At the GPUStack AI Infra Ecosystem and Technology Innovation Conference held on the evening of September 9, GPUStack CEO Qin Xiaokang highlighted that while domestic GPUs and model ecosystems are developing rapidly, challenges persist around compatibility adaptation, performance tuning, and version iteration. He emphasized that GPUStack aims to collaborate with partners, community members, and developers to make domestic computing power more accessible, unlocking the full potential and value of every GPU through continuous optimization.

Since its open-source launch in 2024, GPUStack has served over 110,000 users across 121 countries and regions worldwide, establishing a broad global user base and accumulating rich real-world scenario validation for domestic GPU adaptation, model deployment, and product evolution. Qin Xiaokang pointed out that AI infrastructure encompasses multiple layers—chips, servers, storage, models, and applications—making it impossible for a single company to cover all capabilities. GPUStack will focus on its core strengths while working alongside industry partners to refine products and solutions, simplifying complex computing adaptation and platform construction.

Liang Sheng, co-founder and CTO of GPUStack, contextualized the global GPU industry’s trajectory, noting that market competition extends far beyond hardware performance. A mature GPU ecosystem relies on a comprehensive software framework spanning underlying compute, networking, and storage, extending to model inference and service delivery. As domestic GPUs scale up in production capacity, quality, and performance, the synchronization of the software stack will directly impact their large-scale adoption and global market expansion. GPUStack aims to codify complex deployment and tuning processes into standardized capabilities through its open software platform.

During the conference, GPUStack signed strategic cooperation agreements with four entities—Diou Shuihua Group, Asiainfo Technologies, Rongxin Zhiyuan, and XSKY—to deepen collaboration across the AI infrastructure supply chain. In computing supply and operations, GPUStack and Diou Shuihua Group will work together on computing infrastructure construction, rental services, and model API offerings, enhancing the "Shuihua Do.top" online computing platform to drive standardization of computing resources.

For enterprise-grade AI platform development, GPUStack and Asiainfo Technologies will integrate capabilities in heterogeneous GPU management, model deployment, service operations, and metering/billing, strengthening operational efficiency and large-scale delivery for enterprise AI infrastructure. Extending into intelligent computing hardware, GPUStack will embed its AIOS into Rongxin Zhiyuan's AGC intelligent computing devices, bridging hardware, scheduling, and model services to provide streamlined, integrated deployment solutions. In inference acceleration, GPUStack and XSKY will collaborate on shared KV Cache technology, integrating XSKY MeshFusion to boost inference performance in long-context, multi-turn dialogue, and agent-based scenarios.

These four strategic collaborations connect critical capabilities in computing supply, platform operations, intelligent hardware, and data storage, showcasing GPUStack's ecosystem aggregation strength and signaling a shift from product-level adaptation toward joint solutions and deeper co-construction with industry partners.