XPerf Inc. Unveils Innovative Solutions for GPU Utilization Challenges
ROUND ROCK, Texas, Nov. 17, 2025 /PRNewswire-PRWeb/ — XPerf Inc., a forward-thinking AI infrastructure startup, has officially launched from stealth mode to address the urgent issue of poor GPU utilization in complex, multi-vendor AI data centers. Utilizing advanced AI-native optimization technologies, the company claims to improve GPU utilization rates by a minimum of 20%, significantly enhancing operational efficiency.
As the landscape of AI infrastructure shifts from primarily training workloads to a diverse array of pre-training, post-training, inference, and novel applications, managing GPU performance has become increasingly intricate. The presence of multiple generations and vendors of GPUs and ASICs, alongside varied networking systems, compounds the challenges of GPU optimization. Uptime Intelligence highlights that GPUs only operate at about 80% capacity, with optimized models achieving merely 35-45% of their potential compute performance. Notably, Coreweave suggests that the highest Model FLOPS Utilization (MFU) on Nvidia H100 GPUs reaches only 52% during extensive training benchmarks. This inefficiency not only reduces return on investment but also amplifies operational costs, delays innovation timelines, and increases the associated carbon footprint of AI operations.
XPerf Inc., founded by experienced engineers with a rich history of deploying extensive AI clusters, aims to leverage innovative algorithms that dynamically balance hardware capabilities and workloads to resolve these multifaceted challenges. “Our AI-native approach allows XPerf to enhance GPU utilization significantly while adapting to varying workload demands,” stated Dr. Molly Guo, CTO and Co-Founder of XPerf Inc.
Dr. Rakesh Sambaraju, Executive Vice President of Infraeo, an early investor in XPerf, emphasized the critical importance of optimizing utilization as AI data center deployments increasingly focus on inference. “The solutions XPerf is developing will be invaluable for clients deploying large GPU clusters or designing custom ASICs for AI applications,” he remarked.
XPerf will demonstrate its groundbreaking technology in collaboration with partner Infraeo Inc., a provider of advanced copper and optical connectivity solutions for AI clusters, at SC25, booth #5412.
XPerf Inc. is dedicated to enhancing GPU and ASIC utilization in large-scale AI environments. Founded by engineers with substantial experience at Intel Gaudi, the company harnesses an AI-native strategy to bolster performance, energy efficiency, and operational effectiveness across diverse hardware ecosystems. As a participant in the NVIDIA Inception program and the gBETA accelerator, XPerf is at the forefront of tackling GPU underutilization challenges.
For further information and media inquiries, please contact Molly Guo at XPerf Inc. at 1 609-937-9924 or via email at [email protected].
### Key Highlights:
– XPerf Inc. focuses on improving GPU utilization rates by at least 20%.
– The company addresses the complexities of multi-vendor GPU and ASIC deployments.
– Innovative algorithms balance hardware capabilities and workloads effectively.
– Join XPerf at SC25, booth #5412, to learn more about their solutions.
