Multiverse Computing launched the CompactifAI app and API portal on March 3, 2026, in San Sebastián, Spain. The Spanish startup compresses AI models from OpenAI, Meta, DeepSeek, Mistral AI, and NVIDIA using quantum-inspired technology. It targets rising compute costs and supply chain risks amid 9.2% private company default rates. The firm aims to enable efficient, local AI deployment without cloud dependency.

Lux Capital warned firms to secure compute commitments in writing due to financial instability. Multiverse Computing offers an alternative. Its CompactifAI technology shrinks models by up to 95% with only 2-3% precision loss. This beats industry standards of 20-30% accuracy drops. The app features Gilda, a tiny model that runs offline on mobile devices.

The CompactifAI app acts like ChatGPT but processes queries locally. Users enjoy edge AI with full data privacy. No internet or cloud is needed. Developers access compressed models via the new API portal. They can build apps for phones, laptops, and cars. HyperNova 60B, half the size of OpenAI’s gpt-oss-120b, supports tool calling and coding tasks.

Multiverse Computing serves clients like Iberdrola, Bosch, and Bank of Canada. The company raised $215 million in Series B funding. It discusses a new round. Partnerships with Cerebrium and Plain Concepts boost cloud and on-premise deployment. Nano models like ChickenBrain outperform larger rivals on MacBook Pro and Raspberry Pi.

CompactifAI uses tensor networks to restructure model weights. It captures parameter correlations without retraining. Models cut memory use by 93% and run faster. This unlocks sovereign AI in Europe and North America. The startup challenges big-is-better AI norms. It proves small models deliver frontier performance.

Multiverse Computing leads quantum-inspired compression. Its tech reduces energy needs and costs. Enterprises gain scalable AI without infrastructure limits. The launch pushes compressed models mainstream. It promises sustainable AI growth amid chip shortages.