
Other Europe
Paris
Selected companies
Close quarters
More AI-first software than anywhere else of its size, and the most credentialed founder base in Europe: fewer than one in ten stops at a bachelor's degree, against four in ten in London.

Advanced Machine Intelligence Labs (AMI Labs)
World Models & Embodied AIReal World. Real Intelligence.
AMI Labs builds AI systems that learn how the physical world works instead of predicting the next word. The goal is machines that can plan and act reliably, in settings such as factories, robotics and healthcare. Yann LeCun, a Turing Award winner and former Chief AI Scientist at Meta, founded the company with five others. They include CEO Alex LeBrun, who previously founded the health AI company Nabla, and COO Laurent Solly, a former Meta Europe executive. Pascale Fung, Saining Xie and Mike Rabbat are also co-founders.
Founded 2025

Blynt AI
Custom Speech RecognitionProduction-ready speech recognition for your product, within minutes
Blynt AI builds speech recognition that is tuned to each customer's own field. Off-the-shelf tools often misread specialist vocabulary, accents and noisy rooms. Blynt adapts an existing open model to a customer's setting, such as a medical clinic or a children's voice app, so it makes fewer mistakes. David Leroy led the team behind Sonos Voice Control. He founded Blynt with Alexandre Caulier, who also worked at Sonos, and Benjamin Lalanne, who previously founded a company acquired by Hopper.
Founded 2025

EGIDE
Low-Cost Drone InterceptionEurope's next-generation arsenal
EGIDE builds low-cost interceptors to stop mass-produced drones. Its case is that expensive missiles are a poor answer to cheap drones. The interceptors are electrically powered and designed for mass production. Its software, called Mystique, uses sensors and AI to detect threats in the air, at sea and on land. The company sells this as a continuously updated service. Simon Calonne, a guidance and navigation engineer, and Florian Audigier, a lethality engineer trained specifically in pyrotechnic systems integration, both worked at the missile maker MBDA before founding EGIDE together.
Founded 2025
Gradium
Real-Time Voice ModelsThe voice layer for agentic and interactive apps
Gradium builds voice technology that lets AI assistants speak and listen in real time. Its models convert text to speech and speech to text, and can clone a voice from a short sample or design a new one from a written description. The company grew out of the Paris AI lab Kyutai. Neil Zeghidour, who co-founded Kyutai and previously worked at Google Brain and Google DeepMind, founded it with Olivier Teboul, Alexandre Défossez and Laurent Mazaré.
Founded 2025
Macrodata Labs
Robotics Data InfrastructureBetter data for better robots
Macrodata Labs helps robotics teams turn raw video into data that robots can learn from. Its tools check the quality of each recording, label the steps in a task and pull hand movements out of ordinary footage. The core software is open source, with a managed cloud version for larger teams. Guilherme Penedo and Hynek Kydlíček founded the company after building FineWeb and other widely used open datasets for training language models together at Hugging Face.
Founded 2026

NPco
AI Engineering SimulationThe optimal design already exists
NPco builds AI that simulates how engineered parts behave in seconds rather than days. Engineers use it to test designs in aerospace, defence, automotive and energy without waiting on slow traditional simulations. Safran and EDF are among the industrial companies working with it. NPco is a spin-off of the French research institute Inria. Its co-founders, Matthieu Nastorg and Emmanuel Menier, both earned PhDs at Paris-Saclay on using AI for simulation.
Founded 2025
Orasio
Sovereign Video IntelligenceEuropean video AI that detects, alerts and enables action.
Orasio builds video analysis software for police, intelligence services, cities and operators of critical infrastructure. It flags events in live camera feeds and lets investigators search past footage by typing a plain-language question. It is designed to rely as little as possible on non-European technology and to meet EU data and AI rules. Florian Fournier, a co-founder of the payroll software company PayFit, founded Orasio. His co-founders are Arnaud Delaunay, a computer vision specialist, and Fabio Gennari, a former judge at France's Council of State.
Founded 2025
Thunders
Autonomous Software TestingAI Test Agents that build, run, and maintain tests
Thunders builds AI Test Agents for end-to-end software testing. Teams describe what to check in plain language. The agents build the tests, run them across web, mobile, API, desktop and enterprise systems like SAP, and keep them working as the product changes. Thunders was founded by Karim Jouini and Jihed Othmani, who previously built and sold the expense-management company Expensya in 2023.
Founded 2024
UMA Robotics
General-Purpose Humanoid RobotsWe build intelligent robots that enhance quality of life for everyone
UMA Robotics is building humanoid robots for factories and warehouses. It plans to rent them to companies for repetitive jobs that are hard to staff. The first prototype, Version 0, was built from scratch in nine months by about 30 people from Tesla, Google DeepMind and Hugging Face. The founders are Rémi Cadène, a former Tesla engineer, and Simon Alibert, the chief technology officer. The other founders are Rob Knight, a robot designer with 25 years of experience, and Pierre Sermanet, a former Google Brain and DeepMind researcher.
Founded 2025
White Circle
Runtime AI Safety & ControlRuntime AI safety and alignment infrastructure
White Circle builds software that keeps AI systems in check once they are in use. It tests for failures, enforces company rules on what an AI can say or do, and keeps a record of every decision. Lovable uses it to review what its AI does before actions run. Denis Shilov founded the company. He first became known for a jailbreak that bypassed the safety filters of the major AI models, and was then invited into Anthropic's safety programme.
Founded 2025
As early-stage VCs, most companies we look at are early, some pre-inception. To properly capture the formation layer, we looked back no more than 18 months from a company's founding, and no more than one institutional round raised.
We believe getting to the core truth of a company radiates from the founding team, and that those signs show up long before a company has its breakout success. So we got on the ground, city by city, talking directly to the ecosystem leaders, university professors, angels, operators and micro VCs who see these companies before anyone else does. The names that kept coming up independently, before they showed up anywhere public, were what we went after.
Every one of those names, alongside thousands of other companies we track a year, went through the same internal scoring model: the scale of impact the company could have globally, how genuinely AI-native the product is, the strength of the thesis, and the team, their talent density, where they've built before, how deep that experience really runs. We ended up with ten companies per city that we believe combine all of the above, while capturing the city's own heartbeat and the themes defining it.
Formations counted
569
Every city in the sample on a log scale, 10 to 1,876 formations. Paris is marked in pink. London sits at the far right as a volume outlier (1,876).
- Rank by formation volume
- 2 of 89
- SHARE OF SAMPLE EUROPEAN FORMATION
- 10.5%
- # FORMATIONS VERSUS THE AVERAGE CITY
- 9.24x
Category mix
What Paris is made of
The categories these 569 companies fall into. AI-first software is the largest.
- AI-first software26%
- Conventional SaaS25%
- Fintech11%
- Consumer9%
- Health8%
Specialisation
What Paris builds
On raw volume most European cities look alike. Character only appears once each theme is compared with a typical European city. Every company now carries a category and sub-theme. Pink marks the categories where Paris differs most; provisional comparisons rest on fewer than 10 companies and are never the lead claim.
Above the line means this city forms more of something than a typical European city does. Below means less.
- Centre line: a typical European city (1.0)
- Forms more than average
- Forms less than average
- Provisional: fewer than 10 companies
- Values read as 1.93x the average, then the theme's share of the city's companies
Typical European city (1.0)
AI-first software1.17x the average · 25.8% of the city+ 3 sub-themes
- AI tools for everyday work43.5%
- Marketing, ads & content16.3%
- Data & analytics7.5%
Conventional SaaS1.04x the average · 24.8% of the city+ 3 sub-themes
- General business software80.1%
- Marketing, ads & content7.8%
- Cybersecurity & identity5.7%
Infra0.99x the average · 7.2% of the city+ 2 sub-themes
- Cloud, compute & data infra56.1%
- Developer tooling & APIs34.2%
City 2 of 7
Next city: Berlin