Europe has the talent.
Let's build AI together.
ORCAI is a grassroots collective of European researchers, engineers and volunteers training open-weight AI models: using distributed compute donated by people like you. Think Folding@home, but for AI. No corporations. No bureaucracy. Just talent, compute and a shared mission.
Europe has world-class AI talent.
But we're falling behind.
The biggest AI models come from the US and China. Not because they have better researchers, but because they have better coordination. Europe's talent is scattered across universities, startups and labs, often stuck in long grant cycles and slow decision processes. Meanwhile, the world moves on.
Grants take forever
A Horizon Europe proposal takes 12–18 months from idea to funding. In AI, that's an eternity. By the time the money arrives, the field has moved on three times.
Consortia are slow
Big EU projects mean 20+ partners, endless coordination calls, and decisions by committee. By the time everyone agrees on a data format, three new model architectures have come and gone. Grassroots moves faster.
Dependency is a risk
If Europe can't train its own models, we depend on US and Chinese companies for core AI capabilities, safety standards, multilingual support and access policies.
Grassroots. Open. Fast.
We don't wait for permission. We start with what we have (talented people and spare GPUs) and build from there. Like Folding@home pooled millions of computers to fight disease, we pool compute to train AI models.
Volunteer Distributed Compute
Got a GPU sitting idle? Donate cycles to train European AI models. We use frameworks like Hivemind, Petals and DiLoCo to coordinate training across thousands of machines. No data center required.
Open-Weight Models
Every model we train gets released with open weights, full training recipes and reproducible configs. Open-weight means anyone can inspect, fine-tune and deploy. No vendor lock-in.
Ship Fast, Ship Often
Projects run in 30–60 day sprints. A named lead, clear goals, and a deadline. If a project stalls, we archive it and move on. No zombie projects, no endless meetings.
Security Built In
Volunteer compute needs trust. We use signed workloads, sandboxed execution, redundant verification and a two-lane system: safe tasks for volunteers, sensitive work for partner clusters.
Europe doesn't lack talent.
It lacks coordination.
From CERN to DeepMind London to ETH Zürich Europe trains some of the world's best AI researchers. ORCAI gives them a way to collaborate on frontier AI outside institutional silos.
Who we're looking for
- ML researchers: training, evaluation, alignment
- Infrastructure engineers: distributed systems, GPU clusters
- Data specialists: curation, multilingual, quality
- Security engineers: sandboxing, verification, trust
- Community builders: outreach, documentation, ops
- Anyone with a GPU: donate compute cycles
Our inspirations
We're not the first grassroots collective to punch above its weight.
- CERN: European collaboration that invented the World Wide Web and proved big science works across borders
- LAION: tiny German nonprofit → LAION-5B, the dataset behind Stable Diffusion
- Folding@home: volunteer compute → 2.4 exaFLOPS during COVID, more than any supercomputer
- BigScience/BLOOM: 1,000+ researchers → BLOOM-176B, the first multilingual open LLM
- OpenEuroLLM: 20 European institutions building open multilingual models (launched 2025)
Be part of something bigger.
ORCAI is forming its founding cohort right now. Whether you can contribute research, engineering, data expertise or just GPU cycles, there's a place for you. No institution required. No credentials check. Just show up and do the work.