From gigawatts to your desk: the sovereignty pyramid at every scale
Sovereignty in the AI era is measured in gigawatts and fabs, not in bytes and regulation. The claim comes from Nina Schick’s keynote on “Industrial Intelligence” earlier this month, and one number from the keynote has refused to leave: Europe builds the only machines in the world that can print leading-edge chips, and in the last quarter 99% of ASML’s system sales shipped to Taiwan, South Korea, China and the United States. Europe’s own share does not show up in the numbers at all.
Her argument was made at continental scale. The question this post tries to answer is whether the same logic holds two levels down: for a company, and for the infrastructure on your own desk.
Industrial Intelligence and the sovereignty pyramid
Schick’s core claim is that AI is not a software story but an industrial one. Data centres, chips and energy grids now matter the way oil pipelines did in the last century, and whoever controls them sets the price and availability of intelligence for everyone else. Her shorthand: who owns the compute wins the future.
She draws this as a pyramid. The apex is what every country actually wants, prosperity and security. Everything below it is what you need to own, or at least control, for the apex to be yours rather than borrowed.
- Energy at the base. AI is the most energy-hungry technology ever built. No cheap, abundant, controllable electricity, no sovereign anything above it. In Europe the bottleneck is grid capacity rather than generation.
- Compute: chips, data centres and the networking between them, on your own territory and under your own jurisdiction. Renting hyperscaler capacity that answers to foreign law is access, not sovereignty.
- Models, trained and run on that compute. Otherwise you train on someone else’s terms and timeline.
- Applications, where the economic value is realised and where most European companies and governments actually operate.
- Prosperity and security at the top: the outcome, not a layer you can build directly.
Her critique of Europe is that it starts at the top. It regulates, it builds applications, and it assumes the base will be available when needed. In a geopolitical squeeze that assumption fails first.
The supply chain of intelligence
The pyramid tells you what to own. The supply chain tells you where it can be cut. Schick walks the same stack as five layers, and for each one the question is identical: who controls it, where is the choke point, and what is our dependency?
The two views line up with one shift. Energy, the base of the pyramid, becomes one of the raw materials at the bottom of the chain. The chain also names a fifth layer the pyramid leaves implicit: the strategic leader who fits the other four together. Prosperity and security stay what you get when the strategic leader gets that right. The rest of this post uses the five layers of the chain.
| Layer | What it is | Where the choke point sits today |
|---|---|---|
| 1. Raw materials | Wafers, rare earths, gallium, germanium, specialty chemicals, energy inputs | China dominates rare-earth refining and has already used gallium and germanium export controls as leverage; Japan dominates photoresists |
| 2. Compute | Chip design, fabrication, lithography, memory, packaging, data centres | Nvidia on accelerators; TSMC on leading-edge fabs; ASML on EUV lithography; grid connections for data centres |
| 3. Models | Frontier models and the ability to train, fine-tune and run them | A handful of US labs plus Chinese challengers; Europe subscale |
| 4. Applications | Products and workflows built on the layers above | Most of Europe lives here, on APIs and cloud it does not own |
| 5. Strategic leader | The actor who integrates the four layers into one strategy | The US and China act as strategic leaders; Europe has mostly acted as regulator and customer |
The fifth layer is the one that makes the model useful beyond geopolitics. A strategic leader is not a fab or a model; it is whoever decides what to own, what to multi-source and what dependency to accept. The strategic-leader role exists at every scale.
Europe’s paradox: ASML on our soil, machines shipped elsewhere
The Netherlands holds the single most valuable choke point in the entire chain. ASML in Veldhoven is the only company in the world that builds EUV lithography machines, the tools without which no leading-edge AI chip gets made. On paper that is a sovereignty asset few countries can match.
The shipping data tells a different story. In Q3 2026 ASML sold 121 lithography systems for a record €11.7 billion in total net sales (ASML Q3 2026 results). By destination, Taiwan took 32% of system sales, South Korea 31%, China 23% and the United States 13% (regional ship-to shares). Those four add up to 99%, and the last 1% went to the rest of Asia. Europe, the continent that builds the machines, does not appear in the split.
Schick put it more bluntly on stage: no European country or company ordered a single ASML machine last quarter. That exact claim cannot be checked, because ASML stopped publishing quarterly order bookings in 2026. The shipment shares make the same point with public numbers, and arguably a sharper one: this is not a one-quarter blip but the structure of where chips are made.
So Europe owns layer two’s crown jewel and converts almost none of that asset into chips, models or applications on its own soil. The machines leave, the chips are made elsewhere, the models are trained elsewhere, and European companies rent the result back through an API. Europe has built the pyramid from the top down.
Scaling the pyramid down: the enterprise
Most boards still treat AI as a software question for the CIO. Run the five-layer check on a typical European company and the shape of the problem changes.
| Layer | National version | Enterprise version |
|---|---|---|
| Raw materials | Minerals, energy inputs | Talent, data, capital, contracts: the inputs you cannot conjure |
| Compute | Fabs, data centres, grid | Owned or colocated hardware versus rented cloud, and which jurisdiction that cloud answers to |
| Models | Frontier labs on own soil | Models you can run and fine-tune yourself versus dependency on one vendor’s API |
| Applications | Domestic software industry | Products and workflows, and how much of their core logic sits on someone else’s platform |
| Strategic leader | Government and industrial policy | A board that treats AI as infrastructure and dependency management, not tooling choice |
Most companies will never own a fab or a data centre the way TSMC, Samsung, Tesla or Amazon do. The point is knowing your exposure at each layer. Three questions do most of the work: what happens to your product if your model vendor doubles prices, if your cloud region falls under foreign sanctions, or if export rules change overnight? If the honest answer is “we stop”, you have built an application on a base you do not control.
The practical response is the same one Schick asks of nations: keep data and core logic portable, multi-source at the model layer, and make the decision about what to own a deliberate one rather than a default.
Scaling the pyramid down: your own desk
The last step down is one person at one desk. The scale shrinks; the logic does not.
At personal scale the raw materials change. Energy is still one of them (a machine at home draws power you pay for), but the inputs that matter most are time, attention and control over your own data. From there the other four layers map cleanly.
- Compute: a machine you own and can reach when the internet, the vendor or your account is not cooperating. A Mac mini, a workstation with a GPU, even a Raspberry Pi for the right workload. Versus: everything through a browser tab on someone else’s server.
- Models: an open-weight model that runs locally for anything private or critical, with frontier APIs for the rest. Versus: one subscription, one vendor, one terms-of-service change away from losing a workflow.
- Applications: automations and tools whose logic and data live with you, exportable in plain formats. Versus: a dozen SaaS products, each holding a slice of your work hostage to its pricing page.
- Strategic leader: you, deciding what to own, what to rent and what to accept. Nobody else is going to do that for your infrastructure.
This is not an argument for going off-grid. Frontier models are better than anything that runs on a desk, and most of the time renting is the right call. Sovereignty at personal scale means something narrower: never being locked out of your own work. Own the data, own a machine that can run a fallback, keep the skills that do not depend on one vendor’s interface.
The parallel with Europe is uncomfortable. Many of us hold valuable assets (expertise, data, years of accumulated work) and run them entirely on infrastructure we do not control. We are ASML in Veldhoven, shipping every machine abroad.
Why Aurono builds local-first
This is the thinking behind every Aurono product. Aurono Start automates your investing on hardware in your own home, a Raspberry Pi or a Mac, with your own keys and your own data; the exchange is a counterparty, not a landlord. Aurono Admin keeps a Dutch entrepreneur’s finance administration on infrastructure they control, with the cloud as an option rather than a requirement. Aurono Atlas fetches all your crypto investment data, stores the data locally and aggregates the data into an easy-to-monitor crypto portfolio. Start, Admin and Atlas each refuse to build the application layer on a base the user does not own.
Schick’s thesis, stated at the scale of continents, is that prosperity and security follow from owning the layers beneath them. The same sentence is true with “continent” replaced by “company”, and true again with “company” replaced by “you”. The only thing that changes is how many gigawatts it takes.
So the closing question is the one worth sitting with: if the sovereignty pyramid holds for countries and companies at large scale, why would it stop holding for your personal infrastructure?
Sources
- ASML Q3 2026 results, 14 October 2026: €11.7 billion total net sales, 121 systems sold
- Regional ship-to shares, Q3 2026: Taiwan 32%, South Korea 31%, China 23%, United States 13%; ASML no longer reports quarterly bookings
- Nina Schick, Industrial Intelligence: sovereignty measured in gigawatts and fabs