Trump Blinked on AI. Europe Should Be Terrified, Not Relieved

Here is what happened this week, stripped of the spin. Trump delayed an AI executive order, big tech lobbied it into something toothless, and he signed it anyway. Axios, the Washington Post, and the Guardian all cover this from different angles, but they land on the same fact: the oversight provisions got gutted before the ink dried. This is not a policy story. This is a negotiation story, and the negotiation was a hostage situation where big tech held all the leverage from the start.

Meanwhile Satya Nadella is in London this week telling enterprises that Copilot adoption is not optional anymore, it is inevitable. He is right about the inevitability. He is conveniently quiet about the fact that the regulatory ground these enterprises are supposed to build on keeps shifting under Washington's own hand.

If you run a company and you think this is a US problem, you are wrong. This is the moment that tells you exactly how much weight the EU AI Act is about to carry alone.

The Hostage Negotiation Nobody Called by Its Name

Let's be precise about what a "watered-down executive order" actually means. An executive order is the fastest, most direct tool a US president has to set policy without Congress. When that tool gets softened after lobbying pressure, it tells you the lobbying got to the fastest lever available. Not the slow one. The fast one. That is not compromise. That is capture.

I have said this before and I will keep saying it: regulation written by the people it regulates is not regulation, it is a permission slip with a stamp that looks official. The pattern in Washington right now is the same pattern that gave us decades of financial deregulation before 2008. Move fast, delay enforcement, let the biggest players write the actual rules, call it innovation.

The EU AI Act, whatever its flaws, was built through years of actual friction. Committees fought. Member states disagreed. It got watered down too, plenty of times, but it emerged as a binding legal framework with real enforcement teeth and real fines. Compare that to an executive order that can be gutted in a single lobbying cycle and rewritten by the next administration in an afternoon. One of these is law. The other is a mood.

Why Being the Only Adult in the Room Is Expensive

Here is the part nobody in Brussels wants to say out loud: being the responsible actor when everyone else abandons responsibility does not make you safe, it makes you a target for capital flight.

If US AI development runs faster and looser, capital follows speed. If the EU keeps its compliance bar high while the US bar keeps sinking, every founder building AI agent development platforms in Stockholm, Berlin, or Amsterdam has to answer a brutal question from investors: why build here, under this much friction, when the market that actually buys your product moves faster somewhere else?

This is not a hypothetical. Look at Northvolt. Sweden's flagship industrial bet on the future of batteries, backed by real political will, real subsidies, real national pride, and it is still missing production targets according to recent reporting from Dagens Industri. If Sweden struggles to execute on manufacturing with full political backing behind it, what makes anyone think Europe executes cleanly on AI governance, where the will is split across 27 member states and the technology moves monthly, not yearly?

The lesson from Northvolt is not "Europe should give up on ambitious bets." The lesson is "ambition without execution discipline burns cash and credibility at the same time." The EU AI Act needs the opposite problem solved. It has the legal ambition. What it needs now is an execution layer that does not choke the companies actually trying to build compliant AI development company Europe infrastructure while the giants outside Europe move at a different clock speed entirely.

Sweden and the Nordics: Correct on Paper, Alone in Practice

From Jönköping, here is what I see. Swedish founders take AI governance seriously, mostly because we have no other choice, GDPR trained an entire generation of Nordic builders to think about compliance before shipping. That instinct is not a weakness. It is a genuine competitive asset when you are selling to European enterprises, public sector, or healthcare, where trust is the product. But that instinct becomes a liability the moment your buyer is in San Francisco or Singapore and does not care about your compliance story, they care about your shipping speed. Sweden's AI ecosystem, and the Nordic region broadly, is small, well-funded per capita, technically excellent, and structurally slower to reach global scale than the US or Chinese equivalents. That is not new. What is new is that the gap in regulatory seriousness between Europe and the US just widened again, and it widened in the direction that makes the US faster and looser, not the other way around.

SVT and Breakit have both covered Swedish AI startups increasingly registering US entities or seeking US investors specifically to escape this friction gap, not because Swedish engineering is worse, it is often better, but because the market clock in the US now runs on rules that big tech itself gets to edit in real time. That is the actual competitive threat. Not that American AI companies are smarter. That American AI policy is now optional for the companies large enough to lobby it into submission.

Where This Actually Goes

Play this forward two to five years and here is the trajectory I expect. The US ends up with a two-tier AI regulatory reality: enormous companies operating under self-negotiated light-touch rules, and everyone else, smaller labs, startups, open source projects, operating under whatever is left standing after the big players finish editing it. This is already visible in how differently OpenAI, Anthropic, and Microsoft get treated in policy conversations compared to a five-person startup building on top of their APIs. Europe, if it holds its nerve, becomes the default jurisdiction for AI systems that need to be trusted rather than just fast: healthcare AI, financial AI, government AI, critical infrastructure AI. The AI Act's risk-tiered approach, whatever friction it creates today, is actually the right shape for that future. The problem is not the shape of the law. The problem is execution speed, the same disease that is currently eating Northvolt's production targets. The AGI question sits underneath all of this and most executives are not thinking about it correctly. If frontier labs get closer to general-purpose systems capable of autonomous, multi-step reasoning and action, an executive order with gutted oversight provisions is not just a policy embarrassment. It is a live safety gap at the exact moment capability crosses a line, we cannot currently define precisely, but will recognize the moment we are past it. The EU AI Act's tiered risk framework was designed with exactly that scenario in mind. Watered-down US oversight was not. What I expect concretely: continued regulatory arbitrage, US-based frontier labs setting de facto global standards through product behavior rather than law, and Europe's actual leverage point becoming procurement, not legislation. If EU governments and large enterprises require AI Act compliance to sell into the European public sector, that requirement becomes more powerful than any executive order Washington signs or unsigns.

What to Actually Do About It

Stop waiting for regulatory clarity before you build. It is not coming, not from Washington, and Brussels moves on its own multi-year clock. Build compliant by default, treat data governance and explainability as architecture decisions, not paperwork you add later. If you are building AI agents or AI automation business tools right now, look hard at what is actually working in the open source ecosystem before you reinvent it. Dify and Langflow are both proving that agentic workflow tooling does not need to live inside a single vendor's walled garden, and self-hosting them keeps your data and your compliance story under your own control, which matters enormously if you sell into regulated European markets. Claude Code is also worth serious attention if your team is shipping software faster than your governance process can review it, because that gap is exactly where regulatory risk lives. At HEIMLANDR we build custom AI solutions with this reality baked in from day one, not bolted on after a client asks about compliance. If you are exploring AI agent development for your business and want it built to survive whatever regulatory direction Washington or Brussels takes next, that is a conversation worth having before you commit to a vendor who treats governance as an afterthought.

The Real Signal in All This Noise

Trump's executive order was noise dressed as signal, an announcement engineered to look like action while big tech quietly removed the parts that would have cost them anything. The EU AI Act is signal that looks like noise, slow, bureaucratic, occasionally maddening, but real and enforceable. The market has not fully priced that difference yet. It will. Every founder, CTO, and CEO reading this needs to stop asking "which market has less regulation" and start asking "which market has regulation that actually holds." Those are different questions with different answers, and the gap between them is where the next five years of AI competitive advantage gets decided.

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.