Multiverse Computing funding hits $570M at $1.7B valuation
Multiverse Computing raised a $570M Series C to expand AI model compression, with a $1.7B pre-money valuation after its 2025 Series B.
By Dominic Okoye · Staff Writer
· 3 min read
Multiverse Computing funding reached $570 million in a Series C round that values the AI model compression startup at $1.7 billion before the new capital, the company announced Monday. The round gives the Spain-based company more capital to pursue a core enterprise AI problem: making large models cheaper to run across CPUs, edge devices and private infrastructure.
The Series C was co-led by Forgepoint Capital International, BNPP SIVF and Bullhound Capital. Multiverse said Santander Alternative Investments, Tikehau Capital, HP Inc., Orange Ventures, Scania Invest and NAventures also participated, along with several other investors. The company did not disclose revenue, headcount or the ownership sold in the round.
The valuation is a sharp increase from Multiverse’s last disclosed financing. The company raised a $215 million Series B in June 2025, and the new pre-money valuation represents close to a fivefold step-up from that prior round, according to SiliconANGLE.
What does Multiverse Computing do?
Multiverse sells AI model compression technology designed to reduce the compute and memory burden of running large models. Its main product, CompactifAI, uses tensor networks, a mathematical approach associated with quantum physics, to shrink models so they can run with less hardware.
The company claims CompactifAI can cut the hardware footprint of large language models by 80% to 95% while causing only limited accuracy degradation. That claim is central to the business case: if the compressed models perform well enough, customers can lower inference costs, reduce energy use and run workloads on hardware that would otherwise be too constrained for large models.
Multiverse has positioned the technology for both edge and enterprise deployments. Local inference can reduce latency because prompts and responses do not need to travel to a cloud service, and it can keep sensitive data on customer-controlled hardware. That is the practical pitch for buyers in regulated industries and for companies trying to limit dependence on hyperscaler infrastructure.
The company has also described a routing layer, CompactifAI Router, that decides whether an AI workload should run locally or be sent to the cloud. Multiverse said its broader platform includes compressed models for devices, cloud deployments and on-premises systems.
How much compression has Multiverse shown?
Multiverse recently said it compressed Meta Platforms’ Llama 3.3 70B so it could run on an Intel Xeon 6 processor. In that example, the company said the model’s disk size fell from about 130 gigabytes to 65 gigabytes, a roughly 50% reduction.
That test still required substantial memory, about 1 terabyte of RAM, according to the company. The point of the demonstration was that the model could run on a central processing unit without a high-end graphics processing unit, rather than on small consumer hardware.
Multiverse said it is also building a software layer for so-called AI factories that combines model compression, GPU orchestration, AI deployment, compute management and governance controls. The company says the goal is to let enterprise customers keep AI and data flows inside their own firewalls without replacing existing infrastructure.
The company lists Allianz, Bank of Canada, Bosch, Iberdrola, Indra, PwC and Telefónica among its customers and partners. It says those relationships span manufacturing, finance, energy, aerospace, cybersecurity, defense and life sciences.
Multiverse said the new funding will support more high-efficiency models, research and development for proprietary compression algorithms, strategic investments in AI gigafactory infrastructure and expansion in East Asia, Southeast Asia, the Middle East, Canada and the United States.
This story draws on original reporting from SiliconANGLE.
