Sweden Didn't Win the AI Race, It Just Became the Rack

Four headlines in one week. Mistral AI teaming up with EcoDataCenter on a billion-dollar build. CoreWeave expanding through Conapto on cheap renewable power. HIVE Digital dropping $115 million and locking a 10-year lease on a facility in Boden they'd been renting for eight years already. Google breaking ground with a symbolic €5 million skills fund bolted on like a press release afterthought. If you run a company in this country and you read all that as "Sweden is winning AI," I need you to stop and think about what actually got sold.

None of this is Swedish AI. It's Swedish electricity, Swedish land, and Swedish cold air, wrapped in a press release that lets foreign capital say the word "Sweden" next to the word "billion." The compute stays foreign-owned. The IP stays foreign-owned. The margin, the actual money that compounds and builds the next generation of companies, goes back to Paris, to New Jersey, to Mountain View, to wherever HIVE's shareholders live. We get jobs running cables and clearing snow off substations. That's the trade.

Custom SaaS Development and Software Development Sweden: The Part Nobody's Investing In

Here's what actually builds sovereign capability: custom SaaS development teams that own their product, their data, and their customer relationships end to end. Software development in Sweden that produces IP Swedish companies hold, not rack space Swedish companies rent out at a discount. That distinction matters more than any data center groundbreaking. A data center is a building with a power bill. A software company that ships product is an asset that appreciates, employs engineers who level up, and generates margin that stays in the country when the founders pay taxes here. Jönköping doesn't need another 32 megawatt facility. It needs fifty more companies doing what we do at HEIMLANDR: building product, owning the stack, and selling it globally instead of selling our basement to whoever's cooling their GPUs this quarter.

I want to be precise here because this gets lazy fast in comment sections. I'm not against data centers. Sweden has genuinely excellent conditions for them: cold climate, stable grid, cheap and clean electricity thanks to hydro and nuclear, and political stability that Ireland and increasingly parts of the US can't promise anymore. That's real. Boden and Luleå built an entire regional economy around this over the last fifteen years, going back to Facebook's original data center there. That part of the story is legitimate industrial policy done right.

What's not legitimate is the framing. CNBC calling Mistral "Europe's answer to OpenAI" while it builds its actual training infrastructure on Swedish soil doesn't make Sweden the answer to anything. Mistral is a French company. The IP, the model weights, the enterprise contracts, all of it sits in Paris. We're the basement. A very good, very cold, very efficient basement. But still the basement.

Nordic vs. Global: Who Actually Owns the Stack

Look at how this compares elsewhere. The US argument, whatever you think of it, is at least coherent: American hyperscalers build American-owned compute to train American-owned models to sell American-owned products globally. The capital, the IP, and the infrastructure are aligned under one flag. China does the same thing with even less subtlety. The Gulf states, Saudi Arabia and the UAE pouring sovereign wealth into AI infrastructure, are doing something smarter than us: they're buying equity stakes and board seats alongside the electricity deals, not just leasing land. They understood the leverage they had and used it.

Sweden and the broader Nordic region keep taking the easiest version of the deal: sell power at a discount, get a groundbreaking photo op, call it a win. Dagens Industri and SVT cover these announcements with the same tone every time: jobs, investment, green energy, future-proofing the region. Nobody in that coverage asks the only question that matters, which is who owns what gets built on top of that electricity. That's not journalism failing, by the way, that's regulators and politicians failing to give journalists anything harder to report on, because nobody in government asked the hard question either before signing the deal.

Is Swedish and EU Policy Keeping Up? No.

The EU AI Act is a compliance framework. It regulates how AI systems behave, what disclosures they need, what risk categories they fall into. It says almost nothing about who owns the physical and computational substrate that European AI runs on. That's a massive blind spot, and it's the one that actually determines economic sovereignty five years from now. If every serious foundation model trained on European soil is trained by an American, French, or Gulf-backed company using Swedish power and Swedish land, the EU can regulate the outputs all it wants. It still doesn't own the means of production. That's not a policy detail. That's the whole game.

Sweden's own industrial policy has no equivalent to what Denmark did with wind or what France is at least attempting with Mistral itself, meaning direct state backing of a domestic model company with actual equity upside for the state. We have Energimyndigheten handing out favorable electricity agreements and municipalities racing each other to offer the cheapest deal, competing against each other instead of negotiating as a bloc. Boden, Luleå, and now wherever EcoDataCenter's Mistral facility lands are all competing for the same foreign capital instead of coordinating leverage. That's a self-inflicted wound, and it's fixable, but nobody in Stockholm seems interested in fixing it because the groundbreaking photos look good regardless.

Where This Actually Goes

Play this forward three to five years. Compute demand for frontier AI training keeps compounding, probably faster than most people's spreadsheets currently model, especially once we're past the current plateau and into whatever comes after the current generation of large models hits diminishing returns and someone cracks the next architecture. Whoever needs to train the next generation needs somewhere cold, stable, and cheap to do it. Sweden stays attractive for exactly that reason. More announcements are coming. Bigger ones. The question is whether we use this window to negotiate better terms or whether we keep taking whatever's offered because the headline reads well in an election cycle.

The realistic path toward AGI, however you define that milestone and whenever it actually arrives, runs through control of compute at scale. Not through clever prompting, not through regulation, through who owns the physical infrastructure and the capital to keep scaling it. Nations and companies that only host that infrastructure without owning stakes in what it produces become permanently downstream of whoever does own it. That's not a hypothetical. That's the current trajectory extended honestly. Sweden is choosing, right now, in 2026, whether to be a landlord with equity or a landlord with a lease agreement. Right now we're picking the lease.

The smarter version of this deal exists and isn't complicated. Government-backed equity stakes in exchange for power discounts. Mandatory local hiring quotas with actual teeth, not €5 million skills funds that amount to rounding error against billion-dollar builds. Requirements that some percentage of trained models or resulting IP get licensed back to Swedish companies at preferential rates. None of this is radical. It's just leverage that currently nobody in a negotiating room is using.

What This Means If You're Building, Not Just Watching

If you're a founder or CTO in Sweden reading data center headlines and feeling optimistic, redirect that energy. The opportunity isn't in the buildings going up in Boden. It's in what gets built on top of infrastructure that exists regardless of who owns it. Cheap, stable compute close to home is genuinely useful if you're smart about how you use it. Build the layer above the rack. Own your product. Own your customer relationships. This is exactly the argument I make constantly with clients considering rapid MVP builds versus waiting for "the market to mature." The market matures around whoever ships first and owns their stack. Nobody's waiting for you.

What to Look At

If you're actually trying to build sovereign capability instead of renting someone else's infrastructure, here's where I'd start looking this week:

• Kubernetes: if you're deploying anything at scale on infrastructure you don't fully control, container orchestration you own and understand is non-negotiable. Don't let the cloud provider's abstraction layer be the only thing standing between you and portability.
• awesome-selfhosted: 310,000 stars for a reason. Every tool on this list is a reminder that you can own your stack instead of renting it, from analytics to CRM to internal tools. Worth an afternoon of your team's time.
• Dify: if you're building agentic workflows or RAG pipelines, this lets you prototype and deploy without locking into one vendor's model ecosystem. Deploy on your own infrastructure, in Sweden, on that cheap stable power everyone's fighting over.
• n8n: fair-code workflow automation, self-hostable, 400+ integrations. If you're automating operations and don't want to be permanently dependent on a SaaS subscription owned by someone else, this is the honest alternative.

All four of these share one property that matters more than any feature list: you can run them on infrastructure you control, in a country with cheap clean power, without handing ownership of your data or your IP to whoever's data center you happened to land in.

What To Actually Do About This

Stop reading data center groundbreakings as good news for Swedish tech and start reading them as neutral infrastructure events that only become good news if Swedish companies build on top of them. If you're a founder, the move isn't to wait for the AI ecosystem to mature around you. It's to build now, own your product, and use the cheap compute without giving up equity in what you make with it. If you're evaluating blockchain development or smart contract development for tokenized infrastructure ownership models, that's not a fringe idea anymore either. Tokenizing revenue shares in Nordic data center capacity, letting local investors and municipalities hold real economic upside instead of just tax revenue, is a genuinely interesting structural fix to the exact problem I've spent this whole article describing. Nobody's built it seriously yet. Somebody should. If you're a policymaker, and I know a few of you read this, stop competing municipality against municipality for the same foreign capital. Coordinate. Ask for equity. Ask for IP licensing terms. The Gulf states figured this out years ago. We have better power, better stability, and worse negotiating posture. Fix the posture.

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.