Multiverse Computing, a startup specializing in quantum-inspired AI technologies, announced a €189 million ($215 million) Series B funding round, one of Europe’s largest AI-focused investments in 2025 so far.
The company’s breakthrough is a technology called CompactifAI, which promises to shrink Large Language Models (LLMs) by up to 95%, without sacrificing performance.
Multiverse’s core innovation challenges a long-held belief in AI circles: that smaller models inevitably perform worse. Traditionally, AI model compression relied on methods like quantization (using lower precision numbers) or pruning (removing less important connections).
While these cut down size, they also degrade model accuracy or flexibility, a tough trade-off in sensitive applications like healthcare, finance, or autonomous driving. But CompactifAI claims to break that trade-off entirely. Through tensor networks, a computational technique borrowed from quantum physics that simplifies extremely large mathematical structures, like those inside an LLM.
Instead of brute force, this method gently untangles the model’s inner relationships, stripping away redundant data without disturbing the important parts.
“For the first time in history, we are able to profile the inner workings of a neural network to eliminate billions of spurious correlations to truly optimize all sorts of AI models,” explains Román Orús, Multiverse’s CTO and co-founder, who also happens to be a professor of physics at the prestigious Donostia International Physics Center.
In plain terms: imagine cleaning out a library by removing only useless duplicate books while leaving all the classics perfectly untouched. The result is more shelf space, faster browsing, and no loss of wisdom.
Faster, cheaper, greener AI
The real-world benefits of this “slimming” are eye-popping. Multiverse says its compressed models, called “Slim models” are 4 to 12 times faster than their full-sized counterparts and can reduce inference costs by 50% to 80%. For example, its compressed version of Llama 4 Scout costs 10 cents per million tokens on AWS, down from the standard 14 cents.
Speed and cost aren’t the only advantages. The models also become energy-efficient enough to run on PCs, smartphones, cars, drones, and even the humble Raspberry Pi. In an age where the environmental impact of AI training and inference is under growing scrutiny, that energy saving could prove critical.
According to research from the University of Massachusetts, training a single large AI model can emit as much carbon as five cars over their lifetimes. CompactifAI’s promise of small, efficient models could radically change that equation.
“The prevailing wisdom is that shrinking LLMs comes at a cost. Multiverse is changing that,” said Enrique Lizaso Olmos, the company’s CEO and co-founder, who also has an unusual career path for a tech founder, holding multiple mathematics degrees and having once served as deputy CEO at Spain’s Unnim Bank.
The size of this Series B round reflects how deeply investors believe Multiverse could reshape the AI infrastructure market, currently estimated at $106 billion globally for inference alone.
Leading the round was Bullhound Capital, an investment firm with a history of backing winners like Spotify, Revolut, and Slack. Also joining were HP Tech Ventures, Toshiba, Forgepoint Capital, Santander Climate VC, and several European funds tied to digital transformation, including Spain’s SETT and the regional Capital Riesgo de Euskadi.
“The Multiverse team has solved a deeply complex problem with sweeping implications. The company is well-positioned to be a foundational layer of the AI infrastructure stack. Multiverse represents a quantum leap for the global deployment and application of AI models, enabling smarter, cheaper and greener AI. This is only just the beginning of a massive market opportunity.” commented Damien Henault of Forgepoint Capital.
Even HP Inc., the PC and printer giant, sees potential for bringing AI closer to everyday devices. “At HP, we are dedicated to leading the future of work by providing solutions that drive business growth and enhance professional fulfillment. Our investment in Multiverse Computing supports this ambition. By making AI applications more accessible at the edge, Multiverse’s innovative approach has the potential to bring AI benefits of enhanced performance, personalization, privacy and cost efficiency to life for companies of any size” said Tuan Tran, HP’s President of Technology and Innovation.
Today, most LLMs require vast data centers filled with expensive GPUs. This centralization favors the richest tech firms, Google, OpenAI, Microsoft, while locking out startups, researchers, and small enterprises.
If CompactifAI delivers on its promise, that bottleneck could crack wide open.
Imagine a local business running sophisticated AI tools on a laptop, or a self-driving car using advanced language models without the cloud. A drone programmed to understand complex commands. A Raspberry Pi-powered Christmas display with an LLM-powered talking Santa (yes, even that got mentioned in Multiverse’s release).
For now, Multiverse focuses on open source models like Meta’s Llama 3 and 4, Mistral, and soon DeepSeek R1. Notably, the company isn’t compressing proprietary models from OpenAI or Anthropic, perhaps a quiet hint that its tech favors transparency and openness.
Though inspired by quantum physics, CompactifAI runs on classic computers, not quantum ones. Its methods come from tensor networks, a field originally developed to simulate quantum systems, Orús’ academic specialty.
Multiverse’s compression avoids the major technical obstacles of building real quantum hardware, delivering benefits today, not years from now.
There’s another layer here: geopolitics. As the US and China dominate AI, Europe has sought to carve out space through sovereign technology initiatives. Multiverse could fit perfectly into this ambition.
With over 160 patents, 100 customers globally, and awards like CB Insights’ Top 100 AI Companies for 2025, Multiverse seems poised to lead this charge.
