needs, in short, to be intelligent, that is, to have digital infrastructure that responds to its own priorities, not to those of whoever built it for a different market with a different logic.

The answer most frequently circulating in progressive policy spaces is “technological sovereignty.” The concept has immediate political appeal: it names the problem, identifies the adversary, proposes a direction. The problem is that it is almost empty if not filled with imprecision.

Whose sovereignty? The sovereignty of a large state with its own industrial capacity is not the same as that of a small, open economy with limited technological critical mass. It is not the same in the context of an integrated regional bloc as in that of an isolated country negotiating alone with global providers. The concept that seems universally applicable conceals very different conditions of possibility depending on who speaks it.

Sovereignty over what, exactly? Over data? Over models? Over computing infrastructure? Over decision algorithms? These are different layers of the problem, with different logics, and advancing on one does not necessarily mean advancing on the others. A country can have its own servers and still depend entirely on foreign models. It can have localised data and lack the technical capacity to process it. It can have technical capacity and lack the institutional framework to govern what that capacity produces.

China is often invoked as proof that a state can simply build its way out of digital dependency. The reality is more layered. China has its own software stack, its own general-purpose hardware manufacturing base and, critically, the institutional capacity to design and adjust its own digital policy. In those layers, it is sovereign. But it still depends heavily on foreign-designed, mostly US-controlled, advanced microprocessors3. Recent progress on domestic lithography is real, but it remains at prototype stage, years away from resolving that dependency. China's case does not refute the argument above; it illustrates it. Sovereignty over one layer does not translate automatically into sovereignty over the others.

TNI’s GIP Lab working group discussed this for long. What emerged was not an answer but a more precise tension. For most countries of the Global South, especially small and medium-sized ones, full technological sovereignty might not be a realistic horizon in the short or medium term. Not because they lack capacities, but because the starting point is a structural asymmetry that is not resolved by political will alone. That structural asymmetry pushes the question toward a different terrain: not each state building sovereignty alone, but regional cooperation and shared digital infrastructure as the more realistic path for small and medium economies.

But there is a second trap, less visible than the first. Even when a state manages to build or control its digital infrastructure, even when it advances toward some real form of technological sovereignty,  that guarantees nothing on its own. Infrastructure can be state-owned without being democratic. It can be public without serving the collective interest. It can survive one election and not survive the next. It can be captured by national elites who use it to concentrate power or to reproduce the same extractive logics with a different accent.

The history of state enterprises in the Global South is complex enough to discourage romanticising public ownership. Not every nationalisation is democratisation. Not every state control is popular control. The question is not only whether the infrastructure is public: it is whether it is democratic. Who decides what it does, how it evolves, in whose interest it operates. Who can audit it, challenge it, change it. What accountability mechanisms exist beyond the electoral cycle.

The central tension, then, is not only between dependency and sovereignty. It is between ownership and governance. And that distinction, which the Digital Public Infrastructure literature is beginning to recognise but has not yet elaborated with sufficient rigour,  is the one that should organise the debate on technology and green industrial policy.

There are three questions that should organise any serious discussion about artificial intelligence and green industrial policy. Today they are answered separately, when they are answered at all. And that separation is precisely the problem.

The first: what technology? Not all artificial intelligence is equivalent, and treating it as if it were is one of the most costly mistakes a state can make. A large model trained in the North on Northern data, deployed as a commercial service in the South, has a logic, biases, and dependencies very different from a system built with local data, for local problems, under local governance. Scale matters, but not in the way usually thought: bigger does not mean more useful for a specific context, and more sophisticated does not mean more appropriate for the priorities of a sovereign industrial policy. The question of what technology is not a technical question that engineers can answer alone: it is a political question that requires clarity about what one wants to do, for what purpose, and under what conditions.

The second: what governance? Who decides what the system does? Who has access to its results and who does not. Who can audit it, question it, modify it when it produces problematic results. Who can shut it down? With what accountability mechanisms, to whom, with what frequency. These questions have no technical answers, they have institutional answers, and building those institutions requires time, capacity, and political will that cannot be improvised. Governance is not what gets designed after choosing the technology. It is what determines whether the technology serves the purpose assigned to it or ends up serving another.

The third: for what purpose? This is perhaps the most uncomfortable, because it forces us to name the contradictions. An AI system for energy planning can be genuinely useful for designing a sovereign transition, or it can be primarily a mechanism for extracting strategic information for the companies that provide it. A critical mineral traceability system can serve a state to understand and regulate its value chain, or it can be the instrument through which buying companies in the North impose their standards and conditions on Southern producers, with no real benefit to affected communities. The difference is not in the code. It is in who governs it, with what logic, and in whose interest.

It is not enough for it to be state-owned. It must pursue the eco-social goals that communities set for themselves. Otherwise the green transition is a dressed-up transition: one more “greenwashing” within a capitalism that wants to save the ecosystem in order to perpetuate its extractive culture.

These three questions are inseparable because they condition each other. The choice of technology delimits the governance options available. The governance structure determines what purpose the system ends up serving in practice, regardless of the purpose for which it was designed in theory. And the purpose , if defined clearly from the outset, should be what guides both the technological choice and the institutional design. Starting with any of the three without the other two is the most common, and most costly, way to get it wrong.

There is still no systematic critical analysis that articulates green industrial policy, democratic digital governance, and the political economy of the Global South from a perspective that takes all three seriously at the same time. There is brilliant literature on each one separately. There are partial bridges, conversations beginning to intersect, researchers working at the edges. But the field as such, with its own analytical framework, its constitutive questions, its political language , remains to be built.

That is the problem. And also the opportunity.

This is where the cooperation route sketched earlier needs to be taken seriously, not as a slogan but as a research object. South-South cooperation as an alternative model of technological development exists as an embryo, with concrete experiences that deserve rigorous study. Digital commons models (free software, open data under collective governance, shared infrastructures) exist as technical and political proposals that have not yet been seriously evaluated in relation to the specific requirements of green industrial policy. Participatory governance frameworks for technology exist in the academic literature and in some pilot experiences, but have not been translated into the language of industrial policy. Connecting those dots, evaluating them rigorously, giving them political language, situating them in the concrete debate on energy transition in the Global South, is work that remains to be done.

That is the task. It is not small, and we must not rush toward answers that do not yet have sufficient grounding. But we need to start, with the right questions, with analytical rigour, and with clarity about for whom this knowledge is being built and what it is for.