As financial instability ripples through the artificial intelligence supply chain, Spanish startup Multiverse Computing is pushing its compressed AI models into the mainstream with a dual-pronged strategy targeting both consumers and enterprises. The company’s technology enables smaller AI models to run directly on user devices, potentially reducing reliance on expensive and sometimes unstable cloud infrastructure. This development arrives amid warnings from venture capital firm Lux Capital about securing compute capacity commitments in writing, highlighting broader concerns about AI infrastructure dependencies.
Multiverse Computing’s Compression Technology
Multiverse Computing has developed quantum-inspired compression technology called CompactifAI that significantly reduces the size of AI models while maintaining functionality. The company has successfully compressed models from major AI laboratories including OpenAI, Meta, DeepSeek, and Mistral AI. This compression enables models to operate locally on devices without continuous cloud connectivity. The technical approach represents a significant advancement in model optimization, addressing both computational efficiency and privacy concerns simultaneously.
The company recently launched two key products:
- CompactifAI App: A consumer-facing AI chat application similar to ChatGPT or Mistral’s Le Chat
- API Portal: A self-service platform giving developers direct access to compressed models
The Financial Context Driving AI Efficiency
The push toward compressed AI models coincides with increasing financial pressures across the technology sector. Private company defaults reached 9.2% recently, the highest rate in years, according to market data. This financial instability has prompted venture capital firm Lux Capital to advise AI-dependent companies to secure their compute capacity commitments through formal written agreements rather than informal arrangements.
This financial backdrop creates a compelling case for alternative approaches to AI deployment. Traditional large language models require substantial cloud computing resources, creating significant operational expenses and potential single points of failure. Smaller, compressed models that can run locally offer potential cost savings and reduced counterparty risk, though they come with different technical limitations.
Enterprise Adoption and Use Cases
Multiverse Computing already serves more than 100 global customers including the Bank of Canada, Bosch, and Iberdrola. The company’s technology proves particularly valuable in scenarios where connectivity cannot be guaranteed or where data privacy concerns prevent cloud processing. Specific applications include:
| Industry | Application | Benefit |
|---|---|---|
| Finance | Secure document analysis | Data never leaves device |
| Manufacturing | Quality control in remote facilities | Works without internet |
| Energy | Equipment monitoring | Reduced latency |
| Defense/Security | Field operations | Enhanced privacy |
Technical Implementation and Limitations
The CompactifAI app embeds a model called Gilda that can run locally and offline on compatible devices. However, the technology faces hardware limitations. Mobile devices must have sufficient RAM and storage to support local processing. Many older smartphones, particularly previous-generation iPhones, lack the necessary specifications. When devices cannot support local processing, the app automatically routes requests to cloud-based models via API through a system Multiverse calls Ash Nazg.
This automatic routing creates a privacy trade-off. When processing moves to the cloud, the application loses its primary privacy advantage. Consequently, the consumer app has seen limited adoption, with fewer than 5,000 downloads in the past month according to Sensor Tower data. The company appears to prioritize enterprise applications where the technical requirements can be more carefully managed and controlled.
Performance Comparisons and Advancements
Multiverse’s latest compressed model, HyperNova 60B 2602, demonstrates the narrowing performance gap between compressed and full-sized models. Built on gpt-oss-120b, an OpenAI model with publicly available underlying code, HyperNova reportedly delivers faster responses at lower cost than its source model. This advantage proves particularly significant for agentic coding workflows where AI systems autonomously complete complex, multi-step programming tasks.
Meanwhile, other companies are advancing small model technology. Mistral recently updated its small model family with Mistral Small 4, optimized for general chat, coding, agentic tasks, and reasoning. The French company also released Forge, a system enabling enterprises to build custom models with specific trade-offs tailored to their use cases.
Market Position and Funding Landscape
Multiverse Computing maintains a lower profile than some AI industry peers but is gaining visibility as demand for AI efficiency grows. After raising a $215 million Series B round last year, the company is reportedly seeking approximately €500 million in fresh funding at a valuation exceeding €1.5 billion. This funding would support further research and development while expanding the customer base beyond the current 100+ global clients.
The company’s API portal represents a strategic move toward broader adoption. CEO Enrique Lizaso emphasized that the portal “gives developers direct access to compressed models with the transparency and control needed to run them in production.” Real-time usage monitoring serves as a key feature, addressing enterprise requirements for operational visibility and cost management.
Industry Implications and Future Directions
The emergence of viable compressed AI models signals a potential shift in how organizations deploy artificial intelligence. While large language models will continue dominating applications requiring extensive knowledge bases and complex reasoning, compressed models offer compelling alternatives for specific use cases. The technology proves particularly relevant for:
- Edge computing scenarios where low latency is critical
- Privacy-sensitive applications in healthcare, finance, and government
- Remote or disconnected environments including drones, satellites, and field operations
- Cost-sensitive implementations where cloud expenses prove prohibitive
Apple’s approach with Apple Intelligence, combining on-device and cloud models, represents an alternative strategy for balancing capability with privacy. However, Multiverse’s technology demonstrates that purely local processing can deliver sufficient performance for many business applications without hybrid architectures.
Conclusion
Multiverse Computing’s push into mainstream compressed AI models arrives at a critical juncture for the artificial intelligence industry. As financial pressures mount and concerns about cloud dependency grow, the Spanish startup’s technology offers a viable alternative for specific applications. While consumer adoption faces hardware limitations, enterprise applications in privacy-sensitive, remote, or cost-conscious environments show significant promise. The company’s expanding customer base and rumored funding round suggest growing market recognition that efficient, local AI processing represents more than a niche solution. As compression technology continues advancing and hardware capabilities improve, compressed AI models may increasingly complement rather than merely compete with their larger cloud-based counterparts.
FAQs
Q1: What is Multiverse Computing’s main technology?
Multiverse Computing specializes in quantum-inspired compression technology called CompactifAI that reduces AI model sizes significantly, enabling them to run locally on devices without continuous cloud connectivity.
Q2: How do compressed AI models differ from traditional large language models?
Compressed models are significantly smaller and can run locally on devices, offering potential privacy advantages and reduced cloud dependency. However, they typically have more limited capabilities compared to full-sized models running in data centers.
Q3: What are the main limitations of Multiverse’s CompactifAI app?
The app requires devices with sufficient RAM and storage for local processing. Many older smartphones cannot support this, forcing the app to switch to cloud processing, which eliminates the privacy advantage. Current download numbers remain relatively low.
Q4: Which companies are using Multiverse’s technology?
The company serves more than 100 global customers including the Bank of Canada, Bosch, and Iberdrola. These organizations typically deploy the technology in privacy-sensitive or remote operational environments.
Q5: How does Multiverse’s funding and valuation compare to other AI companies?
After a $215 million Series B round last year, Multiverse is reportedly seeking approximately €500 million at a valuation exceeding €1.5 billion. This positions the company as a significant but not dominant player in the broader AI infrastructure market.
Updated insights and analysis added for better clarity.
This article was produced with AI assistance and reviewed by our editorial team for accuracy and quality.
