Spanish startup Multiverse Computing has secured a massive €189 million (approximately $215 million) in a Series B funding round, all thanks to a breakthrough technology it calls CompactifAI. This innovation is designed to drastically shrink the size of large language models (LLMs) by as much as 95%—without compromising their performance, the company claims. CompactifAI is inspired by principles from quantum computing and uses tensor networks to compress AI models. These networks replicate the behaviour of quantum systems on classical hardware, making them a powerful tool for deep learning optimisation. The result: faster models with reduced hardware demands.

Also Read: Apple Is Making Your iPhone a Digital Babysitter—Here’s How

Smaller Models, Bigger Efficiency

Rather than creating entirely new AI models, Multiverse focuses on compressing existing open-source LLMs. Some of the main models already optimised through CompactifAI include Llama 4 Scout, Llama 3.3 70B, Llama 3.1 8B, and Mistral Small 3.1. The company plans to expand its offerings soon, with compressed versions of DeepSeek R1 and additional open-source and reasoning models. However, proprietary models like those from OpenAI remain unsupported for now.

Also Read: Mizoram Joins the Railway Map After Years of Waiting

These compressed versions—dubbed “slim” models—are not only smaller but also faster and more cost-efficient. Multiverse claims they perform 4 to 12 times faster than their full-sized counterparts. This efficiency translates into significant cost savings, with inference expenses reportedly reduced by 50% to 80%. For instance, its slimmed-down Llama 4 Scout costs just 10 cents per million tokens on AWS, compared to the original's 14 cents.

For anyone tracking the evolution of AI infrastructure, this might be the beginning of a slimmer, smarter future.