
How Computing Power Trading Centers Support 200 Agent Applications: China Unicom’s Xingluo Scheduling Platform 2.0 Delivers the Standard Answer
English Version
At WAIC 2026 in July, China Unicom released a set of eye-catching industrial data. Its Yuanjing MaaS one-stop Token operation platform integrates more than 200 full-modal large models. The Yuanjing Tianrong Agent Supermarket has launched over 150 preset intelligent agents, fully verified by more than 50 benchmark enterprises across industries. Combined with the enterprise-level agent applications on the UniClaw platform, China Unicom has incubated nearly 200 mature intelligent agents within its ecosystem.
A key industrial question follows: what kind of computing power support is required to stably run 200 agent applications in parallel? Where do standardized Tokens come from? How to realize efficient scheduling and refined management of massive resources? China Unicom’s upgraded Xingluo Scheduling Platform 2.0 provides a complete and systematic solution.
From Intensive Management to Shared Empowerment: The Core Scheduling Logic of Xingluo 2.0
In May 2026, China Unicom Cloud officially launched the upgraded Xingluo Advanced Computing Power Scheduling Platform 2.0. The iteration brings a qualitative leap in service capability, realizing the strategic upgrade of computing power services from traditional “intensive centralized management” to modern “full shared empowerment”.
There is an essential difference between intensive management and shared scheduling. The intensive model adopts centralized control and unified resource allocation, which is essentially a passive service model of “users can only use allocated resources”. In contrast, the shared model enables computing power to be accessed on-demand like water and electricity. Users can schedule accurate computing resources and Token quotas according to actual business demands without pre-occupying or locking idle resources. This revolutionary transformation is supported by two core underlying technologies: heterogeneous computing power pooling and global intelligent scheduling.
The platform’s performance data fully demonstrates its industrial advantages: the unit computing power Token throughput is doubled, the average computing power utilization rate increases by 20%, the business load capacity rises by 40%, and the first-byte time (TTFT) of the self-developed inference engine under high-load scenarios is improved by more than 2 times.
The 40% increase in business load capacity means that the same hardware infrastructure that originally supported 100 agent tasks can now stably carry 140 tasks. Doubled Token throughput indicates that the same computing power resources can output twice the effective Token resources. It is clear that the powerful underlying capability to support 200 large-scale agent applications relies not on blind GPU stacking, but on efficient scheduling optimization and refined resource utilization.
In addition, Xingluo 2.0 has built a full-stack localized independent and controllable system covering domestic chips, self-developed models, independent computing power and national security protection. It realizes full independent controllability from underlying hardware, scheduling algorithms to upper-layer software and security systems. At present, the platform has deployed transit, distribution and one-click scheduling services in more than 300 backbone cloud pools nationwide, achieving millisecond-level ultra-low-latency transmission coverage in core regions including the Beijing-Tianjin-Hebei region, the Yangtze River Delta and the Guangdong-Hong Kong-Macao Greater Bay Area.
Token Upgrade: From By-Products of Computing to Measurable, Usable Standard Resources
The Xingluo platform solves the core problem of reasonable computing power allocation, while intelligent agent operation relies on standardized Tokens — the basic output unit of large model operations. There is an essential conversion logic between underlying computing power and application-end Token resources.
China Unicom takes the lead in completing industrial innovation: converting idle and heterogeneous computing power into standardized, salable and measurable Token services. In May 2026, relying on its nine-in-one full-stack intelligent computing system, Unicom Cloud launched the Xingluo Token Service Platform, realizing full-link technical upgrades covering Token production, circulation, application and full lifecycle operation management.
The platform acts as an efficient “computing power to Token converter” for the AI industry. Enterprise users and developers no longer need to pay attention to underlying hardware investment, GPU quantity and running hours. They only focus on actual Token consumption and business output results. This marks China Unicom’s complete transformation from “selling computing power resources” to “delivering intelligent Token service capabilities”.
Wang Kai, General Manager of the Smart Model Platform R&D Department at China Unicom, interpreted the core logic: “The essence of Token operation layout is a fundamental strategic transformation. We are no longer limited to basic channel services, but shift to intelligent operation, exporting integrated capabilities of computing power, model empowerment and intelligent application services.”
The shift from traffic billing to Token billing is not merely a change of measurement units, but a fundamental upgrade of infrastructure positioning — from traditional communication infrastructure operators to new AI intelligent infrastructure operators.
Front Store & Back Factory Model: The Secret Behind the Massive Growth of 200 Agents
Powerful scheduling computing power and standardized Token resources lay the foundation for industrial AI implementation, and the Yuanjing Tianrong Agent Supermarket provides the carrier for large-scale agent incubation.
Officially launched in March 2026, the platform adopts an innovative “front store + back factory” industrial operation model, perfectly solving the problem of mass agent development and landing.
The “front store” builds a one-stop service entrance for end customers, supporting online trial operation, visual experience and one-click rapid deployment of various intelligent agents. The “back factory” provides full-link ecological empowerment for developers and industrial partners, covering intelligent agent development tools, low-cost computing power support, professional performance testing and standardized compliance review.
Equipped with precise Token metering and pay-as-you-go billing mechanism, the platform effectively reduces the threshold for AI innovation. The overall development cost of intelligent agents is reduced by about 50%, enabling small and medium-sized enterprises and individual developers to conduct low-cost trial and error without one-time high investment in computing power resources, and only pay for actual Token consumption.
The platform has achieved remarkable practical results in multiple vertical industries. In the industrial manufacturing field, the visual quality inspection agent controls the missing detection rate within 0.05% and reduces customer complaints caused by assembly errors by 50%. In the medical industry, the intelligent pre-consultation system increases diagnosis efficiency by 50%, and the anesthesia pre-evaluation agent improves evaluation efficiency by 50% and accuracy by 30%. In government affairs scenarios, the intelligent document review agent shortens the enterprise declaration cycle from two working days to just two hours, greatly improving administrative service efficiency.
From efficient computing power scheduling to standardized Token conversion and finally to large-scale agent implementation, China Unicom has built a complete, closed-loop industrial pipeline for AI intelligent landing.
Unexhub: The Market-Oriented Variable Complementing Token Industrialization
China Unicom adheres to the state-owned enterprise operation model of “platform standardization + package popularization” to realize standardized Token popularization and industrial landing. As a market-oriented supplement to the industry, Unexhub builds a pure market Token trading system to activate industrial vitality.
Positioned as a global Token aggregation, distribution and computing power trading center, Unexhub reshapes the traditional Token distribution pattern with three core advantageous mechanisms, bringing tangible benefits to both suppliers and developers:
Real-Time Full Settlement, Zero Cash Flow Pressure. Small and medium-sized computing suppliers have long been trapped by long settlement cycles and upfront capital occupation. Unexhub realizes synchronous fund return corresponding to Token supply volume, completely eliminating corporate financial pressure and ensuring healthy capital turnover.
Permanent 0 Commission for Early Settled Merchants. Early cooperative suppliers enjoy a 0 platform commission policy, directly cutting the largest operational cost and maximizing profit margins.
API Peak-Valley Bidding Mechanism. Referencing the grid electricity pricing model, API calling prices float dynamically with market supply and demand. Developers can obtain ultra-low-price computing resources during off-peak periods to control costs; suppliers gain premium returns during peak demand periods to activate idle resources efficiently.
China Unicom Xingluo platform undertakes standardized “wholesale and retail” of industrial Tokens, while Unexhub acts as a professional “Token exchange”. Both platforms share the same industrial judgment: Tokens have become the most basic pricing unit and core trading asset in the AI era.
Industry Summary & Future Outlook
The large-scale landing of 200 intelligent agents is not accidental. The efficient scheduling capability of Xingluo 2.0 solves the problem of insufficient computing power efficiency, the Token service platform converts underlying computing power into measurable and controllable industrial resources, and the Agent Supermarket greatly reduces the threshold for AI development and implementation. The superposition of the three capabilities creates the necessary conditions for the large-scale popularization of industrial agents.
It is worth noting that the Token economy ecosystem is still in the early stage of development, and there is still room for improvement in standardized Token pricing, cross-platform interoperation and low-threshold access for small and medium-sized suppliers.
However, the industrial trend is irreversible. The construction of computing power trading centers is an inevitable demand for AI infrastructure maturity. China Unicom’s Xingluo 2.0 provides a standardized state-owned enterprise solution for industrial AI implementation, while Unexhub complements the market-oriented trading system. The dual-path coordination is turning the vision of “Tokens evolving from model by-products to universal currency in the intelligent era” into reality.
Unexhub reshapes global Token trading and distribution logic and builds a one-stop Token aggregation and computing power trading center. The platform has gathered full-spectrum AI suppliers, covering top large models such as Gemini, ChatGPT, Claude, DeepSeek and GLM, as well as high-quality customized models from individual developers and SMEs. Relying on millisecond-level intelligent routing, massive AI resources participate in open market bidding.
End users pay strictly based on actual Token consumption and can enjoy a minimum discount of 30% off the market price. Suppliers can access the platform with one click to open diversified sales channels and realize real-time settlement of call revenue shares with zero threshold.
中文版本
2026年7月WAIC世界人工智能大会上,中国联通公布了一组亮眼的产业落地数据:元景MaaS一站式词元运营平台已汇聚超200款全模态大模型;元景·天融智能体超市上架150余款预置智能体,历经50余家行业标杆企业实战验证;叠加UniClaw平台企业级智能体应用,中国联通体系内已成功孵化近200个成熟智能体应用。
海量智能体规模化并行落地,背后藏着核心产业难题:支撑200个智能体同时稳定运行需要怎样的算力底座?标准化词元资源从何而来?海量算力、词元资源如何高效调度、精细管控?中国联通星罗调度平台2.0的全面升级,给出了产业标准化标准答案。
从集约管控到共享赋能:星罗2.0的核心调度革新
2026年5月,联通云正式发布全新升级的星罗先进算力调度平台2.0。相较于上一代版本,此次迭代实现能力质的飞跃,完成算力服务从「集约集中管控」到「全域共享赋能」的战略跃迁。
很多人难以区分集约与共享的核心差异:传统集约模式是集中管控、统一分配,本质是平台分配什么、用户只能用什么的被动服务模式;而星罗2.0的共享模式,真正实现算力像水电一样随取随用,用户可根据业务实时需求精准调度资源、匹配词元额度,无需提前锁定、闲置占用算力。这一颠覆性升级,依托异构算力池化、全局智能调度两大核心底层技术实现。
平台实测性能数据极具说服力:单位算力词元吞吐率翻倍提升,算力平均利用率提升20%,业务容载率提升40%,自研推理引擎在高负载场景下首字节返回速度(TTFT)提升2倍以上。
40%的业务容载率提升,意味着同等硬件基建下,原本只能承载100个智能体任务的算力,如今可稳定运行140个任务;翻倍的词元吞吐率,代表同等算力资源可产出双倍有效词元。足以可见,联通支撑200个智能体规模化落地,并非依靠堆砌GPU硬件,而是通过极致调度效率优化、精细化资源盘活实现。
与此同时,星罗2.0搭建起「国芯、国模、国算、国盾」四位一体全栈国产化体系,从底层硬件、调度算法到上层软件、安全防护,实现全链路自主可控。目前平台已在全国300余个骨干云池部署中转、分发、一键调度服务,在京津冀、长三角、粤港澳大湾区等核心产业区域,实现毫秒级超低时延传输全覆盖。
词元价值蜕变:从算力附属产物到可计量标准化产业资源
星罗平台解决了算力资源合理分配、高效利用的难题,而智能体落地运行的核心依托是词元——大模型运算的基础输出单元。底层算力与应用端词元之间,存在关键的价值转化链路。
中国联通率先打通产业闭环,实现算力资源的标准化价值转化。2026年5月,联通云依托九位一体全栈智算体系,上线星罗词元服务平台,完成词元生产、流转、场景应用、全周期运营的全链路技术升级,打造行业专属的「算力转词元」高效转换器。
这套体系彻底降低了企业AI应用门槛:用户无需关注底层GPU数量、算力投入时长,只需聚焦实际词元消耗与业务落地效果。这也标志着中国联通的商业模式彻底升级,从单纯「售卖算力资源」,转向「交付标准化词元智能服务能力」。
中国联通智慧模型平台研发部总经理王凯精准解读转型内核:「布局词元经营,本质是一场核心战略升级。我们不再局限于基础渠道服务,而是全面转向智能化经营,对外输出『算力底座+模型能力+智能应用』的一体化AI服务体系。」
从传统流量计费到全新词元计费,看似是计量单位的迭代,实则是基础设施定位的底层革新,完成了从通信基础设施运营商,到AI智能基础设施运营商的身份跃迁。
前店后厂生态模式:解锁200个智能体规模化孵化密码
高效的算力调度底座、标准化的词元资源,为AI产业化落地筑牢根基,而元景·天融智能体超市的创新模式,实现了智能体的批量孵化、快速迭代与规模化落地。
2026年3月上线的元景·天融智能体超市,独创「前店后厂」产业运营模式,完美破解行业智能体开发难、落地慢、成本高的痛点。
「前店」面向政企、开发者用户,打造一站式智能体选购、试用、部署入口,支持在线体验、一键落地,零门槛复用成熟AI能力;「后厂」面向生态合作伙伴,提供智能体开发工具、低成本算力支撑、性能测评、合规审核全链路赋能,全方位助力生态伙伴创新迭代。
平台搭载精准词元计量、按量付费机制,将智能体整体开发成本降低50%。彻底解决中小企业、个人开发者的AI创新顾虑,无需一次性高额投入算力资源,按需消耗、按量付费,低成本完成业务试错与场景落地。
目前平台已在多行业落地标杆成果:工业领域,视觉质检智能体将漏检率控制在0.05%以内,装配错漏引发的客户投诉率下降50%;医疗领域,智能预问诊系统诊疗效率提升50%,麻醉术前评估智能体效率提升50%、评估准确率提升30%;政务领域,智能公文审查智能体将企业申报流程从两天繁琐流程,压缩至两小时快速办结,大幅提升政务服务效能。
从算力高效调度、词元标准化转化,到智能体规模化落地,中国联通搭建起一条完整、闭环的AI产业落地管道。
Unexhub市场化变量:补齐词元交易生态短板
中国联通依托央企资源优势,以「平台标准化+套餐普惠化」模式,实现词元产业规范化落地、规模化普及。而Unexhub以纯市场化模式入局,作为产业重要补充,激活词元交易流通活力。
Unexhub定位全球词元聚合分发算力交易中心,重构传统词元分发格局,三大核心机制为供需两端带来实打实的权益升级:
实时结算,零资金压力。中小算力供应商长期受困于前置算力投入、长周期回款的现金流难题,平台实现词元供应与资金回流实时同步,彻底清零资金占用压力,保障企业良性周转。
早期入驻永久0佣金。针对首批入驻的算力供应商与生态伙伴,平台开放永久0佣金专属权益,直接砍掉行业最大运营成本,大幅提升供应商盈利空间。
峰谷竞价模式,市场化调价。复刻电网峰谷电价机制,打破传统固定API定价弊端。高峰需求期价格合理上浮,供应商溢价增收;低谷闲置期价格下探,开发者低成本调用,以市场杠杆实现算力资源最优配置。
中国联通星罗平台承担产业词元的标准化批发与普惠零售,Unexhub则打造专业化词元交易交易所。二者赛道互补、理念同源,共同印证行业核心趋势:词元已成为AI时代最基础的计价单位与核心交易标的。
产业总结与未来展望
200个智能体的规模化落地,并非偶然的行业成果:星罗2.0的高效调度解决了算力效率瓶颈,词元服务平台实现算力资源标准化、可量化,智能体超市大幅降低AI开发落地门槛,三重能力叠加,构筑起产业智能体规模化普及的核心底气。
目前词元经济生态仍处于发展初期,在标准化定价、跨平台互通、中小供应商低门槛准入等方面,仍有持续优化空间。
但产业趋势已然确定:算力交易中心的规模化建设,是AI基础设施成熟的必然趋势。中国联通星罗2.0交出了央企标准化落地的标准答案,Unexhub补齐市场化交易的生态短板。双路径并行发力,推动词元从模型附属产物,蜕变为智能时代的流通硬通货,让产业愿景全面落地。
Unexhub重塑全球Token流通交易逻辑,搭建一站式Token聚合分发算力交易中心。平台汇聚全层级AI资源,涵盖Gemini、ChatGPT、Claude、DeepSeek、GLM等头部大厂大模型,以及个人、中小企业出品的优质定制模型。依托毫秒级智能路由技术,海量AI资源公开竞价,为企业、开发者提供极致低价的调用服务。
用户支持Token实付结算,最低可享市面3折超低调用价格;供应商一键入驻即可打通全域销售渠道,调用分成实时结算、快速回款,零门槛实现规模化盈利。