The Donostiarra company Multiverse Computing has begun a collaboration with the American technology company Qualcomm with the aim of deploying "efficient" artificial intelligence (AI) models in data centers spread across the planet.
In a statement released this Wednesday, Multiverse Computing details its agreement with Qualcomm Technologies to incorporate "highly efficient" AI models into data center infrastructure. In this way, the artificial intelligence solutions of the company based in San Sebastián will specifically adapt to the Qualcomm Dragonfly AI200 and AI250 platforms.
According to the company, this alliance "combines the AI acceleration hardware of Qualcomm Technologies with the model optimization technology of Multiverse Computing to provide data center operators with a way to run large-scale AI workloads with higher performance and lower energy consumption."
For the managers of these infrastructures, "the direct benefit is greater capacity with the same hardware," they have pointed out. They have also specified that "by optimizing the AI models before they run on the Qualcomm Dragonfly AI200 and AI250 accelerators," Multiverse Computing "reduces the compute and memory footprint required by each model," which "frees up margin in existing deployments, and allows operators to handle more inference requests, run more models simultaneously, or scale their AI services without adding new accelerators or expanding the data center infrastructure."
The company emphasizes that these "efficiency improvements" were demonstrated live during MWC Barcelona 2026, where Multiverse Computing and Qualcomm Technologies showcased a large-scale open-source and compressed language model (LLM) running on the Qualcomm Cloud AI100 Ultra accelerator.
In a usage scenario focused on real-time emergency medical report generation, the compressed version of the model "achieved response times up to 93% faster and performance up to 44% higher compared to the uncompressed base model, in addition to reducing memory usage by up to 45% and energy consumption by up to 21%, without loss of accuracy," they have indicated from Multiverse Computing.
In another demonstration, a retrieval-augmented generation (RAG) chatbot deployed 'on-premise' to consult confidential financial documentation "worked up to 35% faster and provided performance up to 54% higher, in addition to reducing memory usage by up to 45% and energy consumption by up to 14%, again with no loss of accuracy".
The company concludes that applications running on the new Qualcomm Dragonfly AI200 and AI250 accelerators will deliver benchmark results "even better".