Spain-based Multiverse Computing, which makes software to compress LLMs using “quantum-inspired tensor networks”, raised a €25M Series A at a €100M valuation
Context & Ripple Effects
This early financing round established Multiverse Computing as a commercial bet on quantum-inspired techniques for making large language models smaller. Later coverage records a much larger Series B tied to model-compression technology, followed by reports of a potential financing at a markedly higher valuation.
That arc matters because it shifts the company’s story from a specialist technical approach toward a business proposition centered on reducing the compute and energy burden of deploying LLMs.
First-order effects
- Multiverse gains €25M to develop and sell its tensor-network compression software, while the €100M valuation gives its investors and prospective customers an early market reference point.
- The round puts LLM compression—not only larger model training—at the center of the company’s immediate product and go-to-market case.
Second-order effects
- Providers and enterprises evaluating LLM deployments have another route to lower inference-resource requirements, increasing pressure on optimization vendors to demonstrate measurable savings rather than technical novelty alone.
- If compression preserves sufficient model performance, spending can move toward software layers that improve utilization of existing AI infrastructure rather than solely toward additional compute capacity.
Third-order effects
- The subsequent Series C at a $1.7B valuation suggests that efficiency software can become a substantial AI-infrastructure category when customers treat compute and energy use as operating constraints.
- The broader structural question is whether compression becomes a standard capability bundled by model and cloud providers, or remains a differentiated independent layer; that depends on reproducible performance across customer workloads.
The trend: AI infrastructure is expanding beyond model creation toward software that makes deployed models cheaper and less resource-intensive to run.