En 
It 
Simone Vannuccini / 27 February 2025
Commentary no. 318
In some ways, it feels natural to consider Artificial Intelligence (AI), by virtue of its strategic importance, as one of the policy priorities in the ongoing European attempt to contain vulnerabilities, achieve technological sovereignty and increase competitiveness. However, contrary to what headlines say, AI is not a strategic technology. Yes, it is a technological breakthrough; yet, even the most advanced products based on large language models produce little to no profits. While AI introduces a disruptive new logic of interaction with information, its adoption is mostly limited to end users. It is also a dual-use technology that poses various risks to society, from enabling auto-pilot drones to accelerating disinformation and thereby undermining democracy; however, this is true of other technologies that could be weaponised. A more useful approach is to consider AI “a solution in search of a problem”. Its place in history is more akin to “a key development within the still-evolving information communications technology revolution”.
None of this is to say that AI does not have widespread impacts: every radical innovation lowers some kind of cost and, in so doing, changes incentives and opportunities. AI lowers the cost of prediction (content generation is a special case of prediction), which means societal resources are reallocated to prediction-intensive activities. But why are governments so focused on this technology, as if their ability to remain (or become) competitive in the global landscape depended solely on it? The answer lies in the fact that, after a brief phase in which it was wrongly (and ideologically) conceived as an existential risk, now AI has emerged as the “strange attractor” that represents a deeper transformation: the return of power politics in an increasingly fragmented global system. The European Union’s (EU) discourse on AI is influenced by a desire not to be left out of the club that is deciding the shape of the future global order. Thanks to the Letta and Draghi Reports and the change of the guard in the US, the EU has become aware of its geopolitical fragility and the accumulated investment gap – what Draghi called the “tariff” that Europe has imposed on itself. European policy makers cannot really escape the AI arms race, since this is in fact the race to join the game of repositioning in a scrambled globalisation.
The wave of large investments in AI announced in recent weeks (OpenAI’s $500 billion Stargate project backed by the US administration, Macron’s pledge of €109 billion in response, and the European Commission’s announcement of €200 billion to be mobilised through the InvestAI initiative) makes it clear that when we talk of AI policy nowadays, what we are talking about is industrial policy. The willingness to invest such a scale of resources in AI is in line with the growing consensus on the pro-active role of the state in the economy. To be clear, as we argue elsewhere, this is a welcome development. Furthermore, the economic rationale for industrial policy is strong: strategic industries are normally characterised by private under-investment (as externalities make private returns lower than social ones) and absence of markets, which justify the possibility of direct government intervention. However, as the pro-industrial policy narrative gains momentum, competition policy risks being sidelined. Recent statements by Commissioner Ribera in the EU and by the Competition and Markets Authority in the UK suggest that competition authorities are broadening their focus, trying to balance the protection of a level playing field with the demand for industrial consolidation to reach a scale capable of competing on the global stage, as recommended by the Draghi Report.
In the case of AI, a carelessly crafted industrial policy poses two risks. First, as priorities have rapidly shifted from industrial policy centred on enabling the green and digital transitions to industrial policy focused on ensuring autonomy and security, AI industrial policy is likely to evolve into AI nationalism. In this scenario, we will witness an AI race to the bottom, with competing policies that can undermine trust and cooperation not only among international allies, but also within the EU. We already have evidence of this, for example the scraping of the proposed AI liability directive. Second, the pressure on EU institutions and governments to turn initiatives into quick political returns could lead to the wrong kind of directionality. European AI industrial policy must prioritise reducing the huge fixed costs of producing AI, rather than just the marginal costs of developing consumer market applications. The problem is that there are no easy solutions: unlike projects such as Galileo or Copernicus, where the EU allowed itself ample time to develop alternatives to US solutions, the AI sector has seen European policymakers grant tech giants enough time to establish an un-contestable dominant position as “essential facilities”. The European Commission’s InvestAI and the associated AI factories initiative now explicitly recognise the European gap in AI infrastructure, but in their current form they are applying a patch to a sinkhole.
Does this mean that the EU should give up the possibility to compete on all the layers of AI technology? Given the strength of current geopolitical pressures and the long-term existential effects of losing even more sovereignty and bargaining power at the international stage, the idea of incurring the necessary costs to develop an integrated European AI, ranging from infrastructure to applications, does not seem so radical or far-fetched. The recently launched EuroStack initiative is a useful and comprehensive blueprint in this direction. The key idea echoes the claim that AI is a “system technology”, and that sovereignty must be achieved across the whole vertical stack of digital technologies, in a federated manner – that is, following subsidiarity. The proposal highlights the need to develop a common European digital stack: the sovereign components of the stack should be predominantly public, while building on and involving key companies and existing alliances in Europe. The focus on common and public technology is a fundamental requirement for an AI industrial policy that serves European citizens: while supporting the formation of European champions unconditionally risks exacerbating the tension between competition vs industrial policy that we have discussed, fostering the establishment of European Public (and public-private) champions is the way forward for European AI.
A federated system of coordinated investments in AI championed by the public sector is certainly the right direction to take. However, alongside a federated architecture, we need to enable federal competences. This requires a supranational allocation of resources. To put it differently, beyond deep learning, European technological sovereignty in AI requires deep pockets. To be able to match US and Chinese industrial policy interventions, the EU needs to expand its budget. Federal sovereignty is a pre-requisite to unconditional technological sovereignty. In an age of trade rivalries and the looming threat of the arms race over advanced technologies, the EU must start addressing the root of the issue. AI may not be the technological revolution it promises to be, but if unpacking its place within European industrial policy allows us to openly discuss the tensions that are most vital to the EU’s future, then it may be the innovation that sparks a genuine institutional revolution.
*Professor of Economics of AI and Innovation at Université Côte d'Azur, Nice (France)
