AI gigafactories require political decisions – not just technological investments

• News
• All news
• AI gigafactories requi...

• Politics
• Artificial intelligence

The EU plans to build massive AI gigafactories, and Denmark has thrown its hat into the ring. But according to a new research brief from CAISA - the National Centre for AI in Society - there is a lack of broad political and democratic debate on the purposes and consequences of such facilities.

Five so-called AI gigafactories - enormous data centres with more than 100,000 advanced chips that can be used to develop and run artificial intelligence. They are on the drawing board in the EU, and Denmark is among the countries that have shown interest in hosting one of the AI factories. But before decisions are made, researchers from CAISA - the National Centre for AI in Society - argue that a more fundamental question should be asked: What exactly do we want to achieve with an AI gigafactory?

This is the central theme of a new brief from CAISA. The researchers point out that wjile the debate often jumps straight to benefits and costs without first clarifying the political choices that lie ahead.

'The question is not merely whether an AI gigafactory is a good or bad idea. There is a lack of clear clarification as to what purposes it should serve and who should have access to it, so that Denmark can engage in a political and democratic conversation at local, national and regional levels about the implications of operating an AI gigafactory in Denmark,' says Matilde Bro, a researcher at CAISA and the Department of Political Science at the University of Copenhagen and co-author of the brief.

More than a technology project

The European Commission's plans form part of the EU's action plan for artificial intelligence and are supported by the InvestAI investment initiative, under which around €20 billion has been earmarked to mobilise the build-up. At the same time, the location, management and specific application remain unclear.

• The EU is planning five AI gigafactories, each with over 100,000 GPU chips.
• Denmark has expressed interest, but the purpose, governance and application remain unclear.
• The brief highlights four crucial dimensions: usage patterns, purpose, business model and governance.
• The choice between AI training and application has a major impact on who benefits from the investments.
• Depending on the model, a gigafactory could have significant implications for energy consumption, the climate and Denmark's understanding of digital sovereignty.

According to the CAISA brief, the discussion should be based on four key dimensions: the usage patterns the gigafactories are intended to support, their purpose, the business model under which they will operate, and how governance and management should be organised.

'AI gigafactories are often presented as neutral infrastructures. But they are anything but neutral. They reflect political priorities and choices that have implications for business, research and society,' says Matilde Bro.

Major consequences, locally and nationally

The brief highlights that there is a significant difference between whether a gigafactory is primarily used to train new AI models or to apply existing models in practice - so-called inference.

'Right now, the demand for deployment is far greater than for training. This is not just a technical issue, but a political choice that determines who actually benefits from the investments,' explains Matilde Bro.

The researchers identify three broad models for how AI gigafactories could be organised, and discuss the economic, societal and political consequences of each model. The aim is not to identify a single correct solution, but to highlight the various scenarios that are rarely discussed openly today.

Regardless of the model, an AI gigafactory will have significant consequences. Among other things, the researchers estimate that a single facility could increase Denmark's total electricity consumption by around 1,5 per cent. Climate impact, local infrastructure and energy consumption are therefore just some of the factors that, according to the brief, decision-makers should actively address.

'When talking about digital sovereignty and technological independence, one must also discuss the local and national consequences. This requires political choices and a broader democratic conversation about which paths to take,' says Matilde Bro.

The brief 'A Danish gigafactory for artificial intelligence?' is written by Matilde Bro and Anders Søgaard for CAISA, the National Centre for AI in Society. Read it here.

Contact

Matilde Bro, PhD fellow
CAISA and Department of Political Science, University of Copenhagen
Email: [email protected]
M: +45 60 85 05 83

Simon Knokgaard Halskov, press officer
UCPH Press Department
Email: [email protected]
M: +45 93 56 53 29

• Politics
• Artificial intelligence

More stories