{"id":150320,"date":"2025-06-14T01:05:49","date_gmt":"2025-06-14T01:05:49","guid":{"rendered":"https:\/\/teknomers.com\/en\/a-basque-ai-startup-has-just-raised-189-million-euros-with-a-brilliant-idea-compressing-ai\/"},"modified":"2025-06-14T01:05:50","modified_gmt":"2025-06-14T01:05:50","slug":"a-basque-ai-startup-has-just-raised-189-million-euros-with-a-brilliant-idea-compressing-ai","status":"publish","type":"post","link":"https:\/\/teknomers.com\/en\/a-basque-ai-startup-has-just-raised-189-million-euros-with-a-brilliant-idea-compressing-ai\/","title":{"rendered":"A Basque AI startup has just raised 189 million euros with a brilliant idea: compressing AI."},"content":{"rendered":"\n

Multiverse Computing: Revolutionizing AI Compression<\/h2>\n

In the past, we compressed files with ZIP. Now, what we increasingly need is to compress \u00a0AI\u00a0 to make it smaller and more efficient. This is exactly the idea behind \u00a0Multiverse Computing\u00a0, a Spanish startup that is becoming the new crown jewel of our AI industry. Its founders (in the image, from left to right, \u00a0Rom\u00e1n Or\u00fas\u00a0, \u00a0Enrique Lizaso Olmos\u00a0, and \u00a0Samuel Mugel\u00a0) alongside \u00a0Alfonso Rubio\u00a0 have a lot to celebrate.<\/p>\n

<\/p>\n

Investment Round Success<\/h2>\n

Multiverse Computing<\/strong> has recently closed an investment round of \u00a0189 million euros\u00a0 (215 million dollars). The round, labeled \u00a0Series B\u00a0, was led by \u00a0Bullhound Capital\u00a0, with participation from major players such as \u00a0HP Tech Ventures\u00a0, \u00a0SETT\u00a0, \u00a0Forgepoint Capital International\u00a0, \u00a0CDP Venture Capital\u00a0, \u00a0Santander Climate VC\u00a0, \u00a0Quantonation\u00a0, \u00a0Toshiba\u00a0, and \u00a0Capital Riesgo de Euskadi – Grupo SPRI\u00a0. Earlier in March, the company received an investment of \u00a067 million euros\u00a0 from the Government of Spain.<\/p>\n

<\/p>\n

\n
\n
\n
<\/div>\n<\/p><\/div>\n<\/p><\/div>\n<\/div>\n

AI Inference at Its Core<\/h2>\n

While the spotlight currently shines on \u00a0big tech companies\u00a0 investing billions of dollars in data centers to train large language models (LLMs), there is an increasing focus on the other side: the part we users engage with when asking questions to AI systems like \u00a0ChatGPT\u00a0. This is known as \u00a0AI inference\u00a0, and it is estimated that by 2025, the value of this industry will soar to \u00a0106 billion dollars\u00a0. Multiverse Computing aims to claim a significant share of this market with its unique technology.<\/p>\n

<\/p>\n

Introducing CompactifAI<\/h2>\n

CompactifAI<\/strong> is the name of the AI model compression technology developed by Multiverse Computing. This technology enables the conversion of monolithic AI models\u2014those that are costly to “run”\u2014into far smaller and more efficient models, making them more manageable and saving substantial resources (and time) during inference.<\/p>\n

<\/p>\n

How to Compress an AI Model<\/h2>\n

Rom\u00e1n Or\u00fas, the company\u2019s scientific director, led a study<\/a> in May 2024 that explained the concept of \u00a0tensor networks\u00a0, inspired by quantum principles, which allow for the compression of these models. The process involves decomposing the weight matrices of neural networks by “truncating” and retaining only the most significant values. Essentially, the concept revolves around discarding less relevant information to focus on what truly matters within the model.<\/p>\n

<\/p>\n

Does This Compromise Model Accuracy?<\/h2>\n

Indeed, it can, but the degree of truncation is controllable to strike a balance between compression and precision loss. Despite compressing these models, Multiverse Computing asserts that the drop in accuracy is merely between \u00a02% to 3%\u00a0.<\/p>\n

<\/p>\n

Same Performance at 95% Smaller Size<\/h2>\n

To counteract potential accuracy declines, this system includes a rapid retraining phase known as \u00a0“curation,”\u00a0 which can be repeated multiple times to achieve an accuracy even closer to the original model. Ultimately, the company claims they can compress an AI model by up to \u00a095%\u00a0 while maintaining performance.<\/p>\n

<\/p>\n

Making AI More Affordable<\/h2>\n

According to their data<\/a>, a model like \u00a0Llama 3.1 405B\u00a0 incurs an operational cost of around \u00a0390,000 dollars\u00a0 when running locally (needing \u00a013 GPUs H100\u00a0 and drawing \u00a09100 W\u00a0). However, with the help of CompactifAI, that cost can be slashed to just \u00a060,000 dollars\u00a0 (requiring only \u00a02 GPUs H100\u00a0 and consuming \u00a01400 W\u00a0).<\/p>\n

<\/p>\n

\n
\n
\n
\n \"Meta\n <\/div>\n<\/p><\/div>\n<\/p><\/div>\n<\/div>\n

Slim AI Models<\/h2>\n

The “slim” models provided by the company\u2014derived from \u00a0Llama 3.3 70B\u00a0 or \u00a0Llama 4 Scout\u00a0\u2014are compressed versions that theoretically maintain accuracy. They can be executed through the \u00a0AWS\u00a0 platform or via licenses that also allow for \u00a0on-premise\u00a0 use, meaning local infrastructure. According to their metrics, these models run between \u00a04 and 12 times\u00a0 faster than their non-compressed counterparts, translating to an inference cost that is \u00a050% to 80% lower\u00a0.<\/p>\n

<\/p>\n

Image | Multiverse Computing<\/p>\n

As Multiverse Computing continues to innovate in the realm of AI, the implications for businesses and consumers alike could be monumental. By reducing operational costs and enhancing efficiency in AI application, they are not only paving the way for future advancements but are also enabling broader and more accessible AI utilization across various sectors.<\/p>\n


\n
General News – 2<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Multiverse Computing: Revolutionizing AI Compression In the past, we compressed files with ZIP. Now, what we increasingly need is to compress \u00a0AI\u00a0 to make it smaller and more efficient. This is exactly the idea behind \u00a0Multiverse Computing\u00a0, a Spanish startup that is becoming the new crown jewel of our AI industry. Its founders (in the […]<\/p>\n","protected":false},"author":1,"featured_media":150321,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[36399],"tags":[36967,21561,36968,4849,5700,679,2513,12272],"class_list":["post-150320","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technology","tag-basque","tag-brilliant","tag-compressing","tag-euros","tag-idea","tag-million","tag-raised","tag-startup"],"_links":{"self":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts\/150320","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/comments?post=150320"}],"version-history":[{"count":0,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts\/150320\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/media\/150321"}],"wp:attachment":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/media?parent=150320"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/categories?post=150320"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/tags?post=150320"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}