The AI Class War Is Here: 20% Are Eating the 80% Alive

The numbers are in. They're worse than you think.

PwC drops its latest AI economic impact report today and the headline number should stop every CEO in Europe mid-coffee: 75% of AI-generated economic value is concentrating into roughly 20% of companies. Not 20% of industries. Not 20% of countries. Twenty percent of companies. And the gap between the leaders and everyone else is not stabilizing. It is accelerating. If you run a mid-market European business and you are still in "pilot project" mode with AI, you are not being cautious. You are being eaten.

I have been saying this from Jönköping for two years. Nobody at the big Swedish consultancies wanted to hear it. The narrative was always: "AI is early. There's time. Let's do a proof of concept. Let's wait for the EU AI Act to settle." Well, the PwC data is here. And while Swedish enterprises waited, the 20% didn't.

What the 20% actually did differently

The PwC report makes one thing clear that most summaries will skip: the leading companies are not just using AI for productivity. They are using it for growth. That distinction matters enormously.

Most AI "adoption" in Europe looks like this: take an existing process, bolt a chatbot or document summarizer onto it, call it transformation, put it in the annual report. The 20% did something else entirely. They rebuilt revenue models. They created new product lines. They used AI agents not to answer customer emails faster, but to identify customer segments that didn't exist before, to price dynamically, to enter markets they previously couldn't serve.

Growth, not efficiency. That is the dividing line.

The efficiency play is real. It saves money. But it does not compound. A company that cuts costs 15% with AI automation hits a ceiling. A company that uses AI to find and serve new customers at near-zero marginal cost does not. It compounds. And after two or three years of compounding, the gap becomes a canyon.

Sweden's position: the SCB data doesn't lie

SCB, Sweden's national statistics bureau, releases its 2025 enterprise data on AI adoption today. I have been waiting for these numbers because the vibes in the Swedish tech scene and the reality on the ground are two different things.

The narrative: Sweden is a tech leader. We have Spotify, Klarna, King. We are digital natives. We will figure out AI.

The reality: Swedish enterprise AI adoption is concentrated in the same handful of large companies it was concentrated in two years ago. The mid-market, the 10-to-500-employee companies that make up the backbone of Swedish industry, remains overwhelmingly in the experimental phase. Pilot projects. Internal chatbots. "We're evaluating tools." The SCB data confirms this pattern.

This is not a technology problem. Sweden has the talent. We have world-class engineering schools. We have some of the best internet infrastructure on the planet. From Jönköping, I can push code to production faster than most teams in San Francisco can get a meeting scheduled. The problem is organizational. It is cultural. And increasingly, it is regulatory.

The EU AI Act: protection or paralysis?

I want to be careful here because I actually believe in regulation. The Wild West approach the US took in 2023-2024 created real problems. But the EU AI Act, as currently being implemented, is creating a different kind of problem: paralysis by compliance anxiety.

Talk to any Swedish CTO at a mid-market company. Ask them why they haven't deployed AI agents into their customer-facing operations. Nine times out of ten, the answer involves some version of "we're waiting to understand our obligations under the AI Act." Meanwhile, their American and Chinese competitors are deploying and iterating. The compliance burden falls hardest on the companies least able to absorb it. A company with 50 employees doesn't have a regulatory affairs team. A company with 50,000 does.

The EU AI Act was supposed to create trust. Instead, for the European mid-market, it is creating hesitation at exactly the moment when speed is everything. Swedish regulators need to hear this: if you want AI value created in Europe rather than just consumed from American platforms, you need to make compliance achievable for companies that don't have a legal department the size of a small village.

The hyperscalers are coming for the integration layer

Here is the other headline that should terrify every IT services company in the Nordics: OpenAI launches its Deployment Company today for direct enterprise integration. Anthropic, meanwhile, is building a new AI services company backed by Blackstone, Goldman Sachs, and Hellman & Friedman.

Read that again. The AI model companies are not content to sell APIs. They are now building the integration and consulting businesses themselves. They are coming directly for the layer that companies like Accenture, Capgemini, and every mid-tier Nordic IT consultancy assumed was theirs.

This changes the math completely. Until now, a Swedish company wanting to deploy custom AI solutions would hire a consulting firm, who would then build on top of OpenAI or Anthropic APIs. That consulting firm captured a huge margin for what was essentially a middleware position. Now OpenAI and Anthropic are saying: why should we let someone else capture that margin?

For European mid-market companies, this is actually a double-edged situation. On one hand, it might get cheaper and easier to deploy AI directly. On the other, it means the entire value chain is consolidating toward the model providers. Your data, your processes, your competitive intelligence, all flowing into systems owned and operated by three or four American companies. If European sovereignty over AI means anything, this is the moment it gets tested.

At HEIMLANDR, we have been building AI agent development capabilities precisely because we saw this coming. The companies that will hold their ground are the ones with partners who build on open architectures, who can work with multiple model providers, who keep the client's data and logic under the client's control. Not locked into one hyperscaler's ecosystem.

Where this goes: 2027-2030

Let me sketch the trajectory as I see it from where I sit.

By end of 2027: The 20/80 split PwC identifies today becomes a 15/85 split. AI-native companies will have compounded their advantages for another 18 months. Many mid-market European companies that are currently in "pilot" mode will not have shipped anything meaningful. Some will have been acquired. Some will have lost enough market share that the math stops working. The AI automation business gap between leaders and laggards becomes structurally permanent for those who waited too long.

By 2028-2029: The path toward more capable AI systems, whatever you want to call them, AGI-adjacent, ASI-lite, strong generalists, makes the current model look quaint. Today's AI agents handle defined tasks in defined contexts. The next generation will handle ambiguous tasks across contexts. Companies that built their AI infrastructure now will be able to absorb these capabilities. Companies that didn't will face a complete architectural rebuild. That is a two-year project minimum. Which means if you start in 2028, you are operational in 2030. Your competitors who started in 2025 will have five years of compounded advantage.

By 2030: The concept of an "AI development company Europe" as a separate category stops making sense, because every serious software company is an AI company. The distinction dissolves. The question is whether European companies are building this capability or renting it from Mountain View.

The regulatory gap that nobody is talking about

Here is what I think Swedish and EU policymakers are missing entirely: the AI class war PwC describes is not just a business problem. It is an economic sovereignty problem. If 75% of AI value concentrates in 20% of companies, and the vast majority of those companies are American, then Europe is not just losing a technology race. It is becoming an economic dependency. We went through this with cloud. We went through it with social media. We are about to go through it with AI, except this time, the value at stake is an order of magnitude larger.

Swedish industrial policy has traditionally been good at picking spots: telecom, automotive, fintech. But I see no coherent AI industrial strategy coming from the government. Vinnova funds some research. A few incubators run AI programs. It is not enough. Not close.

What to look at

If you are a CTO, founder, or senior engineer trying to move from the 80% into the 20%, here are the things I would actually spend time on this week:

1. Langflow (149k+ stars on GitHub). This is one of the most practical tools I have seen for building and deploying AI-powered agents and workflows without starting from scratch. If you are a mid-market company and you need to get from "we have ideas" to "we have a deployed agent" in weeks rather than months, this is worth your engineering team's attention right now.

2. Claude Code (130k+ stars). Anthropic's agentic coding tool is sitting at the intersection of developer productivity and AI agent architecture. If your engineering team is not using tools like this already, you are leaving 30-40% of your development velocity on the table. That is not a guess. That is what I see in our own numbers at HEIMLANDR.

3. awesome-llm-apps (113k+ stars). Over 100 AI agent and RAG applications you can actually clone, customize, and ship. This repo is the fastest way I know to go from "what could we build?" to "here's a working prototype" for a specific business problem. Stop theorizing. Clone something. Run it. Show your team what is possible.

4. The PwC report itself. Actually read it. Not the summary. Not the LinkedIn posts about it. The methodology section. Understand how they define the 20%. Then honestly assess where your company sits. If the answer is uncomfortable, good. Discomfort is the precondition for action.

What you should actually do Monday morning

I am not interested in writing articles that people nod along to and then do nothing about. So here is what I would do if I were running a 50-to-500 person European company today:

Kill your pilot projects. All of them. If they haven't shipped to production in 90 days, they are not pilots. They are excuses. Pick one. Ship it. Learn from it. Then pick the next one.

Audit your AI vendor dependencies. If everything you are building runs through a single provider's API with no abstraction layer, you are one pricing change or one Terms of Service update away from a crisis. Build on open architectures. Use multiple models. Keep your options open.

Hire or partner for AI agent development, not AI strategy. You don't need another PowerPoint deck. You need someone who can build and deploy AI agents that work in your specific business context. Whether that is an internal hire, an AI development partner, or both, the priority is execution capability.

Talk to your board about growth, not savings. If your AI business case is built entirely on cost reduction, you are playing the wrong game. The 20% are using AI for revenue. Your board needs to understand that distinction today.

The window is not closing. It is almost closed.

I write this from Jönköping on a Sunday morning. It is quiet here. The lake is flat. Sweden is beautiful and calm and, right now, dangerously comfortable. The PwC numbers, the SCB data, the hyperscaler moves into services. These are not abstract trends. They are the sound of a door closing.

We at HEIMLANDR are building for the companies that hear that sound and choose to move. Punk rock tech means something specific to us: build fast, own your stack, don't wait for permission. The 80% is not a destiny. It is a choice. And it is a choice that gets made every week you defer shipping something real.

Choose differently.

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.