Multiverse Computing: AI Gets Smaller, Faster & More Accessible

With private company defaults soaring to over 9.2% – the highest rate in years – VC firm GearTech recently advised companies relying on Artificial Intelligence to get their compute capacity commitments confirmed in writing. Amidst financial instability rippling through the AI supply chain, a handshake agreement simply isn’t enough. However, there’s a compelling alternative: reducing reliance on external compute infrastructure altogether. Smaller AI models, capable of running directly on a user’s device – eliminating the need for data centers, cloud providers, and counterparty risk – are rapidly becoming viable, and Multiverse Computing is leading the charge.

The Rise of On-Device AI and Multiverse Computing

The Spanish startup, Multiverse Computing, has been quietly building a reputation for innovation in AI compression. While maintaining a lower profile than some of its competitors, the growing demand for AI efficiency is propelling them into the spotlight. Multiverse has successfully compressed models from leading AI labs including OpenAI, Meta, DeepSeek, and Mistral AI, and has now launched both the CompactifAI app – a demonstration of their compressed models’ capabilities – and a self-serve API portal, expanding access for developers.

Introducing CompactifAI: AI on the Edge

The CompactifAI app functions as an AI chat tool, similar to ChatGPT or Mistral’s Le Chat. Users can pose questions and receive responses powered by Gilda, a remarkably small model designed to run locally and offline, according to Multiverse. This represents a significant step towards AI on the edge, offering users data privacy and functionality without requiring a constant internet connection.

However, there are practical limitations. The app requires sufficient RAM and storage on the user’s device. Older iPhones, for example, may lack the necessary resources and will revert to cloud-based models via the API. Multiverse’s “Ash Nazg” system – a nod to Tolkien’s “The Lord of the Rings” – intelligently manages this routing between local and cloud processing. Crucially, switching to the cloud compromises the app’s primary privacy advantage.

Current Adoption and Limitations

While CompactifAI showcases the potential of on-device AI, mass adoption isn’t yet a reality. Data from GearTech Sensor Tower indicates fewer than 5,000 downloads in the past month. This suggests the app’s primary focus isn’t direct consumer engagement, but rather, demonstrating the technology to potential business clients.

Empowering Businesses with the CompactifAI API Portal

Today, Multiverse is launching its self-serve API portal, providing developers and enterprises with direct access to its compressed models – bypassing the need for platforms like AWS Marketplace. This is a pivotal move, positioning Multiverse as a key enabler for businesses seeking to integrate efficient AI solutions.

“The CompactifAI API portal now gives developers direct access to compressed models with the transparency and control needed to run them in production,” stated CEO Enrique Lizaso. Real-time usage monitoring is a core feature of the API, reflecting the significant cost savings that smaller models offer compared to large language models (LLMs).

The Evolution of Small Language Models

Small models are rapidly closing the performance gap with their larger counterparts. Mistral AI recently released Mistral Small 4, optimized for a wide range of tasks including general chat, coding, agentic workflows, and reasoning. They also introduced Forge, a system allowing enterprises to build custom models, including small models tailored to specific use cases and tolerance levels.

Multiverse’s recent advancements further demonstrate this trend. Their latest compressed model, HyperNova 60B 2602, is based on gpt-oss-120b, an open-source OpenAI model. Multiverse claims it delivers faster responses at a lower cost than the original, particularly beneficial for agentic coding workflows where AI autonomously handles complex programming tasks.

Challenges and Advantages of On-Device AI

Compressing models to run effectively on mobile devices while maintaining utility is a complex engineering challenge. Apple Intelligence addressed this by combining on-device and cloud processing. Multiverse’s CompactifAI app can also route requests to gpt-oss-120b via API, but its core objective is to highlight the advantages of local models like Gilda – advantages that extend beyond mere cost savings.

Privacy, Resilience, and New Use Cases

For professionals in sensitive fields, a locally-run model offers enhanced privacy and resilience, independent of cloud connectivity. However, the most significant value lies in the new business applications this technology unlocks. Consider embedding AI in drones, satellites, and other environments where reliable connectivity is not guaranteed. This opens doors to a new era of distributed AI.

Multiverse Computing: Growth and Funding

Multiverse already serves over 100 global customers, including the Bank of Canada, Bosch, and Iberdrola. Expanding its customer base is crucial for securing further funding. Following a $215 million Series B round last year, the company is reportedly seeking a €500 million funding round at a valuation exceeding €1.5 billion. This demonstrates strong investor confidence in the future of on-device AI and Multiverse’s position within this rapidly evolving market.

The Future of AI: Accessibility and Efficiency

The trend towards smaller, faster, and more accessible AI models is undeniable. As computational costs continue to rise and concerns about data privacy intensify, the ability to run AI locally will become increasingly important. Multiverse Computing is at the forefront of this revolution, offering a compelling alternative to the traditional cloud-centric approach. The company’s innovative compression technology, coupled with its developer-friendly API portal, positions it as a key player in shaping the future of AI – a future where intelligence is not confined to the cloud, but is readily available at the edge, empowering individuals and businesses alike.

Key Takeaways:

• Reduced Costs: Smaller models significantly lower compute expenses compared to LLMs.
• Enhanced Privacy: On-device processing keeps data secure and private.
• Increased Resilience: Functionality isn't dependent on a constant internet connection.
• New Applications: Enables AI integration in previously inaccessible environments (drones, satellites, etc.).
• Growing Market: Demand for efficient and accessible AI is rapidly increasing.