AI Goes Local: Small Models Break Free From Cloud

Facing financial instability in the AI supply chain, VC firm Lux Capital advised AI companies to secure compute capacity in writing. An alternative emerges from Multiverse Computing, a Spanish startup developing AI models optimized for on-device execution. Their CompactifAI technology compresses large AI models, enabling them to run locally and offline. The CompactifAI app provides a glimpse of this edge AI, ensuring user data remains on the device, though sufficient RAM and storage are necessary.

When local processing isn't feasible, the app automatically routes requests to cloud-based models. This hybrid approach, managed by their 'Ash Nazg' system, retains privacy benefits only when local processing is active. The primary focus, however, is on enterprise clients. Multiverse is launching a self-serve API portal, granting developers direct access to their compressed models without relying on third-party marketplaces.

This move aims to address the growing demand for AI efficiency and lower compute costs. Multiverse's latest compressed model, HyperNova 60B 2602, reportedly offers faster responses and lower costs than its OpenAI-derived base. The company already serves over 100 global customers, including notable financial and industrial firms, and is reportedly seeking a significant new funding round.