Brands
Latest top stories
Start-ups
Technology

Multiverse: How a Spanish startup is using quantum ideas to make AI cheaper and greener

24 February 2026

 

San Sebastián (Donostia), Spain — As artificial intelligence spreads across industries, a less glamorous but more urgent question is coming to the fore: how to run powerful AI models without unsustainable cost and energy consumption. Multiverse Computing, a Spanish deep-tech company founded in 2019, is addressing that problem by applying quantum-inspired mathematics to compress large machine-learning models, including large language models (LLMs).

In June 2025, Multiverse raised €189 million to scale its flagship software platform, CompactifAI, which the company says can dramatically reduce the size and operating costs of AI models. At the same time, it continues to win research contracts in quantum-adjacent fields, including a $1.4 million project with the German Aerospace Centre (DLR).

Multiverse occupies a niche that is still under-reported but strategically important: it translates ideas from quantum computing into software that can be deployed today, on classical hardware, for commercial AI use.

The hidden bottleneck in AI: cost, energy, and infrastructure

Large language models have become the engine behind generative AI, powering chatbots, coding assistants and knowledge tools across sectors. But their success has exposed a structural weakness: LLMs are expensive to train, expensive to run, and energy-intensive.

For enterprises, this is no longer a technical footnote. Cloud inference bills, GPU shortages, power usage and cooling requirements increasingly shape AI strategy. In some cases, they limit whether advanced AI can be deployed at all.

Multiverse’s argument is simple but consequential: scaling AI will require efficiency, not just larger models.

What “quantum-inspired AI” actually means

Quantum computing is often associated with experimental hardware and long timelines. Multiverse’s technology takes a different path.

The company uses quantum-inspired algorithms, meaning mathematical techniques that originated in quantum physics research but can run on standard computers. Crucially, no quantum hardware is required.

Reuters summarised the approach succinctly: Multiverse combines ideas from quantum physics and machine learning “but doesn’t need a quantum computer”. This distinction matters in a market prone to exaggerated claims and blurred terminology.

CompactifAI: compressing large language models

Multiverse’s commercial product, CompactifAI, is designed to reduce the computational footprint of AI models — particularly LLMs — while preserving performance.

The company reports that CompactifAI can compress models by up to 95%, reducing memory usage and operating costs. These figures are presented as company-reported results and have been widely cited in coverage of its 2025 funding round.

How the technology works (without the jargon)

CompactifAI is built around tensor networks, a mathematical framework used in quantum physics to describe complex systems efficiently. In an AI context, tensor networks allow a model’s internal relationships to be represented more compactly, reducing redundancy.

In a January 2024 research paper published on arXiv, Multiverse researchers describe applying this method to LLaMA-based models. In reported benchmarks, the authors show:

  • 93% reduction in memory footprint

  • 70% reduction in parameters

  • 2–3% accuracy loss, depending on configuration

These results combine tensor networks with other techniques such as quantisation. While arXiv papers are pre-peer-review, they provide technical transparency that goes beyond marketing claims.

Why compression is becoming a strategic AI capability

Model compression has long existed at the margins of machine learning. That has changed.

Today, the economics of AI increasingly depend on:

  • Cost of inference at scale

  • Energy consumption and sustainability targets

  • Ability to deploy models on constrained infrastructure

  • Regulatory and data-sovereignty requirements

Multiverse positions CompactifAI as an enabler of deployment, not just an optimisation tool. According to Reuters, the company claims compression can reduce AI operating costs by up to 80%, and CompactifAI is available through the Amazon Web Services AI Marketplace.

For senior executives, this shifts the conversation from “Can we build this?” to “Can we afford to run this?”

Funding and market validation

In June 2025, Multiverse announced a €189 million investment round, one of the largest AI funding rounds in Spain to date. Reuters reported that investors included Bullhound Capital, HP Inc, Forgepoint Capital and Toshiba.

The framing of the round is notable. Rather than betting on future quantum hardware, investors backed near-term enterprise software addressing a clear pain point in AI adoption.

Multiverse and its investors also state that the company holds more than 160 patents and serves over 100 customers worldwide, including organisations in energy, manufacturing, finance and research. These figures are company-reported but indicate commercial traction beyond pilot projects.

Awards and industry recognition

Multiverse has received several signals of credibility from industry bodies:

  • Future Unicorn Award 2024, awarded by DIGITALEUROPE

  • Inclusion in CB Insights’ AI 100 list (2025)

Such recognition does not replace independent benchmarking, but it reflects how the company is perceived within the European and global AI ecosystem.

Staying close to science: the German Aerospace Centre contract

Multiverse’s identity is not purely commercial. In February 2024, the company announced a $1.4 million research contract under the German Aerospace Centre’s Quantum Computing Initiative (DLR QCI), in partnership with Single Quantum.

The project focuses on improving superconducting materials used in single-photon detectors, critical components for quantum communications and sensing. While separate from CompactifAI, the work reinforces Multiverse’s position as a company operating between fundamental research and applied AI.

A measured critique: what still needs to be proven

Multiverse’s proposition is compelling, but caution is warranted.

First, AI compression performance is context-dependent. Results vary by task, dataset and deployment environment. While Multiverse provides published benchmarks, broad third-party evaluations across production workloads remain limited in the public domain.

Second, the term “quantum-inspired” is often misused in the industry. Multiverse has been careful to avoid claims about quantum advantage or quantum hardware, but ongoing clarity will be essential as the market matures.

The strongest counterweight to scepticism so far is adoption: enterprise customers, a large funding round, and integration into major cloud marketplaces.

Why Multiverse matters now

AI’s next phase will be shaped less by novelty and more by viability. Cost, energy consumption and infrastructure constraints are becoming first-order strategic issues.

Multiverse Computing represents a pragmatic response to that reality. By translating quantum-derived mathematics into deployable AI software, it shows how deep scientific ideas can be commercialised without waiting for future hardware breakthroughs.

This is not a story about speculative quantum futures. It is about making today’s AI workable at scale.

What comes next, according to reporting

In February 2026, Spanish business newspaper CincoDías reported that Multiverse was in talks to raise approximately €500 million, potentially valuing the company at unicorn level. While discussions do not guarantee outcomes, the report underscores growing interest in efficiency-focused AI infrastructure companies.

Whether or not that round materialises, Multiverse’s trajectory illustrates a broader shift: the convergence of quantum thinking and commercial AI deployment.

 

Further reading on this subject on MoveTheNeedle.news:

  • Microsoft’s European quantum bet: what its new Danish lab means for the future of computing in Europe – How Microsoft’s expanded quantum research footprint in Denmark could influence the future of quantum hardware and broader computing innovation in Europe. Read more

  • 2025 in Innovation: The Breakthroughs That Shaped Tomorrow – A year-end technology retrospective that highlights advancements in quantum materials, error-correction, and AI research efforts shaping computing’s next phase. Read more

  • Light Over Silicon: Q.ANT’s Big Bet on Sustainable AI Compute – A look at photonic computing as another alternative path toward energy-efficient AI infrastructure, going beyond ASIC and GPU-centric ecosystems. Read more

 

Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.