Back to Articles

The Great AI Consolidation is coming: European Systemic Risk Report

How a US-dominated AI oligopoly threatens systemic stability — and what Europe must do about it

By Sukalp Muzumdar

Update (February 2026): And now this…

Just two months after I wrote the piece below about the systemic risks of a US-dominated AI oligopoly, we got a perfect — and painful — illustration of the brain drain playing out in real-time.

This week, Peter Steinberger, the Austrian developer behind the viral open-source agent framework OpenClaw (formerly Moltbot), was poached by OpenAI. His autonomous agent — which lets users manage calendars, clear inboxes, and run complex workflows entirely on their own hardware — was a genuine European success story, racking up 1.5 million agents generated in just weeks.

Unsurprisingly, Silicon Valley pounced. Steinberger reportedly had competing offers from both Mark Zuckerberg and Sam Altman. He went with OpenAI, where he’ll lead their “personal agents” division. OpenClaw will technically stay an independent open-source foundation, but the optics speak for themselves: the brightest spark of recent European AI ingenuity got absorbed into the US heavyweight machine. When Big Tech comes calling with massive compute and distribution, it’s a fight Europe keeps losing.

Original thoughts on this dynamic from December below.

The ESRB Sounds the Alarm

The European Systemic Risk Board (ESRB) just dropped a report that should be required reading for anyone in Brussels. Published on December 4, it warns that the global financial system faces serious systemic risks because AI power is concentrated in the hands of a few US-based providers.

Strip away the bureaucratic language, and the message is stark: we’re building the future of the global economy on an oligopoly. Training frontier models takes so much capital and compute that the market is effectively cornered by OpenAI, Google, and Anthropic.

That creates a massive single point of failure. If one of these foundational providers goes down — whether from a model collapse, a cyberattack, a board-level governance meltdown (we’ve seen that movie before), or widespread AI “hallucinations” driving automated trading — the ripple effects could destabilize global markets overnight.

Beyond Mistral: Europe’s Fragmented Resistance

The ESRB report puts a spotlight on a tension that’s been building for a while. The EU has gotten very good at regulating AI — the AI Act is genuinely impressive as a piece of legislation — but when it comes to actually building foundational AI, the ecosystem is fragmented.

That’s not to say there’s no LLM innovation happening in Europe. Mistral AI continues to punch well above its weight, consistently shipping competitive open-weight models from Paris. Germany’s Aleph Alpha has quietly become the go-to for European government and enterprise AI. Black Forest Labs is dominating open-weight image generation. And Poolside is going all-in on foundational models for software development.

But none of this is happening at a scale that can seriously challenge the US oligopoly. These are strong, specialized players — not continent-scale contenders.

But the core vulnerability hasn’t changed: we have the talent, yet the speed at which capital actually flows in Europe is glacial compared to the US. It’s not innovation we’re short on — it’s the hyperscale infrastructure and aggressive venture funding needed to keep that innovation on this side of the Atlantic.

The Swiss Exclave: A Blueprint for Sovereign AI?

Here’s something interesting: the most viable blueprint for escaping this “single point of failure” might actually lie just outside the EU. Switzerland has taken a completely different approach — optimizing not for venture-backed hype, but for sovereignty and precision.

With its highly regulated finance and healthcare sectors, Switzerland simply couldn’t afford to depend on US-hosted APIs. So the Swiss AI Initiative — led by ETH Zurich and EPFL — used the massive “Alps” supercomputer at CSCS in Lugano to train and release Apertus in late 2025.

Apertus is a fully open, multilingual 70B-parameter foundation model, built from scratch on Swiss infrastructure and transparent right down to the training data. For Swiss enterprises and startups — think deep-tech spin-offs like EthonAI and DeepJudge — it means they can finally build next-gen applications without routing sensitive banking or legal data through servers in California.

So What Now?

If Europe is serious about tackling the systemic risks the ESRB is flagging, the Swiss model deserves a hard look. We need to stop settling for being the world’s strictest AI referee, start pooling our compute resources, and build transparent, sovereign infrastructure. Otherwise, this continent will stay what it is today: a customer of American compute — and a poaching ground for American talent.

References:

  1. Advisory Scientific Committee. (2025, December 4). Advisory Scientific Committee publishes report on artificial intelligence and systemic risk. European Systemic Risk Board.
  2. ETH Zurich. (2025). Apertus: a fully open, transparent, multilingual language model.
  3. Ha, A. (2026, February 15). OpenClaw creator Peter Steinberger joins OpenAI. TechCrunch.
  4. Steinschaden, J. (2026, February 16). OpenClaw Developer Peter Steinberger Joins OpenAI; His AI Agent Will Stay Open Source. Trending Topics.

#AIPolicy #Europe #Oligopoly #SystemicRisk #SovereignAI #ESRB #MistralAI #OpenClaw #SwissAI