Multiverse Computing pushes its compressed AI models into the mainstream

Amidst growing concerns about the financial stability of the AI supply chain, Spanish startup Multiverse Computing is positioning on-device processing as a viable alternative to reliance on external cloud infrastructure. The company has launched a self-serve API portal to provide developers with direct access to its compressed AI models. This move comes as venture firms like Lux Capital advise companies to secure compute capacity in writing due to high default rates, highlighting a market appetite for solutions that mitigate counterparty risk by running AI locally.

Multiverse’s core offering is its quantum-inspired compression technology, showcased in its CompactifAI chat application. The app is designed to run a small model named Gilda locally on a user's device, ensuring data privacy and offline functionality. However, it incorporates a system called 'Ash Nazg' that automatically routes requests to cloud-based models if the device lacks sufficient RAM or storage, a limitation for older hardware. The new API portal bypasses this consumer-facing layer, granting enterprises direct access to compressed models from labs like OpenAI and Mistral AI, complete with tools for monitoring real-time usage and controlling production deployments.

While its app has seen limited consumer adoption, Multiverse's primary focus is the enterprise market. The company’s technology addresses key business needs for lower compute costs, enhanced data privacy, and operational resilience in settings with poor connectivity, such as drones or satellites. With a client list that already includes Bosch and the Bank of Canada, and a rumored €500 million funding round in the works, Multiverse is capitalizing on a broader industry shift towards more efficient, specialized AI models. This trend is also evident in efforts by competitors like Mistral AI, which recently updated its own family of small models optimized for enterprise tasks.

By launching a direct-access API, Multiverse Computing is moving beyond technical demonstrations to build a commercial platform that directly targets enterprise needs for AI solutions independent of major cloud providers, capitalizing on concerns around cost, privacy, and supply chain stability.