Multiverse Computing has announced a collaboration with Qualcomm Technologies to bring more efficient artificial intelligence models to data centers worldwide. Under the collaboration, Multiverse Computing will optimize its AI models for Qualcomm Dragonfly AI200 and AI250 accelerators. The companies aim to help data center operators improve AI workload performance while reducing memory requirements, power consumption and infrastructure costs.
The initiative combines Qualcomm Technologies’ AI acceleration hardware with Multiverse Computing’s model compression and optimization technology.
By optimizing models before they run on the Dragonfly accelerators, Multiverse Computing expects to reduce the amount of compute capacity and memory required for each workload.
The companies said this could allow data center operators to process more inference requests, run additional AI models at the same time and scale AI services without immediately purchasing more accelerators or expanding their physical infrastructure.
The collaboration reflects growing demand for AI systems that can deliver stronger performance without a proportional increase in energy use, hardware investment and operating expenses.
Data center operators are facing increased pressure to accommodate large-scale AI workloads while managing constraints involving electricity, cooling, memory capacity and accelerator availability.
Multiverse Computing’s technology is designed to reduce the size and computational requirements of AI models while maintaining their performance and accuracy.
The companies previously demonstrated the potential benefits of this approach at Mobile World Congress in March 2026.
During the event, Multiverse Computing and Qualcomm Technologies showcased a compressed open-source large language model running on a Qualcomm Cloud AI100 Ultra accelerator.
In a real-time emergency medical reporting demonstration, the compressed model delivered response times that were up to 93% faster than those of the uncompressed base model.
The demonstration also produced up to 44% higher throughput while reducing memory usage by as much as 45% and power consumption by up to 21%.
Multiverse Computing said the performance improvements were achieved without a loss in model accuracy.
The companies also demonstrated an on-premises retrieval-augmented generation chatbot designed to search and analyze confidential financial documents.
That system ran up to 35% faster than the uncompressed model and delivered up to 54% greater throughput.
The financial-document chatbot reduced memory usage by as much as 45% and lowered power consumption by up to 14%, again without a loss in accuracy, according to the companies.
Multiverse Computing and Qualcomm Technologies believe applications running on the newer Dragonfly AI200 and AI250 accelerators could produce even stronger benchmark results.
The companies did not disclose specific performance targets for the Dragonfly accelerators or provide a timeline for the availability of jointly optimized models.
The collaboration is intended to support enterprise and cloud customers deploying AI systems in environments where power efficiency, infrastructure utilization and data control are increasingly important.
For data center operators, reducing the compute and memory footprint of an AI model could increase the amount of work performed by existing hardware.
This may help customers extend the useful capacity of current data center deployments while delaying or reducing the need for additional accelerators, servers and supporting infrastructure.
The approach may also be useful for enterprises running AI models within their own data centers rather than relying entirely on external cloud platforms.
Multiverse Computing focuses on sovereign and efficient AI, developing specialized models that organizations can operate within their own infrastructure.
The company’s technology is designed for customers that require greater control over sensitive information, governance policies and regulatory compliance.
Multiverse Computing serves more than 100 customers globally, including Iberdrola, Bosch and the Bank of Canada. The company is headquartered in Donostia-San Sebastián, Spain, and has offices in the United States, Canada and Europe.
The collaboration with Qualcomm Technologies extends Multiverse Computing’s efforts to pair model-level optimization with specialized AI hardware.
The companies believe the combination can provide a practical path for data center operators and enterprises seeking to expand AI capacity without equivalent increases in energy consumption and hardware spending.
KEY QUOTES:
“Qualcomm Technologies has built its leadership in segments such as mobile and IoT, where maximizing performance within strict power and efficiency constraints has always been essential. Those same principles are now becoming critical in AI data centers.”
“By combining Qualcomm Technologies’ highly efficient and performant AI accelerators with Multiverse Computing’s model compression and optimization technology, we can help customers achieve breakthrough improvements in performance, cost and energy efficiency at scale.”
Victor Gaspar, Chief Sales Officer at Multiverse Computing
“As AI adoption accelerates across enterprise and cloud infrastructure, customers need solutions that deliver high performance with greater efficiency.”
“Qualcomm Technologies brings industry-leading AI acceleration, while Multiverse Computing works closely with customers, their AI models and their specific use cases to optimize real-world deployments. Together, we can deliver highly efficient AI solutions tailored to the needs of modern data centers.”
Dino Flore, Vice President of Technology at Qualcomm Europe
