Mozilla’s CTO thinks AI should be built like the internet
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Mozilla reported last month that the performance gap between top open-source models and proprietary systems like Claude and ChatGPT has narrowed to 3%.
- Alibaba's open-source model Qwen was downloaded more times in February 2026 than the next eight models combined, illustrating what Krikorian called a 'multi-hundred-billion-dollar commercial layer.'
- Meta released a new open-weight model on August 10 with more promised, adding fresh momentum to the open-AI debate.
- Mozilla CTO Raffi Krikorian argued Washington is making a 'categorization error' by treating AI as a product to rent rather than infrastructure to build — a framing he said the rest of the world does not share.
- Enterprise IT and HR teams are quietly migrating workflows to open models for price-performance and to keep data within their own firewalls, especially in regulated industries, according to Krikorian.
- Krikorian drew a sharp distinction between true open-source models — which require full training-data transparency — and the open-weight models (downloadable, modifiable, but with opaque training data) that dominate today's deployment landscape.
Why it matters: Mozilla's data shows open-source models now trail Claude and ChatGPT by just 3%, and Krikorian says enterprise IT teams are already migrating to them for price-performance and data sovereignty. His argument: US policy that backs Anthropic and OpenAI treats AI as a product to rent, while the rest of the world wants to buy it as infrastructure — and that mismatch is the real story.
Ask SkimNews
