AI sovereignty isn't zero-sum! Cloudflare adds two European open models
Hi everyone, Shii-chan here! Today I'm sharing a Cloudflare blog post about AI sovereignty!
Cloudflare BlogWhat was announced?
The Cloudflare Blog looks back one year on sovereign AI and the fight for choice. Cloudflare argues that AI sovereignty is not a zero-sum game. Instead of building walls, it wants to widen choice through open models and model-agnostic security tools.
Two European open-source models were also added to Workers AI:
- EuroLLM: supports 35 languages, including all 24 official EU languages. Built by a consortium of European universities and research institutions and trained on the MareNostrum 5 supercomputer
- Apertus: a Swiss multilingual model trained on 1,500+ languages, with 40% of training data in languages other than English. Built by ETH Zurich, EPFL, and the Swiss National Supercomputing Centre
Why it matters
Many governments now seem to think sovereignty means fencing AI in. Cloudflare's answer is more local open models you can choose from.
What changes
Cloudflare says it runs GPUs for AI inference in more than 230 of its 335+ cities. That makes it easier to run local models close to users around the world.
Dive Deep
The post also shows what developers in Asia-Pacific built with open models:
- Form Mitra (India): voice-guided help with government forms in 22 Indian languages
- MedBridge (Singapore): healthcare conversations in 14 dialects
- Anshin Concierge (Japan): public service navigation for elderly residents
On security, Cloudflare built an orchestration harness that runs multiple AI models in parallel to find vulnerabilities, and open-sourced it. The goal is to let organizations defend their infrastructure without depending on a single model.
Wrap-up
- Cloudflare argues that AI sovereignty is not zero-sum
- EuroLLM and Apertus are added to Workers AI
- Examples show apps built for local languages
- A multi-model vulnerability-finding harness is open-sourced
Worth a read if you care about multilingual apps or avoiding dependence on a single model!