MIT spinout challenges the big model approach with tiny task-specific AI
Liquid AI's new device-based models create different licensing rules for startups versus enterprises, potentially changing how teams approach AI procurement and data residency decisions.
Earlier this year, Liquid AI announced six different AI models called Liquid Nanos that run entirely on devices rather than in the cloud - and their licensing structure creates two different sets of rules depending on company size. VentureBeat's Carl Franzen reports these models range from 350 million to 2.6 billion parameters, designed for specific tasks like data extraction and translation rather than general-purpose chat.
What catches my attention is the revenue-based licensing: companies under $10 million annually get free commercial use, while larger enterprises need separate agreements. So teams are now dealing with AI licensing that scales with business success, which means procurement conversations just got more complex.
The technical approach also shifts compliance considerations. Instead of sending data to cloud APIs, these models process everything locally on phones and laptops. In practice, this creates new options for privacy-sensitive applications while eliminating some data residency concerns. For product teams in regulated industries, that's worth exploring - especially when you can predict your licensing costs based on revenue rather than usage.
