Plan for seven AI Gigafactories gets under way. Digital Europe must not repeat the inequalities of industrial Europe

There is a deeper reason why that direction is not for us, and it concerns the very nature of the infrastructure. A model trained on hardware owned by a single vendor, powered by a specific energy mix, and hosted in specific jurisdictions, inherits the characteristics of that chain. Infrastructure is not neutral: it is the first form – material and silent – in which a worldview becomes embedded within a machine. The call for proposals has addressed the legal issue – the coordinator must be European, and operational data must remain in Europe – but jurisdiction is the most subtle layer of sovereignty. Beneath jurisdiction lies the hardware; beneath the hardware, energy; beneath it all, the data and the categories used for training. A continent can control the registered office and yet control nothing.

The good news is that the alternative is not to give up. Centralised computing is not a physical necessity: it is a choice of model. Low-communication and geographically distributed training techniques have shown that it is possible to train foundational models on heterogeneous clusters, and even on devices at the edge of the network. This shifts the issue from the realm of ethical aspiration to that of architectural choice: a distributed computing infrastructure – rooted in local communities, shared by a broad and diverse community – becomes technically feasible, not merely morally preferable.

This is where the Harmonic Innovation paradigm has something specific to say. A single intelligence / infrastructure, whatever models it may take, is a threat to the world even when it professes benevolent intentions, not because of what that model actually does, but because of what its position makes possible in principle: political domination, commercial drift, uncorrected systemic error, and inability to adapt to contexts. This is the lesson that Montesquieu had already formulated with regard to political power, and which applies, in exactly the same way, to cognitive power.

Monoculture must be countered by the biodiversity of ideas: many small-scale models, each governed by the community that creates it, each rooted in a tradition, a local area and a particular vocation. Monoculture is productive in the short term but fragile in the long term; a single pest can devastate an entire harvest. Diversity is resilience, and in a global cognitive system it is also the only self-correcting mechanism available: the biases of one model are offset by those of another, the errors of one by the successes of the other.

The Mediterranean can and must be the setting for a possible and necessary response. Europe’s future depends on this sea in this context too. The first reason is physical, and concerns the limits of what already exists. The major digital hubs in the north of the continent – Frankfurt, Amsterdam, Dublin – are congested, and in more than one instance have had to suspend the connection of new users to the network. The Rhine remains Europe’s industrial backbone, but it is a saturated backbone: saturated in terms of land, energy and public support. Nine hundred megawatts of continuous load cannot be added where the grid is already at breaking point. The margins for growth in Europe lie in the South.

The key constraint for a gigafactory is not silicon, but energy: 150 megawatts of continuous power, for years on end. And the energy required is available in the Mediterranean Basin. Paradoxically, this lack of development becomes, in a sense, a competitive advantage. Furthermore, the wind and solar projects currently in the pipeline in the area exceed 360 gigawatts, with two-thirds expected to be operational by 2030; to this must be added the capacity derived from hydroelectric reservoirs; the Italia–Tunisia interconnector is progressing, whilst the European Bank for Reconstruction and Development is funding studies for the link between Egypt and Greece.

A computing system powered by Mediterranean renewables, with data centres designed to return waste heat to district heating systems and microchips optimised for efficiency rather than mass, is not an environmental concession but an architectural choice that reduces structural costs rather than offsetting them retrospectively. In this field, the principle of harmony goes hand in hand with industrial common sense.

The second reason is market-related, and it is the one that the Italian debate seems to overlook with the most stubborn consistency. The demand for computing power set to grow over the next twenty years is not European: it is African and Middle Eastern, where the youngest population on the planet lives and where digital services are being built right now. And the geography of latency is non-negotiable: whoever serves those markets will do so from this side of the Atlantic, or not at all. A continent that locates its cognitive infrastructure where the population is ageing most rapidly, and as far away as possible from the market that is set to grow, is prioritising the present at the expense of the future.

But the decisive reason is another, and it concerns what a model learns. Europe’s competitive advantage does not lie in scale: we have already lost on that front. It lies in stratification. The Mediterranean is the only place on the planet where Arabic, Greek, Hebrew, Berber, Latin and the Romance languages have coexisted for 3,000 years without merging, leaving behind archives, libraries, law, medicine, theology and poetry. No other body of knowledge in the world possesses such a density of cultural layers, and no model trained elsewhere possesses it.

A Mediterranean gigafactory that drew on that heritage would not be a consolation prize for the South: it would be the only European advantage that does not consist of imitating someone else. The rest – the parameters, the processors, the megawatts – can be bought. Stratification cannot.

None of this, however, justifies treating the gigafactory as an end in itself, to be pursued at any cost. A 150-megawatt plant consumes as much energy as a medium-sized town, requires water for cooling from a catchment area that is already treating water management as a structural emergency, takes up land and draws on grid capacity that would otherwise be needed for households and businesses; and, once fully operational, the number of permanent jobs is small compared to its environmental footprint. These are real costs, to be weighed up and assessed beforehand, not justified afterwards.

The conditions are simple to set out but challenging to meet: that each site should generate additional renewable energy rather than simply drawing on existing supplies; that it should be located on brownfield sites rather than on greenfield land; that it adopts cooling systems with low water consumption; that it reserves a share of capacity for the host community and provides a verifiable return in the form of training and research. A facility that extracts energy, water and consent from a local area in order to export its value elsewhere is the old extractive model in digital form, and the fact that it produces intelligence rather than concrete does not make it any less extractive.

This is the direction in which Colnissar is moving – the sapient artificial intelligence initiative developed within the Harmonic Innovation Ecosystem. We are writing about it here not to propose it as a solution – that would be precisely the contradiction that the principle of plurality prohibits, because Europe must not equip itself with a single sapiential intelligence, nor can a single institutional model adapt to such diverse legal and cultural contexts – but because this example serves to demonstrate that the arguments set out so far can be translated into verifiable architectures: distributed ownership, non-profit development tied to a mission, structural openness of weights, dedicated and optimised computing capacity, and the rejection of indiscriminate scaling. Other initiatives are possible, and are desirable. Plurality is part of the proposal, not merely a stylistic flourish.

After all, the decision does not concern a facility but a plan. Today in Europe there are four networks that are unaware of one another. The network of innovation ecosystems – Málaga, Casablanca, Rabat, Tangier, El Ghazala, Berytech, Sophia Antipolis, and Southern Italia – which encompasses the regions and businesses. The network of universities, which has researchers and languages at its disposal, and which, within the Mediterranean Pact, has been given a framework – the Mediterranean University Initiative – that has yet to be filled. The network of science parks and technology transfer, which possesses the expertise to take a result from the laboratory to the market and is currently the weakest link. And the network of gigafactories, which will possess computing power.

As long as there are four separate policies, with four timetables and four directorates-general, they will produce four partial results: ecosystems lacking power, universities with no industrial links, technology transfer without the necessary knowledge base, computing capacity sold to the highest non-European bidder.

They must become a single infrastructure, with a single mandate and a single governance structure: not four coordinated programmes, but a continuous chain stretching from basic research to the trained model, with the Mediterranean at its centre rather than on its periphery. It is this, and not the number of sites, that is at stake in a new Mediterranean and European leadership.

If one is looking for a place where those four networks are already, at least in part, co-present, two names stand out above all others: Sicily and Calabria. Not out of any southern regionalist sentiment, but because conditions converge there that are found separately elsewhere in Europe. Sicily is the point where the Mediterranean narrows: almost all the cables connecting Africa to Europe pass through its Strait, Palermo has for years been the region’s key internet hub, the interconnection with Tunisia will come ashore there, and the Tyrrhenian Link is connecting it with 1,000 megawatts per branch to Campania and Sardinia, whilst the connection with Calabria has been operational since 2016. Meanwhile, in Catania, STMicroelectronics is building the world’s first fully integrated silicon carbide plant: this is no minor detail, as silicon carbide is precisely the technology that enables efficient power conversion in data centres. Energy, data and components all in one place: this is the definition of a supply chain.

Calabria adds two elements to this picture. The first is physical: electricity generation that has historically exceeded regional demand, supported by wind power in the Crotone area and the hydroelectric reservoirs of the Sila, and a supply of industrial sites that no longer exists elsewhere. The second is institutional. On 24 September, the Region signed an agreement with the Department for Digital Transformation of the Prime Minister’s Office and the University of Calabria that brings together artificial intelligence, quantum computing and high-performance computing, digital healthcare, satellite connectivity for inland areas and – the point of greatest interest here – the mapping of sites suitable for hosting data centres, with an assessment of their technical, energy, environmental and urban planning sustainability. This is the preliminary task that almost no other Italian region has yet carried out, and it is the prerequisite for being well-prepared when the game really gets underway.

The first request concerns Italy’s bid. Having a site in the South is necessary but not sufficient, because hosting is not the same as owning, and a region that receives machinery without acquiring the expertise to manage it remains nothing more than a colony with a pleasant climate.

The proposal must explicitly state a Mediterranean focus: a share of computing capacity reserved, by statute and not merely as a courtesy, for universities and research centres on both sides of the Mediterranean, building on the Mediterranean University Initiative established by the Pact, which is still awaiting substantive implementation.

The second concerns the Essential Levels of Innovation, and puts them to the test in relation to the one asset that really matters. If access to computing power is now a prerequisite for scientific citizenship, then it cannot be a privilege determined by geography or budget: it must be an enforceable, measurable and verifiable threshold. A right to computing, defined in the same way as essential service levels are defined, is the most concrete way to prevent digital Europe from repeating the unequal geographical distribution of industrial Europe.

The third concerns timing, and here the window of opportunity is narrow. The four billion for the second phase depends on the 2028–2035 budget, which is currently being negotiated; the second edition of the Action Plan for the Pact is expected in the coming weeks. These are two forums that are currently not communicating with one another, yet which need each other: without the Mediterranean dimension, the European plan is flawed from the outset, and without that plan, the Pact remains merely a declaration of friendship between the two shores. Aligning the two timetables is not a procedural detail: it is the only political move that can currently produce something irreversible.

In the previous chapter of this reflection, we had referred to Cassiodorus’ *Vivarium*, and the network of scriptoria that sprang from that hub, spanning centuries that might otherwise have erased ancient knowledge. It is worth recalling how that network prevailed. It was not the largest, nor was it the richest; it lacked the scale of the powers that surrounded it. It was the most widespread, the most deeply rooted and the most determined not to be interrupted. The computing power that Europe will bring online over the next two years will determine far more than an industrial ranking: it will determine what memory will be embedded in the machines that will think alongside us. It is wise to choose where to bring it online, and it is wise to do so now.

Francesco Cicione, chairman and founder of Entopan

Sign up to the NextMed newsletter so you don’t miss out on previews and in-depth articles