Sweden Won the Colonial Auction, Not the AI Race

HIVE Digital Technologies, a Canadian crypto miner, just bought the 32MW Big Boden Data Center after 8 years of renting it. Read that again. A foreign company spent eight years leasing Swedish power and cooling, decided it liked the returns so much it bought the building from its own landlord, and everyone is calling this a Swedish infrastructure success story. It is not. It is a tenant deciding the lease terms were so good they should own the building outright. That is not a Swedish win. That is a Swedish exit.

I run HEIMLANDR out of Jönköping. We build AI agents, custom SaaS, and blockchain infrastructure for clients who actually want to own what they build. So when I read the last two weeks of Swedish data center headlines, I do not see a boom. I see an auction. And Sweden is not the buyer.

The Pattern Nobody Wants to Name

Let me lay out what actually happened in the last month, because the framing in most Swedish business press has been embarrassingly celebratory.

Google broke ground in Horndal and attached a €5 million "skills fund" to the announcement. Five million euros. For a company sitting on a market cap that makes that number a rounding error inside a rounding error. It is the corporate equivalent of a landlord handing you a fruit basket while raising your rent. The skills fund is not investment in Swedish capability. It is a PR insurance policy against the headline "Google extracts Swedish power for free."

CoreWeave partnered with Conapto. Mistral, the French company everyone in Brussels calls "Europe's answer to OpenAI," partnered with EcoDataCenter for a billion-dollar infrastructure push. Read the actual structure of these deals, not the press release. Sweden supplies the grid connection, the cold climate that makes cooling cheap, and the political stability that makes insurers comfortable. The model weights, the IP, the revenue from inference, the actual value being generated on top of that hardware? That flows to Paris, to Delaware, to wherever the parent company's tax lawyers decided was optimal this quarter.

This is not a Swedish AI infrastructure boom. It is Sweden becoming the best-reviewed colony in the AI empire. We have great weather for servers and terrible instincts for negotiating what we get in return.

"Sovereignty" Is Doing a Lot of Work in These Press Releases

Every one of these announcements uses the word "sovereignty" somewhere. Mistral gets called Europe's sovereign AI answer to American hyperscalers. Fine. But swap "American flag" for "French flag" and ask what actually changed for a Swedish CTO. Nothing. You still do not own the model. You still do not own the infrastructure decisions. You still do not capture the margin. You have just changed which foreign capital city controls your dependency.

Real sovereignty is not about which country's passport is on the building. It is about who can turn the thing off, who sets the price, and who captures the upside when the thing works. On every one of those three questions, Sweden currently answers "someone else." Swap the flag, keep the colony.

Sweden vs. the World: Who Actually Owns What They Build

Look at how this compares globally and the pattern gets uncomfortable fast.

The United States has its problems, plenty of them, but American hyperscalers own the full stack: the chips, the model labs, the data centers, and increasingly the power generation feeding them. Microsoft is buying nuclear. Amazon is buying nuclear. When Americans build data centers, American companies usually own the thing sitting inside them too.

China does the same thing domestically, aggressively, with state backing that makes Swedish industrial policy look like a suggestion box. Nobody is renting China's grid to a Canadian crypto miner.

France just built a national champion in Mistral specifically so it would not be in Sweden's position, dependent on someone else's model IP. And now Mistral is using Swedish infrastructure to power that French sovereignty play. Sweden supplied the cooling for someone else's independence.

Norway, worth noting, has been smarter about this with its state ownership stakes in energy infrastructure tied to data center deals. Iceland has been more disciplined about demanding local economic return, not just construction jobs, in exchange for geothermal-cooled compute. Sweden, by contrast, treats every foreign data center announcement as an unambiguous economic win regardless of what we actually retain. Business press writes it up as regional development. Politicians cut ribbons. Nobody in the room asks who owns the compute five years from now, or what happens to Norrland when the crypto price cycle turns or the AI capex bubble deflates and the miner or the hyperscaler decides Sweden's grid is no longer the cheapest option on the map.

The CTO Problem: Optimizing for the Wrong Balance Sheet

Here is the part that actually matters for the audience reading this. If you are a Swedish CTO cheering these deals because your municipality gets tax revenue and construction jobs, you are optimizing for 2026's balance sheet. You are not thinking about 2030's dependency structure. Every one of these facilities makes Sweden more load-bearing in someone else's AI stack and less capable of running its own. That is fine if you believe AI infrastructure is a commodity, interchangeable, and Sweden's only job is to be the cheapest rack space in Europe. I do not believe that. I think compute ownership in five years determines who captures margin, who sets terms, and who has actual leverage when AGI-adjacent systems start running critical national infrastructure, defense logistics, healthcare triage, financial systems. You do not want to be renting the machine that makes those decisions. You want to own it, or at minimum, control the software layer sitting on top of it.

Where This Actually Goes

Fast forward three to five years. The current AI capex boom, the one funding every gigawatt data center announcement from Boden to Horndal, is being financed on the assumption that inference demand keeps compounding forever. Maybe it does. Maybe it does not. Either way, the physical infrastructure stays in Sweden regardless of who wins that bet. What changes is who is left holding it.

If the AI boom keeps compounding, Sweden becomes permanently structurally subordinate: supplying power and cooling to foreign-owned compute while capturing construction jobs and modest tax revenue, forever a backend, never a front end. If the boom cools, or if a cheaper compute geography emerges, Swedish municipalities are left holding half-empty megawatt facilities built for someone else's business model, with none of the software IP or customer relationships that made the compute valuable in the first place. Boden has lived this cycle before, with Facebook's data center in 2013 promising thousands of jobs and delivering a fraction of that, mostly automated.

The regulatory gap here is real and it is Brussels' problem as much as Stockholm's. The EU AI Act focuses heavily on model risk classification and deployment transparency. It says almost nothing about infrastructure sovereignty, about who should own the compute layer that European "sovereign AI" actually runs on. Mistral gets celebrated in EU policy circles as proof Europe can compete without American hyperscalers, while quietly building its infrastructure dependency on Swedish grid capacity owned and operated by companies incorporated elsewhere. Nobody in Brussels has asked the obvious question: what is actually sovereign about renting cooling from a country that does not own the compute either?

Swedish policy needs to start attaching real strings to these deals. Not skills funds that amount to rounding-error PR. Actual equity stakes for the state or municipality. Actual requirements that some percentage of compute capacity be available to Swedish and Nordic companies at cost, not market rate, in exchange for grid access and tax breaks. Norway does versions of this with energy. Sweden has the leverage, cheap power, cold climate, political stability, EU market access, and is not using it. We are giving away scarce resources for construction jobs and press releases.

What Builders Should Actually Prepare For

If you are running engineering or product in a Nordic company right now, the lesson is not "boycott foreign data centers." That is naive. The lesson is: do not build your company's core dependency on infrastructure you do not control the terms of. If your AI roadmap depends entirely on a foreign hyperscaler's API pricing and availability, you have the same structural problem as Sweden's grid, just at company scale. You are renting your future to someone else's decisions.

The counter-move is owning your software layer even when you rent the hardware underneath it. Self-hosted, open, portable infrastructure that does not lock you into one vendor's roadmap or pricing whims. This is not ideology, it is risk management.

What to Look At

If you want to build compute independence into your own stack instead of inheriting someone else's dependency, start here:

• Kubernetes remains the baseline for running workloads portably across any cloud or bare metal, which matters more when you are trying not to get locked into a single hyperscaler's terms.
• n8n for workflow automation you can self-host, so your AI agent logic does not live entirely inside a vendor's walled garden.
• Dify for building agentic and RAG pipelines you can deploy on your own VPC instead of renting the entire stack from one provider.
• awesome-selfhosted, the largest curated list of software you can run yourself instead of subscribing to someone else's dependency forever.

None of these fix Sweden's grid ownership problem. That requires actual policy. But at the company level, they are how you avoid being the Swedish grid of your own industry, cheap, reliable, and permanently rented out to someone else's ambition.

What to Actually Do About This

If you are a Swedish founder or CTO reading this, stop treating "we got a data center" as an economic development win by default. Ask what percentage of compute capacity your region actually retains access to. Ask who owns the model layer running on top of the hardware. If the answer is "nobody local," you have a press release, not an asset.

If you are building AI products, the strategic move right now is owning your software and data layer even while renting hardware. That is where actual margin and actual defensibility live in 2026, not in the rack space. This is exactly the gap we work in at HEIMLANDR: AI solutions built to run on infrastructure you control the terms of, not infrastructure that controls you. If you are building tokenized assets or need real smart contract development instead of a whitepaper and a prayer, the same principle applies: own the logic layer, rent the compute underneath it, never the reverse.

And if you are evaluating vendors for software development in Sweden right now, ask them the same question I am asking the Swedish government: what do you actually own when this contract ends? If the answer is nothing, you already know what kind of deal you signed.

Sources worth your actual attention on this, not just mine: Breakit has been tracking the data center land grab in Norrland closer than most, and Dagens Industri covered the HIVE Big Boden acquisition with more skepticism than the press releases deserved credit for.

I am not against foreign investment. I am against calling dependency a victory because the ribbon-cutting photo looked good. Sweden has genuine advantages here: cold climate, stable grid, political trust, EU access. We should be negotiating from that strength, not giving it away for construction jobs and a fruit basket. Build the thing you own. Rent the thing you do not need to. Know the difference before you sign.

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.