³Ô¹ÏºÚÁÏÍø

Generative artificial intelligence (GenAI) models, such as ChatGPT, have become a general-purpose technology with applications in every economic activity. But artificial intelligence (AI) model training costs are growing exponentially and will reach billions of euros in the next years. Global AI computing infrastructure costs may reach a trillion euros. However, productivity gains from AI are growing at a much slower place because they take more time. Unless new technologies can slow down the cost explosion or accelerate productivity growth, this may put a brake on AI innovation.    

The European Commission’s ‘AI Factories’ initiative and Draghi’s AI proposals suggest that the EU aims to catch up with the US by upgrading existing EuroHPC supercomputers. This hardware focus, at the expense of a wider economic view, has characterised EU digital policies for decades. In a forthcoming Policy brief, we show how this is especially misguided in AI. EuroHPC computers are not designed for GenAI models. Upgrading to competitive standards is simply beyond the financial capacity of EU budgets. They lack access to large-scale business models that can generate the revenue required to amortise the huge cost of frontier AI models.  

The EU can prosper below the AI technology frontier. It can use existing open-source and commercial state-of-the-art AI models to produce derived, specialised models and AI application services at much lower costs. They can be rolled-out across sectors and embedded in smaller EU companies with sufficient revenue to bear the investment costs. This private sector approach would contribute more to accelerating EU productivity growth than betting taxpayer money on a few super-expensive supercomputers and AI models. It would give the EU time to address missing or poorly functioning complementary markets, including private equity for AI start-ups and put its regulatory house in order with a razor-sharp focus on competitiveness and innovation.  

Read the working paper by Bertin Martens on the exploding costs of AI model development and keep an eye out for the forthcoming Policy brief on how the EU should position itself in the AI technology race, coming next week.

The Why Axis is a weekly newsletter distributed by Bruegel, bringing you the latest research on European economic policy.