Source: H3C |Original Link
On July 28, the 2026 Trusted Cloud Conference and the inaugural Token Cloud Service Conference, hosted by the China Communications Standards Association and organized by the China Academy of Information and Communications Technology (CAICT), commenced in Beijing. This conference focused on core industrial issues such as computing power for the large-scale deployment of large models, Token production, and trusted cloud services, bringing together stakeholders from industry, academia, research institutions, and enterprises to explore pathways for AI infrastructure upgrades. H3C Group, a subsidiary of Unisplendor Corporation, was invited to attend and deliver a keynote speech, comprehensively showcasing its full-stack AI intelligent cloud technical capabilities and industry implementation achievements. Concurrently, the Distributed Trusted Private Cloud Innovative Application project, jointly submitted by H3C and China Mineral Resources Group Zhoushan Development Co., Ltd., was recognized as an Outstanding Practice in Trusted Cloud User Innovative Applications. Furthermore, the "AI Empowering Digital Infrastructure: Technical Breakthroughs and Construction Guidelines for ICT Agents," co-authored with CAICT, was officially released at the conference.
01 Full-Link System Optimization to Build a Cost-Effective Token Production Foundation
The core agenda of this conference centered on the rapid expansion of the Token cloud service industry. IDC data indicates that daily average Token consumption globally and in China is growing exponentially, with massive inference demands compelling cloud platforms to transition from traditional resource scheduling to AI workload operations.
In his speech, Li Yongping, Director of the Cloud and Digital Intelligence Product Department at H3C Group, noted that the industry currently faces four major pain points: idle computing power, high inference latency, persistently high Token production costs, and a lack of trust in data compliance. He emphasized that relying solely on hardware stacking cannot achieve scalable AI accessibility; only through full-stack collaborative optimization across computing, storage, networking, inference, and gateways can the ultimate cost-effectiveness of Tokens be systematically achieved.
Built upon the CloudOS AI intelligent cloud foundation, H3C has launched a full-stack Token solution that truly delivers efficient computing utilization, economical model usage, practical application deployment, and stable operations. In terms of computing power, the solution achieves unified management of heterogeneous resources across GPUs, NPUs, and general-purpose computing, supporting elastic scaling for clusters of up to 73,728 cards. Its proprietary super-node architecture significantly enhances cluster interconnect bandwidth and single-card inference performance. The storage system incorporates the XCache AI-native caching system, utilizing multi-level KVCache offloading technology to reduce Time To First Token (TTFT) latency by 90% and increase concurrent capacity under equivalent computing power by tenfold. At the network layer, leveraging a topology-aware low-latency collective communication library, communication latency is reduced by 50%, enabling second-level fault detection and minute-level self-healing. The built-in inference engine features video memory optimization, model quantization, and a Prefill-Decode disaggregated distributed architecture. It is compatible with over 20 mainstream training and inference frameworks, including vLLM and TensorRT, and supports more than 100 domestic and international large models. Complemented by an AI intelligent routing gateway and an integrated MaaS platform, the solution enables unified multi-model access, precise Token-level metering, and differentiated cache billing, establishing a complete service closed loop spanning user requests, intelligent scheduling, model inference, and streaming Token output.
Currently, H3C is committed to transforming computing power operations from a resource-oriented to a value-oriented approach. Two benchmark projects have been successfully implemented: a 10,000-card domestic scientific research intelligent computing center and a 100,000-card group-level intelligent computing cloud. These initiatives have significantly improved computing power utilization rates for clients and markedly reduced unit Token production costs, facilitating a transition from hardware leasing to a sustainable TaaS (Token as a Service) business model.
02 Partnering with China Mineral Resources Group Zhoushan Development Co., Ltd. to Create a Benchmark for Domestic Trusted Private Cloud Innovation
The conference also announced the winners of the 2026 Trusted Cloud User Innovative Application Practices. The Distributed Trusted Private Cloud Innovative Application, jointly submitted by H3C and China Mineral Resources Group Zhoushan Development Co., Ltd., was selected as an Outstanding Practice Case, providing a replicable model for central and state-owned enterprises in building domestic unified cloud platforms across multiple data centers.
Based on H3C's distributed cloud architecture, the two parties implemented a multi-data-center deployment achieving unified management across resources, business, and operations. The project adopts a domestic software and hardware ecosystem featuring "one cloud, multiple chips," fully compatible with domestic processors, operating systems, and databases, thereby meeting autonomous and trustworthy assessment requirements. A primary-backup disaster recovery system was established to ensure continuous and stable business operations. Additionally, a tenant-level integrated security resource pool was constructed, encompassing capabilities such as virtual WAF, database auditing, host protection, and unified log auditing, fulfilling the compliance standards of MLPS 2.0 Level 3.
Since its launch, the platform has stably supported business operations at a scale of 300 nodes and 50PB of storage. It has eliminated multi-source data silos, enabling unified O&M, refined management and control, and comprehensive empowerment through data assets. By balancing computing power sharing, data security, and cost control, the project fully demonstrates H3C Intelligent Cloud's mature implementation capabilities in private clouds for central and state-owned enterprises and domestic trusted infrastructure.
Furthermore, the "AI Empowering Digital Infrastructure: Technical Breakthroughs and Construction Guidelines for ICT Agents," co-authored by H3C and CAICT, was officially released at the conference. Focusing on the integrated construction of agents across computing, networking, storage, cloud, security, and O&M, the guideline outlines core technical pathways and implementation specifications. It aims to foster unified technical consensus within the industry and accelerate the standardized, large-scale deployment of AI infrastructure.
Cloud computing has now officially entered a new era of AI intelligent clouds, with Token cloud services becoming the primary vehicle for unleashing AI productivity. As a leader in digital and AI solutions, H3C will continue to leverage its core strengths in "Computing Power × Connectivity." Through continuous full-stack collaborative technical optimization, H3C aims to consistently reduce Token usage costs. Relying on its mature trusted cloud platform, large-scale industry implementation cases, and standardized industrial research outcomes, H3C will provide cost-effective, autonomous, and trusted AI intelligent cloud solutions across all industries. By cultivating new quality productive forces through an efficient intelligent cloud foundation, H3C is dedicated to supporting the high-quality and sustainable development of the domestic AI industry.