Spanish deeptech company that shrinks large language models wants investors to bet that efficiency, rather than sheer scale, is where the next stretch of AI money gets made.

Multiverse Computing, based in San Sebastián in Spain’s Basque Country, has opened a Series C round targeting up to $570m (€500m), the company said on July 27.

The raise would value the startup at roughly $1.7bn (€1.5bn) before the new money arrives, which the company put at about five times its previous valuation.

The pitch rests on CompactifAI, a tool Multiverse says can compress a large language model by up to 95% with what it calls immaterial accuracy loss. It borrows tensor networks from quantum physics to strip redundancy out of a model, cutting the memory, cost, and energy each query burns.

The technique sits inside a broader compression race that has pulled in far larger players.

In practice, the company sells slimmed-down versions of open models such as Meta’s Llama, packaged to run on cheaper hardware or on-premises rather than inside a hyperscaler’s data centre.

That edge-to-cloud framing is the heart of the Series C pitch. As inference, the cost of actually running a model, overtakes training as the dominant expense for many buyers, the argument goes, they will pay for the same output at a fraction of the compute and power.

It is not the only use Multiverse has found for the maths. The firm, which began in quantum software, has applied the same methods to other problems, including a system to predict floods. The through line is squeezing more out of less, whether the target is a neural network or a river.

The round is co-led by Forgepoint Capital International, the BNPP Solar Impulse Venture Fund, and Bullhound Capital, according to the company. If it closes at the top of the range, Multiverse’s total funding would reach about $800m across all rounds.

The company has not said when it expects to finish raising, nor disclosed the terms attached to the lead investors’ stakes. A round that is announced while still open can also close below its target.

In June 2025 the firm closed a Series B worth €189m, or about $215m, led by Bullhound with backing from HP Tech Ventures, Toshiba, Forgepoint, SETT, and Spain’s CDP Venture Capital. Bloomberg reported in February that Multiverse was in talks at a €1.5bn valuation, the same figure now attached to the Series C.

Multiverse was founded in 2019 by chief executive Enrique Lizaso, a former banker, and chief scientific officer Román Orús, a physicist whose work on tensor networks underpins the product. Both the compression business and its quantum tools sit inside a wider European quantum push that has drawn heavy public funding.

Its customer list, per the company, runs to Iberdrola, Bosch, Telefónica, Allianz, Bank of Canada, Indra, and PwC.

Multiverse also reported steep growth to match its ambitions, claiming a tenfold rise in annualised revenue since the last round and first-quarter sales up 96 times year on year. Those figures are self-reported and have not been independently audited.

The bet is a distinctly European one. As models grow more expensive to run, a cluster of the region’s startups is chasing the inference bill rather than the frontier.

Data-centre electricity demand has become one of the industry’s hardest constraints, and trimming the energy a model burns per query is one of the few levers a startup can pull without owning chips or power stations.

Multiverse is selling the opposite trade. If the future of AI is smaller, cheaper, and closer to the edge, a compression startup on the Basque coast would rather own that corner than chase the frontier it is quietly trying to cut down to size.

Get the TNW newsletter

Get the most important tech news in your inbox each week.