For years, the narrative about European AI has been the same: the US builds, Europe regulates. Silicon Valley innovates, Brussels legislates. The continent’s job is to write rules, not set them.
That story is getting old. And it might be wrong.
When investors at some of Europe’s biggest VC funds sat down to name their top AI startup picks recently, the results painted a very different picture. Not a single one mentioned GPT or Gemini. Instead, they pointed to a Prague startup building foundational models with a joke name, a Spanish rocket company, a German nuclear fusion play, and a Welsh startup manufacturing semiconductors in orbit.
Europe’s AI future isn’t about catching up to the US. It’s about building things the US isn’t building at all.
🛡️ The Defence Boom
The war in Ukraine has turned European defence tech from a pariah sector into the hottest ticket in town. Helsing, the Munich-based AI defence startup, is now worth €12 billion — one of Europe’s five most valuable private tech companies. Its autonomous strike drones are already in use by several European militaries, and it just demonstrated its self-driving system piloting a Saab fighter jet.
But Helsing is the headline act. Beneath it, a whole ecosystem is growing:
Alta Ares (Spain) builds AI-powered counter-drone interceptors — cheap solutions for detecting and shooting down drone incursions. Inbolt (France) brings physical AI to factory floors, already active in 70+ manufacturing plants. ARX Robotics (Germany) raised €31 million for autonomous land drones, with a new UK factory in the works.
The numbers tell the story: European defence tech investment rose 500% between 2021 and 2024 compared to the previous three years, according to McKinsey. That’s not a blip. That’s a structural shift.
🔬 Fusion, Space Chips, and Photonics
The most surprising entries on the list are the deep-tech plays — the kind of long-horizon, capital-intensive work that VCs usually avoid.
Proxima Fusion (Germany) just secured $460 million from the state of Bavaria to build a demonstration stellarator fusion power plant near Munich. Not a PowerPoint. A power plant. PLD Space (Spain) launched a suborbital rocket in 2023, raised $209 million from Mitsubishi Electric, and is now developing a reusable orbital launcher for small satellites. Space Forge (Wales) is working on in-space semiconductor manufacturing — building chips in microgravity because it turns out gravity messes with crystal formation.
Cailabs (France) develops photonics for faster data transmission and plans 50 optical ground stations for laser satellite communications. Optics11 (Netherlands) monitors underwater infrastructure with fibre-optic sensing — the kind of tech that matters when undersea cables become geopolitical targets.
These aren’t AI companies in the “another chatbot wrapper” sense. They’re companies using AI to solve physics, materials science, and manufacturing problems that have real-world, physical outcomes.
🧠 The Foundation Layer
Europe is also building at the model level.
BottleCap AI (Czech Republic) — yes, that’s the real name — is developing efficiency-focused foundational LLMs and companion apps. Founded by a team that includes an entrepreneur who sold his previous company to Meta and two AI researchers, BottleCap takes a dual approach: build the models, build the apps on top.
Multiverse Computing (Spain) takes a different tack. Instead of training new models, it compresses existing ones — taking open-weight models from OpenAI, Meta, DeepSeek, and Mistral, and making them smaller and cheaper to run on local hardware. Co-founded by a physics professor, it’s raised $250 million.
Gradium (France), a spinout from Kyutai, is building real-time text-to-speech models that give AI agents natural voices across multiple languages. They raised a $70 million seed round — because apparently competing with ElevenLabs requires serious capital.
Fundamental (UK) emerged from stealth in February with a $255 million Series A at a $1.4 billion valuation. Its Nexus model helps enterprises draw insights from their own data — the kind of boring, lucrative problem that big companies will actually pay for.
🤖 Agents, Automation, and the Boring Stuff That Makes Money
Not every startup on the list is building rockets or fusion reactors. Some are just solving boring problems really well — which, in startup terms, is often where the real money is.
HappyRobot (Spain/US) builds AI agents for complex enterprise workflows, backed by a16z and Y Combinator. Apron (UK) handles invoice management for small businesses. Pennylane (France) is building a unified financial operating system for European SMBs — and it’s already a unicorn. Legora (Sweden/NY) is taking on Harvey in legal AI, and enlisted Jude Law as its brand face. Yes, that Jude Law.
Botify (France) has pivoted from traditional SEO to “generative engine optimization” — helping brands show up in AI search results. Clients include Macy’s and The New York Times. Flower (Sweden) uses AI and battery storage to make renewable energy more predictable. It just raised $60 million in bonds.
🔍 The Bottom Line: Europe Isn’t Playing Catch-Up
There’s a temptation to read this list and think “cute, but none of these will be the next OpenAI.” And maybe that’s true. But that’s also the wrong question.
Europe’s AI strength isn’t in building general-purpose foundation models that compete with GPT-6 or Gemini 3. It’s in applying AI to physical industries — defence, energy, manufacturing, space — where the US tech sector has less domain expertise and where the regulatory environment (yes, the one Europe is famous for) actually creates moats.
Helsing didn’t get to €12 billion by being “the European OpenAI.” It got there by being the best at putting AI on military drones and submarines. PLD Space didn’t get Mitsubishi’s money by building a better chatbot. It built a reusable rocket.
The lesson for Kiwi readers? Same playbook. New Zealand doesn’t need to compete on foundation models. But applying AI to agriculture, renewable energy, ocean tech, and disaster response? That’s a game NZ could win — and these European startups show exactly how.
📋 The Full 21
- Alta Ares (Spain) — AI counter-drone systems
- Apron (UK) — Invoice management for SMBs
- Botify (France) — Generative engine optimization
- BottleCap AI (Czech Republic) — Efficiency-focused LLMs
- Cailabs (France) — Photonics for data transmission
- Cala (Spain) — Knowledge graphs for AI agents
- Flower (Sweden) — AI-powered renewable energy management
- Fundamental (UK) — Enterprise data foundation models
- Gradium (France) — Real-time TTS for AI agents
- HappyRobot (Spain/US) — Enterprise AI agents
- Helsing (Germany) — Autonomous defence AI
- Inbolt (France) — Physical AI for manufacturing
- Legora (Sweden/NY) — AI platform for lawyers
- Macrodata Labs (France) — AI training data infrastructure
- Multiverse Computing (Spain) — Model compression
- Optics11 (Netherlands) — Fibre-optic sensing
- Pennylane (France) — SMB financial OS
- PLD Space (Spain) — Reusable orbital rockets
- Proxima Fusion (Germany) — Nuclear fusion power plants
- Roofline (France) — AI chip deployment software
- Space Forge (Wales) — In-space semiconductor manufacturing
