Tandem Health's €86.5M: Healthcare AI Is Built Around Hospitals, Not In Them

Tandem Health just raised €86.5M (£74M) to build what they call an "AI-native operating system for European healthcare." Read that phrase slowly. Not a tool. Not a feature. An operating system. That is not a startup pitching a product to hospitals. That is a startup pitching itself as the layer hospitals will eventually have to run on top of, whether they like it or not.

I have spent fifteen years building software in Sweden. I know exactly what it means when a private company says "operating system" about a sector that is supposed to be run by the public system. It means the public system did not ship. Somebody else is shipping instead, and they are going to own the rails.

The Number Everyone Is Celebrating Is the Problem

A new Nordic survey making the rounds this week says one in three AI initiatives in healthcare is now live in production, not stuck in pilot purgatory. Everyone is treating this as a win. Karolinska Institutet is getting cited in Nature Medicine for AI diagnostics work. Sweden just launched an AI vård-bot (healthcare chatbot) marketed explicitly on the claim that it "won't make things up," Hittar inte på, as if that is a feature worth a press release instead of the absolute bare minimum you would expect from a clinical tool.

Sit with that for a second. We are running a national marketing campaign around the idea that our healthcare AI does not hallucinate. That is not confidence. That is damage control before the product has even had a scandal. When you have to reassure the public that your AI is not making things up, you are telling me the trust was never solid to begin with, and everyone building this stuff knows it.

So here is the actual story hiding under the celebratory headlines. One in three initiatives live in production sounds good until you ask who built them. Increasingly, the answer is not Region Jönköping, not Region Stockholm, not the offentlig sektor at all. The answer is a startup with venture money and a five year runway to become infrastructure before anyone in government even finishes the procurement paperwork.

Why Public Healthcare Cannot Ship Software Fast Enough

I am not being cynical about public healthcare. Swedish healthcare is genuinely good at the thing it was built to do, which is treat sick people regardless of income, reliably, for decades. What it was never built to do is ship software at startup speed. Procurement cycles run in years. Vendor lock-in from the 1990s is still active in some regions. Data governance rules that exist for good reason (GDPR, patient confidentiality, Socialstyrelsen requirements) become friction that a public IT department cannot move through fast, and a VC-funded startup can. That asymmetry is the whole ballgame. A regional healthcare IT department needs eighteen months and three committees to deploy a clinical decision support tool. Tandem Health needs a pilot contract, six months, and a Series A round. By the time the public system finishes evaluating vendor risk, the startup already has the data, the workflow integration, and the clinicians who now depend on it to get through their day.

This is not a Swedish problem specifically. It is an EU-wide pattern. The EU AI Act classifies most clinical AI as "high-risk," which is the right call from a safety standpoint, but it also means compliance costs that only well-capitalized players can absorb comfortably. Ironically, the regulation meant to keep AI in healthcare honest is also the regulation that filters out anyone except venture-backed platforms with lawyers on staff. Public procurement, already slow, gets slower. Private infrastructure plays, already fast, get a moat.

The Quiet Part: Who Owns the Data Layer in Five Years

Here is where I want you to actually think like a founder or a hospital administrator instead of a headline reader. When a company says "operating system for healthcare," what they mean concretely is: the layer where patient data lives, where clinical workflows run, where every future feature gets bolted on. Whoever owns that layer owns the leverage. Right now, in September 2026, that layer is being built by companies like Tandem Health, and it is being built with private capital, on top of public infrastructure, using data generated inside public clinics. The clinics get a nice dashboard and a chatbot that promises not to lie to them. The startup gets the data pipeline, the clinical relationships, and increasingly, the negotiating position. Fast forward five years. I think we are looking at a real possibility that clinics who did not build or buy into an AI-native data layer early will find themselves in a position where they are effectively renting access to their own historical patient data back from a Stockholm startup, because that startup's platform became the de facto standard for how that data gets structured, queried, and acted on. That is not dystopian speculation, that is how every platform monopoly in the last two decades has played out, from app stores to cloud infrastructure to payment rails. Healthcare is just slower to get there, which means it is happening right now while everyone is busy congratulating themselves on adoption numbers.

Nordic vs. Global: We Are Actually Ahead, and That Is the Risk

Compare this to the US, where healthcare AI adoption is fragmented across a hundred different insurance-driven incentive structures, and most hospital systems are years behind on basic interoperability. Compare it to parts of Asia where state-directed healthcare AI moves fast but under direct government control, which creates a different but equally real concentration of power. Sweden and the Nordics are actually ahead on raw adoption. One in three live deployments is a genuinely good number by international standards. But being ahead on adoption while being behind on ownership structure is arguably worse than being behind on both. If everyone is slow, at least everyone has time to build the right governance model before the platforms lock in. If you are fast on adoption and slow on governance, you get exactly what is happening now: private infrastructure racing ahead of public oversight, wrapped in the comforting language of "AI-native" and "operating system." I say this as someone building AI products for a living. I want the private sector building healthcare tools. I do not want the private sector quietly becoming the substrate the entire public health system depends on without anyone in Riksdagen or the EU Parliament having a real conversation about what that means for patient data sovereignty ten years out.

Where This Actually Goes

Here is my honest forecast, and I will own it if I am wrong. Over the next two to three years, expect two or three companies (Tandem Health being one obvious candidate given this raise) to become the default clinical AI layer across multiple EU countries, the way Epic became the default electronic health record layer in the US. Once that happens, switching costs become enormous, because you are not just switching software, you are migrating years of structured clinical data and retraining an entire workforce's workflow habits. Layer on top of that the trajectory toward more capable general AI models. As models get better at multi-step clinical reasoning, differential diagnosis support, and administrative automation, the value concentrates even more heavily at the infrastructure layer, not the point-solution layer. A company that owns the data pipeline and the workflow integration will be able to swap in better underlying models (better Claude, better GPT, better open models like the ones people are optimizing with tools such as affaan-m/ECC for agent harness performance) faster than any public system could ever re-tender a contract. The infrastructure owner captures the upside of every model improvement. The clinic that just uses the dashboard does not. On the regulatory side, I do not think Sweden or the EU is remotely ready for this. The AI Act handles safety and risk classification reasonably well. It says almost nothing concrete about data sovereignty when the "operating system" running your national healthcare infrastructure is a five-year-old startup with foreign investors on its cap table. That is a gap someone in government needs to close before the platforms are too embedded to unwind, and right now nobody is having that conversation loudly enough.

What to Look At If You Are Building in This Space

If you are a founder, CTO, or clinic operator trying to figure out where to place your bets, here is where I would actually spend my attention right now:

• Graphify (Graphify-Labs/graphify), which turns any codebase, docs, SQL schemas, and PDFs into a queryable knowledge graph with deterministic AST parsing and no vector store. For anyone building clinical decision support that needs traceable, explainable data lineage (which regulators will eventually demand), this kind of deterministic graph approach is a much safer foundation than pure vector retrieval.
• OpenHands (OpenHands/OpenHands), an open AI-driven development framework, worth watching if you want to build your own clinical tooling instead of renting someone else's operating system layer.
• rtk (rtk-ai/rtk), a Rust CLI proxy cutting LLM token costs 60-90% on common dev commands. If you are a smaller clinic-adjacent startup trying to compete with well-funded platforms, your unit economics on AI usage matter enormously, and tools like this buy you runway.
• Read the actual EU AI Act high-risk classification requirements for medical devices, not the summary. Most founders in this space have not, and it will bite them later.

What You Should Actually Do About This

If you run a clinic, a regional healthcare authority, or a health-tech startup in the Nordics, stop thinking about AI adoption as the finish line. Adoption is table stakes now. The real question is: who owns the data layer your operation depends on, and what happens to your negotiating position in five years if that layer belongs to someone else. If you are building in this space and do not want to end up renting your own data back from a platform, you need custom AI solutions built on infrastructure you control, not a black box SaaS dashboard that happens to have a friendly chatbot attached. At HEIMLANDR we build exactly this kind of thing for clients who understand the stakes: proper AI solutions designed around data ownership from day one, AI agents that automate clinical and administrative workflows without locking you into someone else's platform, and when the situation calls for verifiable, auditable data structures, we bring in blockchain development to give patients and institutions real, provable ownership over their own records instead of trust-me dashboards. This is not theoretical for us. Building an AI automation business that survives contact with regulated industries like healthcare means designing for data sovereignty and auditability from the start, not bolting it on after a VC round forces you to move fast and skip the hard questions.

The Real Lesson From Jönköping

I am not against Tandem Health. €86.5M from serious investors for serious infrastructure ambition is a legitimate bet, and honestly, someone was going to build this. My problem is with the framing everyone in Swedish media is using, treating this purely as a healthcare innovation win, when the more accurate framing is: the public system failed to ship, so private capital stepped in to build the missing layer, and it is going to own that layer for a very long time. Sweden should be proud that it adopts fast. Sweden should also be nervous about who ends up owning the infrastructure underneath that adoption. Those are two different things and right now we are only talking about the first one.

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.