TCS Just Sold SKF a Digital Transformation. That's the Problem.

SKF is 122 years old. It invented the modern ball bearing, basically built the industrial world's ability to spin things without friction. This week it signed over its "AI transformation" to TCS, an Indian systems integrator, in a global managed services deal. Read that again. A Swedish industrial icon just outsourced the thinking part of its future to a vendor headquartered in Mumbai. Not the servers. Not the electricity. The intelligence layer itself.

Meanwhile atNorth just opened a new data center on the Gothenburg metro grid, and Vinnova is funding MASSIV+, a serious climate-data modeling initiative with real public R&D money behind it. So Sweden has the concrete, the cooling, the grid capacity, and the government grant money. What it does not have, apparently, is the confidence to build the actual software that runs on top of all that. We built the house. Someone else moved in and is charging us rent.

What SKF actually bought, and what it actually gave away

Let's be precise about what happened, because "AI transformation deal" is the kind of phrase that hides more than it reveals. SKF bought managed services and an AI roadmap from TCS. That means TCS decides the architecture. TCS owns the implementation knowledge. TCS's engineers, not SKF's, understand why the system works the way it works two years from now when something breaks or needs to evolve. SKF gets outcomes. TCS gets the intelligence layer, the institutional knowledge, and the leverage that comes with being the only people who understand the black box they built inside a 122-year-old company.

This is not a criticism of TCS. They're good at this. It's a criticism of the decision to buy rather than build. There's a version of this deal that makes sense: augmenting internal teams, offloading commodity infrastructure work, buying speed you genuinely cannot build in time. There's another version where a company that makes physical, precision, safety-critical products decides that custom AI solutions built around its actual manufacturing data, its actual failure modes, its actual customer relationships, are a commodity you can procure like office furniture. SKF picked the second version.

The infrastructure paradox

Here's what makes this maddening from where I sit in Jönköping, an hour from SKF's Gothenburg backyard. Sweden is not short on capability. atNorth is building serious Nordic data center capacity, cheap green power, good latency to Europe, the whole pitch that made this country attractive for AI infrastructure in the first place. Vinnova is not sitting on its hands either, MASSIV+ is real money going into real climate and industrial data models with actual Swedish research institutions behind it.

So we have the power, the cooling, the compute, and the public funding for foundational research. What's missing is the layer in between: the applied engineering teams who take that infrastructure and that research and turn it into AI agents and systems that actually run a factory floor. That middle layer, the part that translates capability into product, keeps getting handed to foreign systems integrators because Swedish industrial leadership still thinks of software as a cost center to procure rather than a capability to own.

The Nordic pattern versus the global one

Look at what's happening elsewhere. In the US, industrial giants are not outsourcing their AI transformation wholesale to a single integrator, they're buying platforms and building internal teams on top of them, keeping the model layer and the data layer close. Palantir's entire business model is convincing manufacturers to never let the intelligence layer leave the building. In Asia, especially China and increasingly India itself, industrial AI is treated as sovereign infrastructure, something you build domestic capacity around because whoever owns the model owns the pricing power over the next decade of your industrial base.

Sweden's pattern, and I'd argue it's a pattern now, not an isolated incident, is different. We build excellent physical and energy infrastructure, we fund excellent early research through Vinnova and the universities, and then at the exact moment where that research needs to become deployed, revenue-generating, competitively defensible product inside a real company, we call in an outside vendor to do the last mile. Northvolt's collapse and SKF's TCS deal are different failures but they share a root cause: Swedish industry keeps treating the applied software and AI layer as something you buy at the end, not something you build alongside the physical product from day one.

Breakit and Dagens Industri have both covered how thin Sweden's actual AI-native industrial software bench is compared to the amount of public money and infrastructure poured into the space. That gap is not going to close itself. Someone has to build the applied layer, and right now the answer keeps being "someone in Bangalore or Redmond."

Is Swedish and EU policy even looking at this?

No, not really, and this is where I get genuinely frustrated. The EU AI Act is focused on risk classification and compliance paperwork, which matters, but it says nothing about industrial sovereignty over the AI layer itself. There is no Swedish policy instrument that asks "should our largest industrial employers own the intelligence layer of their own transformation, or is it fine if that layer is permanently licensed from a foreign vendor?" Vinnova funds research. It does not fund the harder, less glamorous work of getting a 122-year-old manufacturer to build internal applied AI capability instead of procuring it. Nobody in Swedish industrial policy is asking whether outsourcing your AI transformation is a national competitiveness risk the same way outsourcing your semiconductor supply chain would be treated as one. It should be asked. It currently isn't.

Where this goes in the next two to five years

Here's the trajectory that worries me. As AI models get more capable, the value in industrial AI moves away from the model itself and toward two things: the proprietary data you feed it, and the operational integration that turns predictions into actions on a factory floor. SKF has genuinely valuable proprietary data, decades of failure mode data, sensor data, supply chain data that no foreign vendor can replicate. But if TCS is the one architecting how that data gets used, TCS captures the integration value even if SKF technically owns the data.

Fast forward three years. AI agents are doing autonomous predictive maintenance scheduling, autonomous supply chain rerouting, autonomous quality control decisions on the line. Whoever built that agent layer, understands its failure modes, and controls its evolution has effectively become an unremovable layer of SKF's operating stack. That's not a vendor relationship anymore, that's structural dependency. And when you're structurally dependent on a vendor for the intelligence layer of your core operations, your pricing power, your ability to move fast, your negotiating leverage on the next contract, all of it erodes. This is exactly the mistake European telcos made with network equipment vendors twenty years ago, except this time it's happening to the layer that actually thinks.

As we get closer to genuinely general-purpose AI systems, agents that can reason across domains rather than execute narrow pipelines, this gap compounds. The companies that own their applied AI stack today will be the ones who can absorb frontier model capability fastest, because they already understand their own data and their own integration points. The companies renting their intelligence layer will be waiting on their vendor's roadmap. SKF just chose to be in the second group, for one of the most important transformations in its 122-year history.

What builders should actually look at

If you're a Swedish manufacturer, or any industrial company reading this and recognizing your own procurement instincts, here's where I'd start looking instead of signing the next big managed services contract.

• OpenHands, an open AI-driven development framework that lets internal engineering teams build and iterate on their own automation and agent tooling instead of waiting on an external integrator's release cycle.
• Graphify, which turns your existing codebase, SQL schemas, and documentation into a queryable knowledge graph. This is exactly the kind of tool that lets an internal team retain institutional knowledge instead of handing that understanding to a vendor who bills you to re-explain it back to you later.
• Daytona, secure elastic infrastructure for running AI-generated code safely, useful if you're serious about building internal agent capability on top of the atNorth-grade infrastructure Sweden already has sitting there unused by most manufacturers.
• rtk, a Rust CLI proxy that cuts LLM token consumption 60-90 percent on common dev commands. If cost is the excuse for outsourcing instead of building, this closes that gap fast.

None of these replace a real engineering strategy. But they're proof that the tools to build your own applied AI layer, in-house, on infrastructure you already have access to in this country, are open, cheap, and getting better every month. The excuse that you "have to" buy this off the shelf from a foreign integrator is getting weaker every quarter.

What to actually do about this

If you run a Swedish or Nordic manufacturing company and you're staring at a TCS-shaped contract on your desk right now, ask one question before you sign: who owns the intelligence layer in three years, you or them? If the honest answer is "them," you're not buying a transformation, you're buying a subscription to someone else's understanding of your own business.

The alternative is not "build everything yourself with no outside help," that's naive for most companies. The alternative is building the applied layer with partners who transfer capability rather than hoard it. That's the difference between hiring a team to build you AI agents that your own people can maintain and extend, versus buying a managed service where the knowledge never leaves the vendor's walls. It's the difference between a rapid MVP that proves out an idea with your team owning the code, versus a multi-year managed services contract that locks you in before you've even seen a working prototype. Sweden has the power, the compute, and the research money. What we need now is industrial leadership willing to build the applied layer instead of renting it, and policy that treats owning your AI intelligence layer with the same seriousness we'd treat owning your energy supply. Right now we're building excellent infrastructure for other people's intelligence to run on. That's not a strategy. That's landlording, and landlords don't win industrial competitions.

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.