Qualcomm partners with Multiverse Computing to shrink AI models without losing brainpower

Qualcomm Technologies and Spain-based Multiverse Computing have teamed up to deploy compressed AI models on Qualcomm’s data center accelerators, a partnership that could reshape how efficiently large language models run at scale. The collaboration, announced on August 5, 2026, pairs Multiverse’s CompactifAI compression platform with Qualcomm’s Dragonfly AI200 and AI250 chips. What the compression actually does Multiverse’s CompactifAI platform uses what the company calls quantum-inspired tensor network compression. It borrows mathematical techniques from quantum computing to strip away redundant parameters in AI models, shrinking them significantly while preserving their accuracy. The early results are genuinely striking. Compressed large language models running on Qualcomm hardware achieved response times up to 93% faster than their uncompressed counterparts. Performance improved by 44%, memory usage dropped by 45%, and energy consumption fell by 21%, all with no accuracy loss. Prior deployments of Multiverse’s technology have delivered over 60% parameter reduction and 84% greater energy efficiency. The two companies showcased these capabilities at MWC Barcelona 2026, where compre...