Multiverse Computing Brings Compressed AI Models to the Masses
▼ Summary
– Due to high private company defaults and AI supply chain instability, Lux Capital advises firms to secure compute capacity agreements in writing.
– Multiverse Computing offers an alternative with its CompactifAI technology, which compresses large AI models to run locally on devices, reducing reliance on external cloud infrastructure.
– Its CompactifAI app features a small model called Gilda that can operate offline for privacy, but it automatically switches to cloud-based models if a device lacks sufficient RAM or storage.
– The company’s primary focus is on businesses, launching a self-serve API portal that gives developers direct access to its compressed models for production use, emphasizing cost savings and control.
– Compressed models like Multiverse’s HyperNova 60B are narrowing the performance gap with larger LLMs, enabling new business applications in areas with poor connectivity, such as drones and satellites.
In a climate of financial uncertainty, where securing reliable computing power for artificial intelligence is becoming a major concern, a shift toward more efficient, self-contained models is gaining momentum. Rather than depending on external cloud infrastructure with its associated costs and risks, businesses are increasingly exploring the potential of compressed AI models that operate directly on local devices. This approach not only reduces dependency on third-party providers but also enhances data privacy and operational resilience. One company positioning itself at the forefront of this movement is Multiverse Computing, a Spanish startup that is making its compressed AI technology more accessible to developers and enterprises.
While Multiverse has maintained a relatively quiet presence compared to some industry giants, its focus on model compression is now drawing significant attention. The company has successfully compressed foundational models from leading AI labs like OpenAI, Meta, and Mistral AI. To demonstrate the practical applications of this technology, Multiverse has launched two key products: a consumer-facing chat application called CompactifAI and a new self-serve API portal for developers.
The CompactifAI app functions similarly to other AI chat tools, allowing users to ask questions and receive answers. Its defining feature is the inclusion of a model named Gilda, which is compact enough to run locally on a device without an internet connection. This offers users a genuine taste of edge AI, where personal data never leaves their smartphone or computer. However, this capability comes with hardware requirements; the device must have sufficient RAM and storage. If these are not met, the app automatically switches to using cloud-based models through an API, a process managed by an internal system named Ash Nazg. This fallback to the cloud, while functional, means the primary privacy benefit is temporarily lost.
Given these technical limitations, the app’s current user base remains modest, indicating its role may be more demonstrative than aimed at mass consumer adoption. The company’s strategic focus appears firmly on the business sector. The newly launched API portal allows developers and companies to directly integrate Multiverse’s compressed models into their own production environments, bypassing traditional marketplaces and offering greater transparency and control.
A major selling point for these smaller, efficient models is the substantial reduction in computing costs, a critical factor for enterprises considering alternatives to expensive large language models. Real-time usage monitoring via the API provides businesses with the data needed to manage these efficiencies. The performance gap between compact and large models is also closing. Multiverse’s latest offering, HyperNova 60B 2602, is built on a public OpenAI model and reportedly delivers faster responses at a lower cost, which is particularly advantageous for complex, automated coding tasks.
Creating models that are both powerful and small enough for mobile devices is an immense technical hurdle. Other companies, like Apple with its Intelligence system, use a hybrid approach that combines on-device and cloud processing. Multiverse’s app can also route to larger cloud models when needed, but its core mission is to prove the standalone value of local AI. For professionals in sensitive or remote fields, such as finance, defense, or field operations, a model that operates entirely offline provides unmatched privacy and reliability. This technology unlocks business use cases in environments like drones, satellites, or industrial equipment where constant connectivity is unreliable or impossible.
With a client roster that includes major organizations like the Bank of Canada and Bosch, Multiverse is building a solid enterprise foundation. Its growth trajectory is underscored by significant financial activity; following a $215 million Series B round last year, the company is reportedly seeking an additional €500 million in funding at a valuation exceeding €1.5 billion, signaling strong investor confidence in the future of efficient, decentralized artificial intelligence.
(Source: TechCrunch)
