Build vs Buy Is Dead. Now It's Build vs Rent vs Own the Model

Somebody in your leadership Slack right now is building a slide deck titled "Build vs Buy." Close it. That framework died sometime around when Claude and GPT stopped being chatbots and started being engineers. The real question in August 2026 is not build or buy. It's build, rent, or own the model that runs your business. Most Swedish companies are still answering a question nobody's asking anymore, and they're about to sign three-year contracts for it.

I run HEIMLANDR out of Jönköping. We build custom SaaS and AI systems for companies who used to think custom software meant six months and a six-figure invoice. That math is gone. What replaced it is more interesting, and more dangerous if you don't understand it.

The Old Build vs Buy Math Is a Museum Piece

The 2019 logic was simple. Building custom software cost too much time and too much money for anything except your core differentiator. Everything else, you bought. HR software, CRM, support tickets, internal tools. You rented convenience because building was expensive and slow. Spiceworks and IT Brew are both pointing at the same thing this week: AI-augmented teams can now build what used to require a full department. A team of three with Claude Code, OpenHands, and a decent CTO can ship an internal tool in a week that would have taken a twelve-person outsourced shop three months in 2021. That's not a productivity boost. That's a different category of company being able to build software at all. I've seen it directly. A logistics client in Gothenburg came to us wanting a "simple AI-powered dashboard" quote from a SaaS vendor. Eighteen thousand euros a year, locked contract, their data living on someone else's infrastructure with export limitations buried in section 14. We built them an equivalent system with a rapid MVP sprint in three weeks. They own it. No renewal negotiation. No vendor deciding to sunset a feature they depend on.

"Buy" in 2026 Means Renting Someone Else's AI Stack

Here's what nobody selling you a SaaS platform wants said out loud: they're using the same AI tools you could use yourself. That "AI-powered" CRM you're about to sign for? Built by a small team using the exact agent harnesses and codebases you have access to right now. You're paying margin on top of infrastructure you could own. Look at what's trending in developer tooling this week. ECC is optimizing agent harness performance for Claude Code, Codex, and Cursor at a level that used to require a specialized ML team. Graphify turns your entire codebase, docs, SQL schemas, and configs into a queryable knowledge graph with local deterministic parsing, no vector store needed. This is the same tooling category the vendors are quietly running behind their pricing pages. The gap between "vendor building AI product" and "you building AI product" has basically closed. What's left is who owns the output. This is the real trap in "buy" right now. It's not that vendors are bad. It's that buying has quietly become a euphemism for renting a model you'll never own, running on infrastructure you don't control, with a pricing curve that goes up every year your usage does. IBM and Red Hat pushing "trust infrastructure," Google pushing sovereign cloud partnerships to its enterprise partners, these aren't innovations. They're defensive plays. Big Tech knows the AI-augmented build wave is coming for their SaaS margins, so they're repackaging lock-in as sovereignty and trust before you notice you don't need them.

Sweden's Blind Spot: We Think We're Cautious. We're Actually Just Renting Slower.

Here's where I get annoyed as a Swede. We pride ourselves on being careful, deliberate, GDPR-compliant, security-first. Fine. But careful has become an excuse to default to buying instead of building, even when building is now the more sovereign, more secure, more controlled option. Dagens Industri and Breakit have both covered the Google push for Swedish and EU partners to run "sovereign Google Cloud" instances. Read that carefully. Google, the company least aligned with EU data sovereignty by default, is now selling sovereignty as a product feature. That should tell you everything. If the hyperscaler is scrambling to sell you sovereignty, it means sovereignty is now a competitive weapon and they're worried you'll get it somewhere else. Or worse for them, you'll just build it yourself. Post-Schrems II, Swedish companies have spent five years in a defensive crouch about US cloud dependency, and the answer the market gave them was: rent a different flavor of the same dependency, just with better PR. Nobody's selling Swedish founders the actual third option, which is build sovereign, own your model, own your data pipeline, skip the SaaS tax entirely. That's not a hypothetical. That's a service line we run at HEIMLANDR because the demand is already there and nobody else is filling it. Compare this to the US market, where venture-funded startups are burning cash building in-house AI infrastructure specifically to avoid OpenAI and Anthropic pricing exposure at scale. Compare it to what's happening in parts of Asia, where state-backed sovereign AI stacks are being built at national level, no vendor negotiation required. Sweden's instinct is to wait for a trusted vendor to solve this for us. That instinct is now a liability. The tools exist today for a Jönköping manufacturing company or a Malmö fintech to build and own its AI infrastructure without waiting for Google, IBM, or anyone else to grant them sovereignty as a favor.

Where This Actually Goes: The Three-Way Split

Over the next two to five years, I think the market splits into three tiers, and most companies will misjudge which tier they're in. Tier one: rent everything. Fine for companies with no technical differentiation and no data sensitivity. Payroll, generic scheduling, whatever. Keep buying SaaS here. It's genuinely still cheaper than building. Tier two: build fast, own the code. This is the tier that exploded this year. AI-augmented teams of two to five people building internal tools, customer-facing products, and MVPs that would have needed twenty people in 2020. This is where the build-vs-buy conversation actually still applies, except now build wins almost every time on cost and control. Tier three: own the model. This is the tier nobody's really talking about yet outside AI labs and a handful of sovereign-minded enterprises. It's not just owning your code. It's owning or fine-tuning the actual model your business runs on, so you're not exposed to a foundation model provider changing pricing, changing terms of service, or getting acquired and shifting strategy overnight. As we get closer to genuinely general-purpose AI systems, the companies that own their model layer will have a structural advantage over companies renting API access to someone else's frontier model. If GPT-6 or Gemini 4 or whatever comes next triples in price or gets nerfed for your use case, tier three companies shrug. Tier one and two companies scramble. The path toward AGI makes this more urgent, not less. The closer we get to models that can genuinely reason and act autonomously across domains, the more dangerous it is to have your core business logic sitting on rented intelligence you don't control. Regulation is not remotely ready for this. The EU AI Act addresses risk categories and transparency, but it says almost nothing about the strategic risk of vendor dependency on foundation models themselves. That's a policy gap, and Swedish and EU regulators haven't caught up to it because they're still fighting the last war on data privacy, not the next one on model dependency.

What to Actually Look At This Week

If you're a CTO or founder trying to figure out where you sit in this shift, here's where I'd start: Graphify for understanding your own codebase and data structure before you decide what to build versus rent. If you can turn your legacy systems into a queryable knowledge graph in an afternoon, you remove the biggest excuse for staying locked into a vendor: "we don't understand our own data well enough to migrate." OpenHands for actually testing whether AI-driven development can replace a chunk of what you're paying a vendor for right now. Don't theorize about it. Run a real sprint against a real vendor feature and see the gap. rtk, the Rust CLI proxy cutting LLM token consumption 60 to 90 percent on dev commands. If you're going to build AI-augmented in-house, your token bill matters, and most teams are burning cash on inefficient prompting they don't even notice. And honestly, talk to a team that's done this transition already before you sign anything. We do AI solutions and fullstack development work specifically for companies trying to figure out whether their next system should be rented, built, or owned at the model layer. That conversation is worth more than another vendor demo.

What You Should Actually Do About It

Stop asking build vs buy. Ask three questions instead. What am I currently renting that a small AI-augmented team could build and own in weeks, not months? What data or logic in my business is sensitive enough that model-level ownership matters, not just code ownership? And which vendor contracts am I about to renew out of habit rather than genuine necessity? Audit your SaaS spend this month. For every recurring line item, ask whether it's tier one, genuinely commodity, or tier two dressed up as tier one because a vendor's pricing team is good at their job. You'll find more of the latter than you expect. If you're a Swedish founder specifically, stop waiting for a hyperscaler to hand you sovereignty. Build it. The tools are cheap, the talent exists here, and the regulatory pressure toward data sovereignty is only going to increase, not decrease.

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.