The rise of artificial intelligence rests on physical infrastructure: processors, memory, networks, energy and manufacturing plants. Competition is taking place in software and silicon alike.

Why accelerators are essential

Training and running models require an enormous number of calculations to be performed in parallel. Specialised accelerators are designed for these operations and can be far more efficient than a general-purpose processor.

Hardware alone is not enough. Software libraries, development tools and the availability of engineers create an ecosystem that is difficult to reproduce.

Memory and networking become critical

A model must move large quantities of data quickly. Memory bandwidth and connections between chips can therefore limit performance before raw computing power does.

Modern data centres are designed as coherent systems in which computing, memory, storage and cooling advance together.

A question of sovereignty

Manufacturing the most advanced chips depends on an extremely specialised global supply chain. Several regions are investing to secure supplies and strengthen their design or production capabilities.

This strategy will take time: a cutting-edge factory requires massive investment, rare equipment and expertise built up over many years.

On 3 June 2026, the European Commission presented a Chips Act 2.0 proposal to strengthen semiconductor design and production. It remains a legislative proposal: announced industrial capacity is not the same as completed factories.