The rise of large language models (LLMs) has brought unprecedented capabilities but also significant challenges. These colossal models demand immense computational power and energy, leading to soaring inference costs and limiting their deployment to powerful cloud infrastructure. This bottleneck has created an urgent need for solutions that can democratize AI, making it more accessible, affordable, and sustainable.
Enter Multiverse Computing, a Spanish deep-tech startup that has emerged as a frontrunner in addressing this critical industry tension. The company announced a monumental Series B funding round of €189 million (approximately $215 million), propelled by its groundbreaking "CompactifAI" technology. This quantum-computing inspired compression technology promises to radically reshape the economics of AI by drastically reducing the size of LLMs—by up to 95%—without compromising their performance.
Multiverse Computing's "slim" models are not merely smaller; they are engineered for efficiency. The company asserts that its compressed versions of popular open-source LLMs, such as Llama 4 Scout, Llama 3.3 70B, Llama 3.1 8B, and Mistral Small 3.1, run 4x to 12x faster than their uncompressed counterparts. This translates directly into a staggering 50% to 80% reduction in inference costs. For instance, Multiverse highlights that its Lama 4 Scout Slim model costs a mere 10 cents per million tokens on Amazon Web Services, a significant improvement over the 14 cents for the standard Lama 4 Scout.
The implications of such efficiency are profound. Multiverse's technology enables LLMs to run on a far wider array of devices, from personal computers and smartphones to cars, drones, and even low-power devices like the Raspberry Pi. This capability paves the way for truly ubiquitous AI, bringing advanced language models closer to the edge, where data is generated and real-time processing is paramount. The company is also expanding its portfolio, with a version of DeepSeek R1 slated for release soon, alongside more open-source and reasoning models. Notably, CompactifAI focuses exclusively on open-source models, with proprietary models from OpenAI and others not currently supported.
The technical prowess underpinning Multiverse Computing is rooted in deep academic and industry expertise. Co-founder and CTO Román Orús, a distinguished professor at the Donostia International Physics Center in San Sebastián, Spain, is a pioneer in the field of tensor networks. Joining Orús at the helm is co-founder and CEO Enrique Lizaso Olmos, who brings a formidable blend of mathematical acumen and extensive financial sector experience, including a notable tenure as the former deputy CEO of Unnim Bank.
This substantial Series B round was led by Bullhound Capital, a venture capital firm with a proven track record of backing disruptive technology companies such as Spotify, Revolut, DeliveryHero, Avito, and Discord. The round also saw significant participation from a diverse group of strategic investors, including HP Tech Ventures, SETT, Forgepoint Capital International, CDP Venture Capital, Santander Climate VC, Toshiba, and Capital Riesgo de Euskadi – Grupo SPR.
