The rapid rise of artificial intelligence has become one of the most powerful forces shaping semiconductor manufacturing. Training and serving large models requires vast fleets of accelerators, and each of those chips depends on the most advanced silicon a foundry can produce. As a result, decisions about where and how much fab capacity to build are increasingly being made with AI workloads in mind.

Why AI changes the investment maths

Traditional demand for chips came from a broad base of consumer electronics, communications and industrial products. AI concentrates demand differently. A relatively small number of very large, very expensive processors now account for a growing share of leading-edge wafer starts. This pushes foundries to prioritise their newest process nodes, where performance-per-watt gains matter most for data-centre economics.

The knock-on effects extend well beyond the logic die itself. High-bandwidth memory, advanced packaging such as 2.5D and 3D integration, and sophisticated interposers have all become bottlenecks. Investment is therefore flowing not only into wafer fabs but into packaging and test capacity that was once treated as an afterthought.

A more concentrated capital landscape

Because leading-edge fabs cost tens of billions to build and equip, only a handful of companies can sustain the pace. AI demand reinforces that concentration, since customers want assurance of long-term, high-volume supply. At the same time, governments in the United States, Europe and Asia are offering incentives to broaden where this capacity sits, wary of over-reliance on any single region.

• Leading-edge logic capacity is being prioritised for accelerators.
• Advanced packaging and high-bandwidth memory are now strategic constraints.
• Public subsidies aim to diversify the geography of new fabs.

The mature-node counterweight

It would be a mistake to assume every new fab chases the smallest transistors. AI systems still need power management, networking, sensors and interface chips built on mature nodes. Investment in these established processes continues, giving the market a more balanced footing than headlines about cutting-edge accelerators might suggest.

For engineers, this reshaped landscape is a reminder that opportunity spans the whole stack, and building strong design fundamentals is the surest way to stay relevant as the industry evolves.