Sweden Didn't Attract AI Infrastructure. It Got Colonized By It.

Every time a foreign company announces a new data center in Sweden, someone in government gets a press release ready within the hour. Jobs. Investment. Sweden as an AI superpower. I've read four of these releases this month alone. None of them mention who owns the building, who owns the power contract, or where the depreciation gets written off. That's the part that actually matters, and it's the part nobody wants to talk about.

Here's the pattern from this week alone. HIVE Digital Technologies, a Canadian company that started as a crypto miner and rebranded into "AI infrastructure," is buying the Boden data center it has rented for eight years and locking in a ten-year lease backed by $115 million in new notes. CoreWeave, American, is expanding into Sweden through Conapto. Google breaks ground on new capacity and throws in a EUR 5 million "skills fund," a rounding error dressed up as generosity. Even Mistral, the company Europe keeps pointing to as proof we can build sovereign AI, runs on EcoDataCenter's Swedish grid and cooling to actually function.

Notice what's Swedish in all of this. The hydro. The cold air. The political stability. The grid. Everything expensive and hard to replicate. Notice what's foreign. The capital, the racks, the lease structures, the equity, the upside. This is the exact same model Sweden has run for a hundred years with iron ore and timber. We had the raw material, someone else had the capital and the market access, and the margin walked out the door in a shipping container. Now the raw material is electrons and the container is a fiber line to Frankfurt.

Software development Sweden should be building sovereign wealth, not renting it out

I run a company that does custom SaaS development and AI work out of Jönköping. I talk to founders across the Nordics who want to build here, hire here, and keep ownership here. What they're discovering is that "AI infrastructure investment" and "Sweden getting richer" are not the same sentence anymore. They used to be. Now they're diverging fast.

When a Swedish startup wants GPU capacity, they're renting compute that sits on Swedish soil but is priced, controlled, and profit-optimized by a company headquartered in Toronto, Mountain View, or increasingly Paris. The startup pays EU-sovereign prices, meaning GDPR compliant, EU-jurisdiction, everything the brochure promises, but the actual economics of that transaction, the chips, the model weights, the platform margin, all of it gets extracted straight back out. Sweden becomes the colocation colony. We provide the room temperature and the rule of law. Somebody else keeps the profit.

This matters more for software than people realize. If you're doing AI solutions work, your unit economics are downstream of whoever owns the compute layer. A Canadian miner-turned-infra-company setting lease terms in Boden has more influence on what a Stockholm AI startup can charge its customers than the Swedish government does. That's not sovereignty. That's a rental agreement with better branding.

Sweden versus the rest of the world: everyone's doing this, but not everyone's losing equally

The US does this to other countries. It doesn't do it to itself, because the capital, the chips, and the model labs are American. Nvidia, OpenAI, Anthropic, Google, Microsoft: the entire stack from silicon to foundation model sits inside US borders or under US corporate control. When American data centers get built in Texas or Virginia, the extraction happens inside the same economy. The value stays home even if the land doesn't.

Asia is more mixed but instructive. Japan and South Korea protect their chip and memory industries with real industrial policy, not press releases. Taiwan controls the most important choke point in the entire AI stack, TSMC, and everyone knows it, and everyone treats Taiwan differently because of it. China is building its own compute stack specifically so it never has to depend on foreign infrastructure for anything that matters. Whatever you think of the politics, that's a country that understands the difference between hosting infrastructure and owning it.

Europe, and Sweden specifically, is doing neither. We're not protecting a domestic compute stack like Asia, and we're not the origin point of capital and IP like the US. We're the middle ground: stable enough to trust, cheap enough to exploit, regulated enough to feel safe about it. Dagens Industri covers every new data center announcement as an investment win. SVT runs the jobs angle. Almost nobody in Swedish business media is asking who holds the equity five years from now when these facilities are fully depreciated and generating pure margin for shareholders in another country.

The EU's answer to all this has been the AI Act, which regulates model risk and deployment, and says almost nothing about who owns the physical and capital layer underneath European AI. Brussels built a rulebook for how AI systems behave. It didn't build a strategy for who owns the infrastructure those systems run on. That's a massive gap, and it's the gap foreign capital is currently walking straight through, on schedule, on Swedish soil, this month.

Where this goes over the next five years

This gets worse before it gets better, and here's the mechanism. As AI models get more capable, compute demand doesn't plateau, it compounds. Every serious lab is racing toward larger training runs and denser inference workloads. That means more data centers, more power contracts, more land, more grid capacity, all concentrated in places with cheap renewable energy and political stability. Sweden checks every box. So does Norway. So does parts of Finland. We are, structurally, going to keep getting more of these facilities, not fewer.

The question is whether Sweden uses this window to build actual sovereign capability, or whether we let five more years of "record investment" headlines paper over the fact that we own less of our own AI stack every year. If AGI-level systems arrive in the next three to five years, which is the honest timeline most serious lab researchers are now working toward, the entities that own the compute those systems run on will hold leverage that dwarfs anything a government can regulate after the fact. You don't get to nationalize your way out of a dependency you spent a decade building for someone else.

What should actually happen: Sweden and the EU need an industrial policy for the compute layer itself, not just the model layer. That means public or co-owned compute infrastructure, tax structures that reward equity retention over pure land-and-power leasing, and serious investment in domestic chip design even if we'll never compete with TSMC on fabrication. Norway's sovereign wealth fund model exists for exactly this kind of resource. Nobody has proposed a Nordic sovereign compute fund yet. Someone should.

What to look at if you're building instead of complaining

If you're a founder or CTO reading this and thinking "fine, so what do I actually do," here's where I'd point you.

Self-host what you can. The awesome-selfhosted list is the best single resource for reducing your dependency on rented infrastructure you don't control. Every workload you move off a foreign-owned platform and onto infrastructure you actually own is a small act of sovereignty. It adds up.

Build your automation and AI agent layer on tools you can run yourself. n8n and Dify both let you build serious agentic workflows and RAG pipelines without renting your entire logic layer from a platform that can change pricing overnight. We use patterns like this in our own AI agents work. Owning your orchestration layer matters more than people think, because that's the layer that's cheapest to keep sovereign and most expensive to lose control of later.

Own your compliance and observability stack. Netdata gives you full-stack observability without shipping your telemetry to a foreign SaaS dashboard. Small decision, same principle.

If you're building on-chain infrastructure or tokenized assets, the same extraction logic applies. Don't build your value layer on rails you don't control. This is exactly why our blockchain development and smart contract development work always starts with the question of who actually holds custody and control, not just who's hosting the frontend.

What Swedish founders should actually do this quarter

Stop treating "AI infrastructure investment" headlines as good news by default. Read the ownership structure before you celebrate. When you're picking a cloud or compute partner, ask directly: who owns the depreciation schedule, and what happens to my pricing when they've fully written off the asset and are pure-margin harvesting. If you're building a product that depends on foreign-owned compute, build an exit plan into your architecture from day one, not after a pricing shock.

And if you're a founder trying to get to market fast without over-indexing on any single infrastructure dependency, that's a build problem, not a policy problem, and it's solvable now. We do a lot of this through rapid MVP work specifically so founders can prove the model before locking into infrastructure commitments they can't unwind later. Architecture decisions made in month one are the ones that determine whether you're renting your margin away in year three.

Sweden has the cold air, the clean power, and the trust. Those are real assets. The mistake is thinking that's the whole game. The whole game is who owns the thing sitting in the cold air using the clean power. Right now, mostly, it's not us.

Fredrik Brunnberg is the CEO of HEIMLANDR.IO, building AI and software solutions from Jönköping, Sweden. This is the daily HEIMLANDR briefing. If you found this valuable, share it with someone who builds things.