HONG KONG, Sept. 03, 2026 (GLOBE NEWSWIRE) — On September 3, Xunce Technology (3317.HK) officially launched the TokenCloud all-in-one AI model training and deployment computing power platform. As the hardware resources and infrastructure that facilitate the conversion of data resources into tokens, TokenCloud builds a four-tier, synergistic ecosystem based on heterogeneous computing devices, powered by mainstream algorithmic models, fueled by multi-source internal and external data, and serving vertical industry clients. It is dedicated to establishing a full-chain value pathway from data to tokens, enabling every piece of data to securely and efficiently unlock its intelligent value, and helping industries across the board transform data resources into genuine AI productivity.
The launch of TokenCloud marks Xunce Technology’s formal establishment of a full-chain product closed-loop encompassing “computing power—data—tokens—models—applications.” This achievement signifies a leap from a “digital foundation” to an “AI productivity platform,” securing a strategic vantage point in the token economy era while providing the industry with a model for AI infrastructure that bridges the entire “data-to-token” value chain.
Building a Scenario-Driven, End-to-End Closed Loop to Provide Out-of-the-Box AI Infrastructure
As enterprise AI applications shift from “proof of technical feasibility” to “competition in engineering efficiency,” systemic challenges—such as computing power shortages, difficulties in model selection, hardware compatibility issues, challenges in optimized deployment, and significant security concerns—are becoming the biggest bottlenecks hindering AI implementation.
TokenCloud was created precisely to address these pain points. Positioned as the core “hardware and energy foundation” at the upstream of the token industry chain, it adheres to the principle of “ensuring every unit of computing power precisely serves every enterprise AI scenario.” It has built a complete closed-loop process spanning “scenario identification → model selection → hardware matching → training optimization → secure delivery.” Based on business scenarios and tasks, it intelligently recommends the optimal model and matches it with the most suitable hardware. It then uses distillation optimization to enable lightweight, “specialized” models to achieve accelerated inference with high accuracy, all the way through to secure delivery. With each step seamlessly linked, it provides enterprise customers with out-of-the-box AI infrastructure.
In terms of product design, TokenCloud adopts a “full-chain visualization” philosophy, integrating the previously disparate processes of model selection, computing power configuration, security management, and cost assessment in AI deployment into a unified decision-making platform. Comprising five core modules—solution selection, model distillation, data security management, computing power acceleration, and computing resource management—this platform shifts enterprise AI deployment from “rule-of-thumb” approaches to “data-driven decision-making.” It ensures that the return on investment for every computing power allocation is clearly quantifiable, achieving a dynamic balance among performance, cost, and security.
Four Core Capabilities Build Competitive Barriers and Enable Deep Synergy Between Models and Computing Power
The actual value of a token is determined by four key metrics: first-character latency, generation speed, concurrent processing capacity, and service stability. Through Hardware-Software Co-Optimization, TokenCloud enables the large-scale supply of high-quality tokens—ensuring that every piece of data unleashes its maximum token value through the optimal combination of models and hardware.
Specifically, TokenCloud has established industry-leading technological barriers through four core capabilities: a heterogeneous hardware platform, Cloud-Edge Collaboration Architecture, model evaluation and optimization, and secure isolation:
Heterogeneous Hardware Platform—Integrates diverse computing resources such as GPUs, NPUs, FPGAs, and CPUs to enable unified management and intelligent scheduling across architectures. It supports full-scenario coverage ranging from hundreds to tens of thousands of cards, abstracting underlying, disparate hardware resources into unified computing units.
Cloud-Edge Synergy Architecture—Establishes a hybrid deployment model combining “cloud computing power and on-premises computing power,” enabling elastic scaling to handle large-scale training and high-concurrency inference while ensuring that enterprises’ raw data remains within secure domains. This allows enterprises to enjoy cloud computing power while maintaining full data sovereignty at all times.
Model Evaluation and Optimization System—Establishes a comprehensive evaluation framework covering foundational capabilities, scenario adaptation, engineering optimization, and ecosystem value. It supports task-specific distillation optimization, transferring the core capabilities of large models to lightweight scenario-specific models, thereby significantly reducing inference latency and hardware costs while maintaining high accuracy.
Of particular importance is computing power acceleration—while chip computing power continues to rise, actual effective throughput remains far below theoretical peak values. Through end-to-end acceleration capabilities, TokenCloud significantly reduces first-token latency and enhances concurrent processing capacity, transforming acceleration cards from “waiting for computation” to “continuous computation.” This enables enterprises to support larger business volumes without additional hardware, achieving the optimal balance between accuracy, speed, cost, and deployment efficiency, and enabling the large-scale supply of high-quality tokens.
Security Isolation Mechanism—We build a multi-layered security protection system where data visibility is tiered and data flow is fully traceable throughout its entire journey. Highly sensitive data is 100% locked within the enterprise’s local security domain, with a three-pronged approach combining multi-tenant isolation, encrypted transmission, and access control.
Unlike simple computing power rental platforms or model service platforms on the market, TokenCloud achieves deep synergy between models and computing power—the platform centrally determines the degree of model distillation and the hardware for deployment, ensuring the optimal balance of accuracy, speed, cost, and efficiency. It provides enterprise customers with a comprehensive, one-stop service solution—from model selection and hardware matching to distillation optimization—covering the entire “computing power—data—Token—model—application” chain to accelerate the implementation of enterprise AI.
Software and Hardware Synergy Closes the Full-Path Loop; Full-Stack AI Products Accelerate Industry Adoption
The launch of TokenCloud has enabled Xunce Technology’s full-path product ecosystem to form a complete commercial loop. Previously, the company had introduced the world’s first TokenOS operating system; the two systems have clearly defined roles and operate in synergy.
TokenOS focuses on data capabilities, transforming enterprises’ multi-source, heterogeneous data into standardized scenario-based Tokens in real time, thereby addressing the issue of “data usability.” TokenCloud focuses on model training and deployment, as well as “data + model” integration capabilities. It integrates computing power scheduling, model inference optimization, and the refinement, tuning, and orchestration of enterprise-level small models, bridging the entire chain from data ingestion to application deployment and providing hardware service support for Token generation.
Through this synergy between software and hardware, combined with the underlying AIDP data resource processing service system and the top-level TokenRouters model outcomes service and scenario integration and exchange center, the two solutions together form a complete “computing power—data—Token—model—application” closed-loop ecosystem. This enables enterprises to significantly reduce deployment costs and improve matching efficiency through the synergy of computing power, data, and models. Xunce Technology has also undergone a comprehensive upgrade from an AI real-time data infrastructure service provider to an integrated AI-to-B infrastructure service provider.
Furthermore, Xunce Technology has partnered with multiple domestic GPU manufacturers to focus on the integration of “computing power and data.” As the localization of the full AI stack enters the stage of systematic engineering optimization, TokenCloud has become a key hub connecting domestic heterogeneous computing power with enterprise AI applications, offering significant strategic value within the ecosystem.
Currently, the company has expanded into eleven high-value industries, including finance, telecommunications, power, energy, high-end manufacturing, biomedicine, robotics training, commercial space, urban operations, the low-altitude economy, and smart vehicles. In the first half of 2026, the Token business model was successfully implemented in high-value scenarios such as biomedicine, energy dispatch, and industrial quality inspection, laying the groundwork for cross-industry replication.
The ultimate goal of computing power is not to simply stack chips, but to ensure that every piece of data can be securely and efficiently converted into token value. As AI transitions from the “model competition” phase to the “deployment competition” phase, enterprises need not isolated computing power or isolated models, but rather an integrated infrastructure capable of combining computing power, data, and models to truly adapt to enterprise business scenarios. Positioned as an integrated AI infrastructure, TokenCloud enables computing power to serve models, models to serve scenarios, and scenarios to generate value, thereby laying a solid computing power foundation for the AI transformation of China’s myriad industries.
Media contact Brand Name : Xunce Contact Person: Marketing Team Email: ir@xuncetech.com Tele: 0755-86667876 Website: http://www.xuncetech.com/
Media gallery
