AI infrastructure.

AI infrastructure is the full stack on which artificial intelligence is trained and run. At its base are accelerators and specialised chips, housed in data centres and connected by high-bandwidth networking. Around them sit the systems that keep them working: power supply, cooling and physical security. On top run the software layers — orchestration, cloud platforms and model platforms — through which organisations actually use the capacity.

What has changed is scale and weight. Training and operating advanced models now requires dedicated facilities rather than spare capacity in general-purpose data centres. As a result, AI infrastructure has become the physical foundation of the AI economy: whoever controls compute, and the power and land it depends on, shapes who can build, at what cost and on whose terms.

For several years the scarce input was the chip. Increasingly, it is electricity. A large AI campus needs continuous, substantial power and the means to remove the heat it produces. Grid connections can take years to secure, energy prices vary sharply between regions, and permitting for new generation, transmission and buildings moves at the pace of public process rather than product cycles.

Cooling adds its own conditions, from water availability to climate. Together these factors now decide where compute can be built, often more decisively than access to accelerators. Site selection has become an energy question first, and operators, utilities and governments find themselves negotiating over the same scarce capacity. Regions with surplus generation, cool climates and predictable regulation gain an advantage that is difficult to replicate quickly, while others face waiting lists for connection that no amount of capital can shorten on its own.

Europe has world-class research and a strong industrial base, but comparatively little large-scale compute under its own control. Much of the capacity European organisations rely on is owned and operated elsewhere. Closing that gap depends less on ambition than on fundamentals: affordable and reliable energy, political and legal stability, and credible data protection.

Switzerland brings several of these qualities together, including a stable legal environment and a long tradition of trusted data handling. The question for the continent is closely tied to sovereign AI: capacity located and governed at home is the precondition for keeping essential systems independent of decisions taken in other jurisdictions.

AI infrastructure increasingly attracts long-duration infrastructure capital alongside venture and corporate investment. Yet it combines two very different kinds of asset. Land, grid connections, power and buildings are long-lived and behave much like traditional infrastructure. Accelerators and memory depreciate quickly, with each new hardware generation reducing the value of the last.

Investors therefore ask precise questions: how much of the capacity will be used, by whom and under what contract length; how quickly the equipment becomes obsolete; and which parts of a project retain value if demand shifts. Structuring around that mismatch — separating durable assets from short-lived ones — is becoming a discipline in itself. So is matching capital to risk: patient investors for the land and power, and shorter-horizon structures for the hardware that must be refreshed.

AI infrastructure is one of the four themes of Davos Catalyst, alongside enterprise AI, sovereign and defence AI, and AI and capital. The summit brings together operators, investors, corporate buyers of capacity and researchers — the people whose decisions determine where compute is built and how it is financed.

The format favours private one-to-one meetings and deal rooms over presentations. Davos Catalyst takes place from 12 to 15 January 2027 in Davos, in the week before the World Economic Forum Annual Meeting. It is an independent, privately hosted event and is not affiliated with the World Economic Forum.