Multiverse Computing funding centers on AI model compression

Multiverse Computing funding includes a reported €189 million round in June 2025 from Bullhound Capital, HP, Forgepoint Capital and Toshiba. The Donostia–San Sebastián company is selling software meant to make large AI models cheaper and easier to run, but the supplied excerpts provide no revenue, bookings, contract values or independent performance benchmarks to show how far that proposition has translated into a business.

A Reuters report republished by Forgepoint Capital said the company’s approach combines quantum physics and machine learning techniques that mimic quantum systems without requiring a quantum computer. Multiverse’s commercial focus is model compression: reducing the size and compute requirements of existing large language models during deployment.

The distinction matters. Multiverse is not presented in the available material as training a frontier model. Its pitch is to make models already in use work with less computing infrastructure, including on existing enterprise systems or smaller devices. The company calls its product CompactifAI.

What does Multiverse Computing do?

AI model compression changes a trained model so it needs less memory and processing power to run. For an operator, the intended result is lower inference cost, faster responses, or the ability to deploy a model where hardware capacity is constrained.

Multiverse says CompactifAI can cut model size by as much as 95%. Its website lists claimed results of 50% to 80% lower inference costs, up to twice the inference speed and near-total accuracy retention. In separate company material, it says compression can result in a 2% to 3% accuracy loss. Those are company claims; the supplied excerpts contain no independent tests comparing its output with competing compression methods across particular models and hardware.

Reuters reported that Multiverse had released compressed versions of Meta’s Llama, DeepSeek and Mistral models, and that its tool was available through the AWS AI Marketplace. Chief executive Enrique Lizaso said the company was concentrating on commonly used open-source models. Multiverse identifies Lizaso as its co-founder and CEO, and says it has offices in the United States and Europe in addition to its Spanish headquarters.

How much has Multiverse Computing raised?

The €189 million financing reported in June 2025 must be kept separate from a €67 million Spanish government co-investment announced in March 2025. According to an HPCwire item carrying press-release material, the investment was to be made through the Spanish Society for Technological Transformation, or SETT, which would become a shareholder. The excerpts do not establish total capital raised, ownership terms, current headcount, or a company valuation.

A Sifted profile dated August 13, 2026 said in its readable introduction that Multiverse was nearing the close of a €500 million Series C. The material does not confirm that the round closed, identify its investors, or establish its valuation. A separate social-media excerpt makes more specific claims, but provides neither documentation nor terms and is not sufficient verification.

Multiverse also has a role in Spain’s proposed AI-gigafactory consortium. The company said in July 2026 that it held a 4% stake and would serve as technology partner in a group awaiting a European Commission call. The proposal anticipates facilities in Móra la Nova, Tarragona, and Madrid, and up to €5 billion in investment mobilization. It is a prospective bid, not evidence that a facility has been selected, financed or built.

For now, the strongest verifiable case is a well-funded Spanish company pursuing deployment efficiency for open models. The supplied excerpts do not disclose the operating metrics needed to assess whether it can support broader claims of European AI leadership.

This story draws on original reporting from Sifted.