Spanish Startup Shrinks AI Models Without Losing Power
Smaller Footprint, Same Brainpower
In a move that could reshape how businesses deploy artificial intelligence, Spanish startup Multiverse Computing has cracked the code on shrinking massive AI models without sacrificing their capabilities. Their secret? Quantum-inspired compression technology called CompactifAI.
The Compression Breakthrough
The company's flagship HyperNova60B model - originally based on OpenAI architecture - now occupies just 32GB, half its previous size. Remarkably, this dramatic reduction comes with negligible impact on accuracy or performance.
"We're essentially giving these models a crash diet," explains Multiverse's chief scientist. "They keep all their knowledge and skills but take up far less space."
The latest version (HyperNova60B2602) is already available for free on Hugging Face. Multiverse has specifically optimized it for costly operations like tool calls and agent encoding - areas where businesses feel the pinch of computational expenses.
Europe's AI Contender Emerges
Like French rival Mistral AI, Multiverse represents Europe's growing ambition to compete in artificial intelligence. The company boasts an impressive client roster including Bosch, Iberdrola, and even the Bank of Canada.
Their "sovereign AI" approach has resonated politically too. Spain's Agency for Technological Transition (SETT) recently joined a $215 million Series B funding round - a strong vote of confidence in homegrown tech alternatives.
Valuation Soars Toward Unicorn Status
Though Multiverse remains tight-lipped about financial details, industry insiders report negotiations for €500 million in new funding that would value the company at over €1.5 billion.
While still dwarfed by OpenAI's $20 billion revenue machine, Multiverse's €100 million annual recurring revenue puts it within striking distance of Mistral's €400 million benchmark. With plans to open-source more compressed models throughout 2026, this Basque-region innovator appears poised to become Spain's first AI unicorn.
Key Points:
- Space saver: HyperNova60B model reduced from ~60GB to just 32GB
- Performance preserved: Accuracy remains comparable to original versions
- Enterprise focus: Optimized for cost-heavy operations like tool calls
- European champion: Backed by governments seeking tech independence
- Growth trajectory: Potential €1.5B valuation signals rising prominence
