Europe has spent much of the past decade moaning about the technologies it failed to build itself: cloud, mobile operating systems, social media, SaaS and, most recently, foundation models. Now it is slowly trying to catch up, strengthening its technology ecosystem through initiatives such as Gaia-X, IPCEI-CIS, IPCEI-AI and many more.

But technology moves in cycles. Positions of power are not permanent. They shift as new technology waves emerge, roughly every 10–15 years. If Europe wants to get back into the game, we should spend less time rebuilding what we missed and more time asking what comes next.

And Silicon Valley may already be giving us a hint.

1: AI commoditizes bits

Everyone who has tried Claude Code knows: software is becoming cheaper to build. AI can generate code, understand existing codebases, write tests, translate between frameworks and increasingly execute entire development tasks. That also means legacy systems become cheaper to replace and vendor lock-in easier to break.

Citrini Research describes the same development from an investor perspective. As AI lowers the cost of producing software, traditional software moats come under pressure. Switching costs, established workflows and years of accumulated code become less defensible when competitors and customers can increasingly recreate them with AI.

Their question captures the shift nicely: “What can’t you prompt your way out of?” Software increasingly can be rebuilt. Physical capacity cannot. And that leads us from bits to atoms.

2: But AI needs more and more atoms

At the same time, AI needs the physical world much more than classic software did.

Compute means chips, data centers, electricity, grids and cooling. Chips mean fabs, lithography machines, advanced packaging, specialized materials and chemicals. Robotics adds batteries, motors, actuators, sensors and factories.

A16Z calls this the emerging “electro-industrial stack”: minerals and metals become components, electricity is stored and controlled through batteries and power electronics, motors turn it into motion, and compute and software orchestrate the whole system. AI increasingly connects these layers by turning human intent into software and software into control of physical machines.

In other words: the more capable our bits become, the more atoms we need to make them useful.

3: Leverage shifts towards atoms

Software can be copied and deployed almost infinitely. Physical capacity cannot. You cannot spin up another power plant, semiconductor fab, transformer factory or precision optics supplier with an API call. These things take capital, expertise, physical assets and, most importantly, time.

As bits become cheaper while atoms remain scarce, value and leverage will increasingly shift toward those who control these physical bottlenecks.

And we can already see this happening. S&P Global finds bottlenecks emerging across the AI supply chain: large transformers face lead times of several years, demand for processors and high-bandwidth memory is outpacing capacity, and new fabrication and packaging capacity may not meaningfully ease constraints before 2028. Even less visible inputs such as industrial gases, specialty chemicals and passive components are becoming critical constraints.

The physical becomes the bottleneck

The bottleneck is increasingly not what we can code, but what we can physically build. And this may create an opportunity for Europe.

The software era rewarded placing many bets, iterating rapidly and then scaling the winners globally at almost zero marginal cost. That played beautifully to the American VC model. The emerging physical technology stack works differently. You cannot build 1,000 fabs, power plants or factories and see which one goes viral. It requires long-term capital, engineering expertise, infrastructure, coordination and deep supplier ecosystems.

Europe has substantial capabilities in exactly these areas: industrial automation, machinery, energy infrastructure, semiconductor equipment, advanced manufacturing and a highly specialized Mittelstand. That does not mean that the next technology cycle will somehow automatically favour Europe. But it does mean that we should stop measuring our technology position primarily by the platforms we failed to build in the last one.

Technology leadership is not permanent. Each new wave creates new dependencies, bottlenecks and positions of power. So rather than spending the next decade trying to recreate every layer of the American software stack, Europe should ask a different question: Which scarce capabilities will the next technology stack depend on, and where do we have a realistic chance of becoming indispensable?

That is a much more optimistic technology narrative for Europe. Every cheaper token makes intelligence more abundant. Every coding agent makes software easier to reproduce. But every new data center needs electricity, chips and cooling, and every new robot needs motors, sensors, materials and factories. The technology frontier is not becoming less sophisticated. It is becoming physical again. Europe's opportunity is to own some of its chokepoints.