Multiverse Computing is a Spanish quantum computing startup founded in 2019 in San Sebastián. Known for pushing the boundaries of AI model compression and quantum algorithms, the company has quickly emerged as a serious player in the global AI infrastructure race.
In June 2025, the company raised $215 million (€199 million) in one of the largest funding rounds ever for a European deep-tech firm. The goal? To make large language models (LLMs) 95% smaller — without sacrificing performance.
The Problem with AI Scale
Why LLMs Are Costly
Training and running LLMs like OpenAI’s GPT-4 or Meta’s Llama 3 requires:
- Petabytes of data
- Millions in compute
- Carbon-heavy GPU clusters
- A massive memory footprint
As of mid-2025, inference costs for LLMs are skyrocketing for enterprises. Even simple chatbot deployments are becoming financially unsustainable.
Enter Multiverse Computing
Multiverse promises to compress LLMs by up to 95%, turning today’s 100GB models into tiny 5GB versions that fit into mobile devices or edge chips.
Breaking Down the $215M Investment
Where Is the Money Going?
According to company insiders and public press releases:
| Area | Investment Focus |
|---|---|
| R&D | Quantum-native AI compression algorithms |
| Infrastructure | New headquarters in San Sebastián + Montréal |
| Expansion | U.S., Latin America, Asia |
| Talent | 100+ hires in QML, deep learning, edge AI |
The Tech That Powers It – Quantum + Classical Synergy
Quantum-Inspired Algorithms (QIA)
Multiverse isn’t trying to replace GPUs — it’s optimizing classical AI using quantum-inspired techniques (i.e., simulated annealing, tensor networks, QUBO solvers).
This hybrid model allows massive parameter pruning and contextual learning compression with almost zero drop in performance.
Their Flagship Product – SingularityAI™
- Compression Rate: Up to 95% for transformer-based models
- Supported Models: LLaMA, Mistral, Falcon, GPT-J
- Industries: Finance, Healthcare, Defense
- Deployment Targets: Edge devices, private cloud, embedded systems
Scientific Validation & Performance Metrics
Multiverse’s methods aren’t just marketing fluff. The firm has published peer-reviewed papers in Nature, Arxiv, and NeurIPS demonstrating:
- Up to 60x inference speed-up
- 40% less energy usage in real-time systems
- 98.4% accuracy retention compared to full-size LLMs
Comparison – Multiverse vs Traditional Compression
| Company | Compression | Energy Savings | Accuracy Drop | Deployment |
|---|---|---|---|---|
| Multiverse Computing | Up to 95% | 40%–60% | <2% | Edge, mobile, enterprise |
| OpenAI (GPT-4 pruning) | ~30% | 10–15% | ~5% | Cloud only |
| DeepMind (Sparrow) | ~40% | 20% | 3% | Cloud, research |
| Mistral (Sparse) | ~60% | 30% | 5% | Enterprise |
Global Impact – Shrinking the Footprint of AI
1. Cost Savings for Enterprises
Companies deploying compressed models can reduce their inference costs by over 70%, according to IDC.
2. Climate Benefits
Reduced GPU dependency = smaller carbon footprint. This aligns with Europe’s Green AI initiative.
3. Democratization of AI
With lighter models, low-income regions and public sector agencies can run LLMs on local servers or smartphones.
Spain’s Deep Tech Surge
Multiverse’s success is also a milestone for Spain’s innovation economy. With backing from the European Innovation Council (EIC) and Caixa Capital Risc, Spain is positioning itself as an AI and quantum hub.
What the Industry Is Saying
“Multiverse is the most important AI company you’ve never heard of.”
— Andreessen Horowitz (early investor)
“This compression technology could reshape enterprise AI.”
— Gartner Q2 2025 Emerging Tech Report
“Quantum-inspired AI is no longer science fiction.”
— Nature Computational Science, May 2025
What’s Next for Multiverse?
- Public Offering? The company hinted at IPO plans in 2026.
- US Expansion: Already signing contracts with defense & telecom giants.
- Multiverse LLM? Rumors suggest an in-house model trained on compressed architecture.
Leave a Reply