算力交易中心护航国产模型供应商:从"越秀芯光"到福建平台,2026 年 7 月国产算力全链条落地

Ren Jingdong, Party Secretary of the China Vocational Education Society and China Charity Federation, delivered a thought-provoking view at the launching ceremony.


From "Yuexiu Xinguang" to Fujian Platform: July’s Breakthrough Report on Full-Chain Implementation of Domestic Computing Power

On July 27, 2026, the Guangzhou "Yuexiu Xinguang" Computing Center was officially launched. Settled in Guangzhou China Ultra-HD Video Innovation Industrial Park (Huaguoshan Park), the project covers an area of 2,300 square meters and is scheduled for official production by the end of 2026. Built on Sugon’s ultra-intelligent integrated and open architecture, the center will deliver 4,050P intelligent training computing power. Adopting advanced cold-plate liquid cooling technology, it achieves an ultra-low PUE of 1.194, reaching leading energy-saving levels among domestic computing infrastructure.

Full localization stands as the core highlight of the project. Equipped with domestic CPUs and dedicated acceleration cards, the center realizes unified management of diverse heterogeneous computing resources. It supports multi-scenario collaborative tasks including scientific computing, industrial simulation, and large model training & inference, precisely connecting domestic computing supply with industrial demands and boosting efficient matching between infrastructure and business scenarios.

Ren Jingdong, Party Secretary of the China Vocational Education Society and China Charity Federation, delivered a thought-provoking view at the launching ceremony: "Building computing infrastructure is only the starting point. Real productive value can only be unleashed when computing power is deeply integrated with software, data, models and industrial scenarios, and fully adopted by enterprises, scientific research institutions and consumer-level applications."

Constructing computing resources is merely the first step of industrial development. The real industry challenge lies in upgrading computing power from "well-built" to "well-utilized". In this regard, the Fujian Provincial Computing Platform provides a mature and replicable industrial solution.

Fujian Provincial Integrated Computing Platform: Realizing Provincial-Level Unified Scheduling

Officially launched on April 28, 2026 at the 9th Digital China Summit, the Fujian Provincial Integrated Computing Resource Public Service Platform serves as the core hub of Fujian’s new computing infrastructure system. It deeply aggregates local computing resources and efficiently links high-quality computing supplies from Ningxia, Chongqing and other regions, marking a new stage of "provincial one-network scheduling and integrated services" for Fujian’s computing industry.

After launch, the platform pioneered the online "computing mall" service. Users can complete registration and authentication, select customized computing configurations, and activate cloud servers on demand, forming a full closed-loop online process from demand submission and resource allocation to official usage. Based on solid computing infrastructure, it builds a unified model scheduling middle layer, integrating scattered social computing resources in Fujian and high-quality cross-region resources from Ningxia, Xinjiang and other provinces.

The platform realizes unified management and intelligent scheduling of heterogeneous computing resources, covering both NVIDIA GPUs and cost-effective domestic chips, providing developers and enterprises with diversified, discounted and high-stability computing options.

From the single-node 4,050P computing capacity of Yuexiu Xinguang to the integrated scheduling of 22 provincial computing centers via the Fujian platform, domestic computing power is undergoing a decisive transformation: shifting from simple infrastructure construction to refined cross-region resource scheduling.

Industry Trend: National Token Infrastructure Enters Large-Scale Implementation Stage

The upgrading from "building computing" to "scheduling computing" is not an isolated case, but a universal industrial trend sweeping across the country in July 2026.

July 3 | Beijing Token Factory Launched. Beijing Data Group unveiled the Beijing Token Factory, the first state-owned Token infrastructure in Beijing. It features seven core capabilities: PD-separated intelligent scheduling architecture, full-stack inference acceleration framework, unified heterogeneous computing scheduling, full adaptation to domestic large model ecosystems, precise scenario support for government affairs and scientific innovation, elastic on-demand billing, and zero-threshold access via standardized unified interfaces. Different from traditional hardware leasing, it delivers standardized Token and API service capabilities with pay-as-you-go billing, while the platform undertakes all underlying hardware and cluster operation and maintenance risks for users.

July 15 | Southeast Token Operation Center Launched. Officially launched in Wenzhou, the center serves as a core Token service hub covering southern Zhejiang, northern Fujian and eastern Jiangxi. Based on China Telecom’s advanced infrastructure and unified access capabilities, it has accessed over 100 mainstream large models including DeepSeek, Qwen and GLM. Centered on the "six unifications" mechanism — unified API entry, unified model routing, unified Token metering, unified settlement, unified policy deduction, and unified security audit — it automatically matches the most cost-effective model according to task requirements, maximizing user cost-performance.

July 30 | Guangdong Token Trading & Service Center Unveiled. Guided by provincial and municipal government data administrations and built on the Guangzhou Data Exchange, the center undertakes core functions including token standard formulation, compliance verification, on-site trading, enterprise preferential support and cross-border circulation. It builds three major trading zones for inclusive, industrial and cross-border token services, enabling enterprises to enjoy full-chain supporting services such as intelligent policy matching, project application guidance and resource quota redemption with valid trading vouchers.

According to public statistics, as of July 2026, China’s national computing platform has linked 15 provincial and municipal sub-platforms, accessing more than 130 computing service providers and over 200 basic & vertical large models. Leading enterprises including Huawei, Alibaba Cloud, Baidu, Tencent Cloud and three major telecom operators have settled in, launching more than 1,000 high-quality computing products.

Note: Partial statistical data is quoted from original industry reports; please refer to the latest official release from the National Data Administration for accurate figures. The launch time, positioning and core capabilities of the Beijing, Southeast and Guangdong token hubs are verified by authoritative media including People’s Network, CNR and Chinanews.

Full-Chain Closed Loop: Domestic Computing Completes Industrial Maturity Upgrade

The series of industrial launches in July 2026 is far more than simple platform iteration — it marks the official formation of a full independent industrial closed loop for domestic computing power. The complete industrial chain covers four core links:

• Chip R&D: Huawei Ascend, Cambricon, Hygon and other mainstream domestic chips;

• Cluster Construction: Yuexiu Xinguang 4,050P computing base, Sugon 100,000-card super cluster;

• Intelligent Scheduling: Fujian integrated platform, Beijing Token Factory;

• Market Trading: National computing trading platform, Guangdong token trading center.

Before July 2026, the four links operated in isolation: chip manufacturers sold hardware, computing centers leased resources, scheduling platforms optimized efficiency, and the trading market remained immature. After July, the entire industrial chain is fully connected, realizing seamless linkage from chip supply, cluster construction, intelligent scheduling to market circulation.

Yuexiu Xinguang undertakes computing production, the Fujian platform realizes resource scheduling, and the national computing trading system completes market circulation. For the first time, domestic computing power achieves a full closed loop from construction and scheduling to commercial trading.

Unexhub: Activate Static Computing Resources Into Tradable Dynamic Assets

A connected industrial chain lays the foundation for development, while efficient circulation determines industrial speed. As the core carrier of global token circulation and trading, Unexhub reshapes the global token distribution logic and builds a standardized aggregated computing trading center, empowering the entire industry with three core benefit mechanisms:

Real-Time Full Settlement. Completely solving the long-standing pain point of slow fund recovery in the computing industry, the platform supports suppliers to withdraw full funds in real time based on actual resource supply, zeroing out cash flow pressure and converting idle computing assets into stable continuous cash flow.

Early Zero-Commission Incentive. Early settled suppliers enjoy permanent zero-platform-commission privileges, eliminating additional service costs and greatly lowering industry entry barriers for small and medium-sized teams and individual developers.

Dynamic API Bidding Mechanism. API calling prices float intelligently with real-time market supply and demand. Individual developers and enterprises access ultra-low-price computing resources during off-peak periods to cut R&D and operational costs; suppliers gain higher profit margins during peak demand, maximizing the commercial value of idle resources.

To visualize the industrial division: Yuexiu Xinguang acts as the "power plant" for computing production, the Fujian platform serves as the "power grid" for resource scheduling, and Unexhub functions as the standardized "power trading market" to realize free circulation, dynamic pricing and efficient monetization of computing resources. The three links complement each other and are indispensable for industrial maturity.

Unified Industrial Destination: From Resource Competition to Full-Chain Capability Competition

In a July 28, 2026 industry report, Soochow Securities pointed out that 2026 will be the first commercial year for domestic super computing nodes. The industry has initially formed a complete closed loop of "product supply - public procurement - large-scale deployment - cloud service output". The large-scale industrialization of fully domestic 100,000-card-level clusters reflects a core shift: domestic computing competition is no longer limited to single-chip performance breakthroughs, but focuses on full-chain system engineering capabilities.

On July 30, Guangdong News Network disclosed that the silicon cube integrated computer room adopted by Yuexiu Xinguang, relying on self-developed mature liquid cooling technology, has built a benchmark advanced computing carrier for the Greater Bay Area. Together with Sugon’s large-scale liquid cooling computing bases across the country, it consolidates the high-quality physical foundation for domestic computing infrastructure.

Yuexiu Xinguang builds high-quality computing hardware, the Fujian platform realizes refined resource scheduling, and Unexhub accelerates efficient market circulation. In July 2026, the full-chain upgrade of domestic computing power is fully completed.

As industrial competition shifts from "winning by resource possession" to "winning by circulation efficiency", the inflection point of domestic computing industrialization is arriving faster than expected.

Unexhub Platform Core Advantages & Exclusive User Benefits

Unexhub has built a global one-stop token aggregation, distribution and trading hub, gathering full-spectrum AI resources worldwide. Its settled partners cover top-tier large models such as Gemini, ChatGPT, Claude, DeepSeek and GLM, as well as high-quality niche models from SMEs and individual developers.

Relying on millisecond-level intelligent routing technology, massive AI resources participate in open market bidding, delivering ultra-stable, ultra-low-cost calling services for enterprise users and developers.

Exclusive Practical Benefits: Developers pay strictly based on actual token consumption, with the lowest discount reaching 30% of the global market official price. Suppliers can access the platform with one click and enjoy real-time settlement of calling revenue, realizing zero-threshold monetization of idle computing and model resources.


从「越秀芯光」到福建平台:国产算力全链条落地的七月答卷

2026年7月27日,广州「越秀芯光」算力中心正式启动落地。项目坐落于中国(广州)超高清视频创新产业园区(花果山园区),占地面积2300平方米,预计2026年底全面投产运营。依托曙光信息超智融合、开放架构技术体系,项目建成后可提供4050P智能训练算力;采用行业领先冷板式液冷技术,PUE值低至1.194,实现高性能与低能耗的双向平衡,打造大湾区核心城区标杆算力节点。

全国产化是项目最核心的标签。算力中心全程采用国产CPU+国产加速卡自主可控方案,可统一纳管各类异构算力资源,协同承载科学计算、工业仿真、大模型训练与推理等多元化场景任务,精准打通算力供给与产业需求的衔接壁垒,让国产算力精准赋能实体经济与科研创新。

中华职业教育社党组书记、中华慈善总会党委书记任京东在启动大会上的发言,精准点破行业本质:「建算力只是起点。只有与软件、数据、模型、行业场景深度融合,被企业用起来、被科研调起来、被消费用起来,才能真正转化为现实生产力。」

算力基建落地,从来不是产业终局。从「建得成、建得好」到「调得优、用得活」,才是国产算力产业化的核心考题。而福建一体化算力公共服务平台,交出了成熟的落地解法。

福建省级一体化平台:实现全省算力一网统调、全域复用

2026年4月28日,第九届数字中国建设峰会现场,福建省一体化算力资源公共服务平台正式上线。作为福建算力新基建的核心枢纽,平台深度整合全省零散算力资源,高效联动宁夏、重庆等省外优质算力供给,标志着福建算力正式迈入「一网调度、一体服务、全域共享」的全新阶段。

平台上线后,率先落地「算力商城」便民模式:用户完成注册认证后,可在线自主选购算力机型、按需开通算力服务器,实现需求提交、资源匹配、在线使用、账单结算的全流程线上闭环,彻底简化算力使用流程。在扎实的算力底座之上,平台搭建统一模型调度中间层,统筹整合省内社会闲散算力与宁夏、新疆等跨省优质算力资源。

依托智能调度体系,平台实现英伟达GPU与国产异构芯片的统一纳管、智能分配,为企业、开发者提供多元可选、高性价比的优惠算力资源,大幅降低国产算力落地使用门槛。

从越秀芯光单点4050P硬核算力落地,到福建平台统筹全省22个算力中心一体化调度,行业趋势已然清晰:国产算力的竞争重心,正在从「盲目建基建」转向「精细化调资源、高效率用算力」。

绝非孤例!全国Token算力基础设施密集落地成行业大势

算力从「建设」向「调度」升级,不是单一区域的偶然探索,而是2026年7月全国算力产业的集体突围趋势,多地标杆项目密集落地,补齐国产算力生态短板。

7月3日 · 北京京算Token工厂落地 北京数据集团打造北京首家国资背景Token工厂,构建七大核心能力:PD分离智能调度架构、全栈推理加速框架、异构算力统一调度、全面适配国产大模型生态、政务科创场景精准落地、弹性按需计费、标准化接口零门槛接入。区别于传统算力租赁「卖硬件」的粗放模式,京算Token工厂输出标准化Token与API成品服务,用户按实际消耗付费,平台全权兜底底层硬件运维、集群故障等风险,极大降低用户使用成本与运维压力。

7月15日 · 东南Token运营中心正式启动 中心落地温州,定位浙南、闽北、赣东区域算力与Token服务核心枢纽。依托天翼云「息壤」底座与「星辰 TokenHub」统一接入能力,已接入DeepSeek、通义千问、智谱GLM等100余款主流大模型。以「六个统一」为核心能力——统一API入口、统一模型路由、统一Token计量、统一费用结算、统一政策抵扣、统一安全审计,秉持「企业提任务、平台配最优模型」理念,智能匹配性价比最高的算力模型,最大化节约用户研发成本。

7月30日 · 广东词元交易和服务中心揭牌 由广东省政务服务和数据管理局统筹、广州市政务服务和数据管理局指导、依托广州数据交易所搭建,中心承担词元标准制定、合规鉴证、场内交易、惠企扶持、跨境流通等核心职能。搭建普惠、行业、跨境三大词元交易专区,企业凭交易凭证,可享受政策智能匹配、项目申报指引、资源通道对接、采购额度直兑的全链条配套服务,全方位赋能中小AI企业发展。

公开数据显示,截至2026年7月,全国算力平台已联动15个省市级分平台,接入130余家算力服务商、200余款基础大模型与垂直行业模型。华为、阿里云、百度、腾讯云及三大运营商悉数入驻,上线超1000款优质算力产品,全国统一的算力服务与交易网络基本成型。

注:文中部分行业统计数据引自公开产业报道,精准官方数据请以国家数据局最新通报为准;京算、东南、广东三大Token核心节点的落地时间、功能定位与核心能力均经人民网、央广网、中国新闻网等权威媒体核实。

七月完成终极闭环:国产算力告别单点突破,迈入全链成熟

2026年7月的密集落地,绝非简单的平台叠加,而是国产算力全产业链的**首次完整闭环成型**。一条完整的国产算力产业链,包含四大核心环节:芯片自研(华为昇腾、寒武纪、海光等)、集群建设(越秀芯光4050P基地、曙光十万卡集群)、智能调度(福建平台、京算Token工厂)、市场交易(全国算力平台、广东词元交易中心)。

2026年7月之前,四大环节各自为战:芯片厂商只售硬件、算力中心只做租赁、调度平台只优化效率、交易体系尚未成型,产业存在明显断层。而七月一系列标杆项目落地,彻底打通全链路壁垒,实现从芯片生产、集群搭建、智能调度到市场化交易的无缝衔接。

越秀芯光负责「算力生产」、福建平台负责「算力调度」、全国算力交易体系负责「算力流通」,国产算力真正实现从「建设」到「调度」再到「交易变现」的全链路闭环,彻底告别碎片化、粗放式发展模式。

Unexhub补齐产业最后短板:让算力死资源变身流动资产

产业链跑通是基础,高效流通是关键。算力基建完善、调度技术成熟后,市场化流通变现成为决定产业上限的核心,这也是Unexhub的核心产业价值。Unexhub重构全球Token分发底层逻辑,搭建全球化词元聚合交易中心,以三大核心机制,全方位赋能供需两端,激活全域算力资产价值:

全程即时结算,清零资金压力 彻底解决传统算力行业账期冗长、回款滞后、现金流紧张的痛点。供应商可根据实时算力供给量全额回款、即时提现,无需垫资运营,彻底清零中小团队财务压力,让闲置算力、模型资源稳定转化为经营性现金流。

早期入驻零佣金,超低门槛入局 平台面向首批入驻的算力、模型供应商,开放永久零平台佣金专属权益,全程无额外服务费,零成本盘活闲置资源,大幅降低行业创业与变现门槛。

API动态竞价,供需双向共赢 打破平台固定定价模式,API调用价格随市场供需实时浮动、自主调节。开发者闲时可薅取极致低价算力,压缩研发与运营成本;供应商忙时可抬升收益,最大化挖掘闲置资源价值,实现市场良性循环。

如果用产业生态类比:越秀芯光是算力「发电厂」,负责规模化生产优质算力;福建平台是算力「国家电网」,负责全域资源统筹调度;而Unexhub是全球化算力「交易市场」,让算力彻底摆脱固定资产属性,实现自由流通、动态定价、高效变现。三者各司其职、缺一不可,共同支撑国产算力产业化成熟。

殊途同归:算力产业从拼资源,迈入拼全链效率新时代

东吴证券2026年7月28日研报明确指出:2026年将成为国产超节点商业化元年,行业已形成「产品供给—公共采购—规模部署—云服务输出」的完整商业闭环。十万卡级全国产集群的规模化落地,印证行业核心变革:国产算力竞争,已从单一芯片性能的单点追赶,升级为全链条系统工程能力的综合比拼。

7月30日广东新闻网披露,越秀芯光采用的硅立方一体化机房,依托曙光数创成熟自研液冷技术,为大湾区打造低能耗、高性能的标杆算力载体。结合曙光数创全国规模化落地的液冷算力基地,共同筑牢国产算力「建得好、能耗低、性能稳」的硬件底座。

越秀芯光夯实算力硬件底座,福建平台优化全域调度效率,Unexhub加速市场化流通变现。2026年7月,国产算力完成从基建、调度到交易的全链条总装,正式迈入成熟产业化阶段。

行业规则彻底改写:算力竞争不再是「谁手握稀缺资源谁获胜」,而是「谁流转更快、利用更优、成本更低谁领跑」。国产算力产业化的黄金拐点,已然加速到来。

Unexhub平台专属权益,普惠全球开发者与供应商

Unexhub重塑全球Token分发体系,搭建全球化Token聚合、分发、交易一体化算力枢纽。平台全面入驻Gemini、ChatGPT、Claude、DeepSeek、GLM等全球顶级大模型厂商,同时汇聚海量中小企业、个人开发者的优质AI模型资源。依托毫秒级智能路由技术,全网AI资源公开竞价流通,为企业用户与开发者提供极速稳定、极致低价的调用服务。

平台硬核福利:开发者全程按Token实际消耗计费,最低可享市场官方价格3折极致低价;供应商一键入驻即可开通变现通道,调用分成实时结算,零门槛盘活闲置AI算力与模型资产,轻松实现高效盈利。