Multiverse Computing secures 215 million dollars for breakthrough in AI compression – language models are reduced by 95% with minimal performance loss.
Spanish startup Multiverse Computing has raised €189 million ($215 million) in a Series B funding round to scale its CompactifAI technology. The funding was led by Bullhound Capital, with participation from HP Tech Ventures, Toshiba and other investors. The valuation of the company rose to over 500 million dollars – a five-fold increase compared to the Series A round of 2024.
The quantum-inspired compression technology CompactifAI reduces large language models (LLMs) by up to 95% of their original size while retaining almost all of their performance. In contrast to conventional methods, which often cause 20-30% accuracy loss with 50-60% compression, CompactifAI achieves 93% size reduction with only 2-3% precision loss.
The cost savings are significant: inference costs are reduced by 50-80% and processing speed increases by 4-12 times. Multiverse’s compressed Llama 3.3-70B Slim version retains 97% of the original model accuracy with 60% fewer parameters and 84% lower energy consumption.
Quantum-inspired tensor networks as key technology
The proprietary technology uses tensor networks from quantum physics to identify and remove redundant parameters in LLMs. This method decomposes weight matrices into optimized tensor structures and prunes unnecessary correlations. After compression, the models undergo retraining to recover lost performance.
The technology allows compressed models such as Mistral Small 3.1 Slim to run on edge devices such as smartphones, Raspberry Pi boards and drones. For example, a Raspberry Pi 5 achieves 5.13 tokens per second with a compressed phi-2-Q4 model – comparable to the performance on more powerful hardware.
Strategic investors and market positioning
Bullhound Capital led the funding round and brings experience from successful investments in Spotify and Revolut. HP Tech Ventures’ investment signals strategic interest in embedding efficient AI into HP’s hypermobility and IoT businesses. Toshiba’s involvement underscores the industrial application opportunities in energy and manufacturing.
With 160 patents and over 100 global customers, Multiverse has grown from its origins in quantum computing for finance to become a market leader in AI optimization. Revenue is doubling annually, driven by license fees and a new AWS partnership that provides API access to compressed models.
Executive Summary
- Multiverse Computing raises $215 million in Series B round, valuation rises to over $500 million
- CompactifAI technology reduces LLMs by 95% with only 2-3% accuracy loss
- Quantum-inspired tensor networks enable efficient model compression
- Inference costs decrease by 50-80%, processing speed increases by 4-12 times
- Compressed models run on edge devices such as smartphones and Raspberry Pi
- Bullhound Capital leads financing round, HP Tech Ventures and Toshiba participate
- Over 100 global customers already using the technology, including Iberdrola and Bosch
- AWS partnership provides API access to compressed models
Source: TechChrunch