The Ambition Behind a $50 Million Subsidy: When a Compute Marketplace Becomes the App Store for AI Agents
Bilingual Article · English First / 中文版见下方
On August 7, 2026, the Beijing Economic-Technological Development Area released a policy document titled Several Measures to Support the High-Quality Development of a Token-Driven Intelligent Economy (Trial).
The ten-point policy quickly earned an industry nickname: the “Ten Token Measures.”
What caught the industry's attention was a line buried in the tenth measure: operators providing multi-capability aggregation services could receive up to RMB 50 million in financial support.
That RMB 50 million is not primarily about buying GPUs or building data centers.
It is about building an “entrance.”
And that entrance could eventually become something much bigger: an App Store for AI agents.
What Does This “Super Entrance” Actually Look Like?
The core logic of the Ten Token Measures can be summarized in nine Chinese characters:
Factories produce Tokens. Platforms distribute Tokens. Communities consume Tokens.
On the production side, Beijing Yizhuang has already launched the city's first Token factory, as well as its first state-owned Token factory. Multiple large-scale Token factories are also planned.
On the infrastructure side, the policy supports the development of inference-oriented chips. On the model side, it encourages collaboration between general-purpose and industry-specific models.
Meanwhile, annual computing-power and data vouchers worth RMB 100 million are intended to help smaller companies access advanced AI infrastructure.
But the real breakthrough may not happen at the production layer.
It may happen at the distribution layer.
The policy encourages technology companies to build an “intelligent application store”-style super entrance that can operate across different models and computing platforms, while providing developers with unified access and settlement services.
In practical terms, imagine an AI version of a massive electricity-dispatching center.
Developers do not need to negotiate separately with every model provider or computing supplier. Instead, one unified interface provides access to models, compute resources, Tokens and settlement.
The policy also encourages infrastructure capable of delivering end-to-end intelligent-agent tasks and exploring regulatory sandbox programs for intelligent-agent payments.
Put simply, the “multi-capability aggregation service operator” envisioned by the policy is essentially a platform combining:
Model aggregation
Compute scheduling
Unified Token settlement
AI-agent distribution
Intelligent-agent payments
That is much closer to an AI operating infrastructure layer than a traditional cloud marketplace.
Unexhub Is Already Moving Toward This Position
This is where Unexhub enters the picture.
Unexhub is building a compute marketplace centered on Token aggregation and distribution.
On the supply side, it brings together distributed AI compute and model providers. On the demand side, it connects developers and enterprise users. In between, Token acts as a unified unit for measuring usage, pricing services and settling transactions.
That architecture closely matches the policy's vision of a cross-model, cross-modal and cross-compute “super entrance.”
The settlement layer is equally important.
The Ten Token Measures calls for unified Token settlement services. Unexhub's planned financial settlement and clearing system focuses on real-time settlement, allowing suppliers to receive corresponding payments as compute resources are consumed.
For smaller suppliers, the benefit is straightforward: less capital locked in receivables and less dependence on lengthy monthly settlement cycles.
Every completed Token transaction can become a corresponding cash-flow event.
The Ecosystem Layer: Give Suppliers a Reason to Join
Beijing Yizhuang has already established a large OPC community, providing more than 40,000 square meters of physical space and serving more than 500 OPC participants.
Unexhub is approaching the same ecosystem problem from the marketplace side.
Early suppliers can access platform incentives including commission rates as low as 0%, while the platform's real-time settlement mechanism is designed to improve cash-flow efficiency.
This combination matters.
A physical ecosystem uses space, services and policy support to attract developers.
A digital marketplace can use lower transaction costs, faster settlement and broader customer access to attract suppliers.
Different mechanisms, but the same objective:
Get the supply side moving first, then let the ecosystem attract demand.
Dynamic Pricing: From a Fixed Price List to a Market Signal
There is another important layer.
The Ten Token Measures aims to transform Token from an underlying production resource into a practical productivity tool for intelligent applications.
Unexhub's proposed API bidding model pushes this concept further.
Instead of every API having a completely rigid price, pricing can respond dynamically to market supply and demand.
During low-demand periods, lower prices can give developers more room to experiment.
During peak demand, higher prices can provide suppliers with stronger incentives to release additional capacity.
The result is a shift from:
“Here is the price.”
to:
“Here is the current market signal.”
That difference could become increasingly important as AI workloads become more volatile.
The Battle for the “Super Entrance” Has Just Begun
The RMB 50 million subsidy is not designed for one specific company.
It is aimed at an entire category of infrastructure operators.
That is the bigger signal.
Beijing Yizhuang is effectively using policy incentives to define a new type of infrastructure:
An App Store for the intelligent-agent economy.
Unlike Apple's relatively closed App Store ecosystem, this envisioned AI super entrance would be more open, cross-platform and centered on unified settlement.
Developers would not need to:
Open separate accounts for every model;
Prepay multiple AI providers;
Maintain dozens of API integrations;
Reconcile separate Token bills;
Build their own distribution infrastructure.
AI suppliers would also have less need to independently build customer acquisition channels, billing systems and collection processes.
The platform becomes the connective tissue.
Why Unexhub Wants to Be Part of This Infrastructure
Unexhub is positioning itself around the same fundamental transformation: turning fragmented AI resources into standardized, tradable and accessible capabilities.
Its marketplace brings together AI providers of different sizes, including major model ecosystems such as Gemini, ChatGPT, Claude, DeepSeek and GLM, while also opening the door to high-quality models supplied by individuals and smaller companies.
Instead of forcing developers to negotiate with every provider separately, Unexhub aims to aggregate supply and use market competition to improve pricing efficiency.
For developers, the practical benefits are particularly straightforward:
Pay for what you actually use
Developers can settle according to actual Token consumption rather than committing to large amounts of unused capacity.
Potentially lower AI costs
Through supplier aggregation and competitive pricing, Unexhub says developers can access Token services at as little as 30% of prevailing market prices, depending on the applicable model and supply conditions.
That can make a meaningful difference for high-volume applications, agent workflows and development-stage experimentation.
Millisecond-level routing
The platform is designed to route requests across available model and compute resources, reducing the friction of switching between providers.
One integration instead of many
Developers can potentially access multiple AI providers through a unified interface rather than maintaining separate integrations for every model.
More efficient supplier monetization
For suppliers, the proposition works in the opposite direction: connect once, access a broader demand pool, and receive transaction-based revenue through the platform's settlement mechanism.
The Real Value of the $50 Million
The most important part of the Ten Token Measures may not actually be the RMB 50 million.
The bigger message is that governments are beginning to recognize AI capability distribution itself as infrastructure.
The evolution looks something like this:
GPU → Compute → Token → Model → Agent → Application → Marketplace
At the beginning of the AI era, companies competed for GPUs.
Then they competed for models.
Now the next battle may be over who controls the distribution layer connecting compute, models, agents and users.
Beijing Yizhuang's Ten Token Measures, Unexhub's Token marketplace, Token aggregation stores and AI-agent marketplaces emerging in different regions—all point toward the same structural shift:
AI capabilities are moving from fragmented resources toward standardized, programmable and tradable infrastructure.
Yizhuang already has more than 40,000 square meters of OPC community space and has served more than 500 OPC entities. Its long-term plan targets an intelligent-economy industrial scale exceeding RMB 400 billion during the 15th Five-Year Plan period.
If the “super entrance” eventually works as envisioned, the compute marketplace will no longer be merely a place to buy computing power.
It could become the central nervous system of the intelligent economy.
And that makes the competition much bigger than a race for cheaper Tokens.
It is a race to determine who becomes the gateway through which AI capabilities are discovered, accessed, priced, distributed and monetized.
The battle for the AI super entrance has only just begun.
中文版
“超级入口”背后 5000 万补贴的野心:当算力交易中心变成智能体的 App Store
2026年8月7日,北京经济技术开发区发布《关于支持词元驱动智能经济高质量发展的若干措施(试行)》。
全文共十条,业内很快将其称为 “词元十条”。
真正让行业高度关注的,是第十条里的一句话:
对提供多能力聚合服务的运营主体,给予最高5000万元资金支持。
5000万元,并不是单纯用来买 GPU,也不是用来建厂房。
它真正瞄准的,是一个更重要的东西——“入口”。
而这个入口未来可能演变成 AI 智能体时代的 App Store。
“超级入口”究竟是什么?
“词元十条”的核心逻辑,可以概括为九个字:
工厂供词元、平台分词元、社区用词元。
生产端,北京亦庄已经建设词元工厂,并规划进一步扩大大规模词元基础设施。
芯片侧支持推理专用芯片研发,模型侧推动通用模型与行业模型协同发展,同时通过每年1亿元规模的算力券、数据券等政策工具,降低中小企业使用先进 AI 基础设施的门槛。
但真正可能引爆词元经济的,并不只是生产端。
而是分发端。
政策鼓励建设类似“智能应用商店”的超级入口,跨越不同模型和不同算力平台,为开发者提供统一调用、统一调度和统一结算服务。
简单来说,它就像 AI 世界里的大型“电力调度中心”。
开发者不需要分别寻找模型、购买算力、管理 API,也不需要面对十几套不同的计费规则。
一个入口,就可以连接模型、算力、Token 和结算体系。
这实际上已经不是传统意义上的云平台,而是一层新的智能经济基础设施。
Unexhub,恰好正在这个位置上
Unexhub 正在建设面向 Token 聚合与分发的算力交易中心。
上游聚合不同规模的 AI 模型与算力供应商,下游连接企业用户和开发者,中间以 Token 作为统一的计量、定价与结算单位。
这与“词元十条”所描述的跨模型、跨模态、跨算力平台的“超级入口”逻辑高度一致。
在结算层,政策强调建立统一的词元结算体系。
Unexhub 则进一步将重点放在即时结算上:供应商提供算力并产生调用后,可以同步获得对应的交易收益。
对于中小供应商而言,这意味着资金不必长期沉淀在应收账款里,也不必完全依赖月底统一结算。
每一次 Token 调用,都可以对应一次现金流回流。
生态层:先让供应商愿意进来
北京亦庄已经建设 OPC 社区,拥有超过 4万平方米空间载体,并累计服务超过 500家 OPC 主体。
Unexhub 解决的则是线上生态的冷启动问题。
对于早期入驻供应商,平台提供包括最低0佣金在内的激励政策,同时通过即时结算机制提升资金周转效率。
这两个模式虽然不同,但逻辑高度一致:
先把供给侧做起来,再吸引需求侧进入。
对于供应商而言,降低交易成本、加快回款速度、获得更多客户触达机会,本身就是非常现实的权益。
动态定价:从“固定价目表”走向“市场信号”
“词元十条”希望推动词元从底层生产资料转化为上层应用真正可以使用的生产力工具。
Unexhub 的 API 竞价模式则进一步把 Token 的市场属性放大。
API 价格不再只是一个固定数字,而可以根据供需变化动态调整。
需求较低的时段,价格可以下降,为开发者提供更低成本的实验空间。
需求高峰期,价格则可以上升,为供应商释放算力资源提供更强的经济激励。
价格因此从:
“平台告诉你多少钱。”
变成:
“市场告诉你现在值多少钱。”
随着 AI 工作负载越来越复杂,这种动态定价可能成为算力市场的重要基础设施。
“超级入口”之争,才刚刚开始
5000万元补贴并不是专门给某一家企业准备的。
它真正针对的是一类新的基础设施运营主体。
这才是“词元十条”最值得关注的地方。
北京亦庄实际上正在通过政策激励定义一种新的产业基础设施:
智能体时代的 App Store。
它并不是苹果 App Store 那样的封闭生态,而更像一个开放、跨平台、统一结算的 AI 超级入口。
开发者不再需要:
为不同模型分别开户;
分别充值;
维护十几套 API;
对接不同 Token 计费体系;
手工整理多份账单。
供应商也不再必须独立寻找客户、建设销售渠道、搭建结算系统并承担长期回款压力。
平台成为连接供需两端的基础设施。
为什么 Unexhub 要争夺这个位置?
Unexhub 的核心定位,本质上也是同一场变革:
把分散的 AI 资源变成标准化、可调用、可交易的 AI 能力。
目前 Unexhub 聚合不同规模的 AI 供应商,包括 Gemini、ChatGPT、Claude、DeepSeek、GLM 等主流模型生态,同时也面向个人及中小企业提供的优质 AI 模型开放接入。
通过聚合大量供应商并引入市场竞争,Unexhub 希望让开发者不再需要逐一寻找供应商,而是通过一个统一入口获取更多 AI 能力。
对于开发者而言,最直接的价值包括:
按实际 Token 使用量付费
开发者可以根据实际 Token 消耗进行结算,不必为了获得模型能力而提前承担大量闲置资源成本。
更具竞争力的 Token 价格
通过供应商聚合和市场竞价,Unexhub 表示,开发者在符合相应模型与供给条件的情况下,最低可享受市场价格约3折的 Token 服务。
对于高频调用、AI Agent、多轮推理以及开发测试场景而言,这类价格差异可能直接影响整体运营成本。
毫秒级路由
平台通过路由机制连接不同模型和算力资源,降低开发者在不同供应商之间切换的技术与运营成本。
一个入口连接多家模型
开发者无需为每一个模型重复搭建完整的调用体系,可以通过统一接口获得多模型能力。
供应商更快实现算力变现
对于供应商而言,则可以通过一次接入触达更广泛的需求方,并根据实际调用获得相应收益。
这意味着:
开发者得到更多选择,供应商得到更多订单,平台负责连接双方。
5000万元背后,真正值得关注的是什么?
“词元十条”最值得关注的,也许并不是那 5000万元。
真正重要的信号是:
AI 能力的分发本身,正在被视为一种基础设施。
整个产业链正在发生这样的变化:
GPU → 算力 → Token → 模型 → Agent → 应用 → 市场
AI 早期,企业争夺的是 GPU。
后来,企业争夺的是模型。
而下一阶段真正的竞争,很可能转向:
谁能够掌握连接算力、模型、智能体与用户的分发入口。
北京亦庄的“词元十条”、Unexhub 的 Token 算力交易中心,以及全国不同地区正在出现的 Token 集合店、智能体超市,都指向同一个趋势:
AI 能力正在从碎片化资源,走向标准化、可编程、可交易的基础设施。
北京亦庄目前已经拥有超过4万平方米 OPC 社区空间,并累计服务超过500家 OPC 主体;按照规划,“十五五”期间区域智能经济产业规模力争突破 4000亿元。
如果未来这个“超级入口”真正成熟,那么算力交易中心就不再只是一个“买算力”的地方。
它可能成为整个智能经济的:
“中枢神经系统”。
到那时,竞争也就不再只是“谁的 Token 更便宜”。
真正的竞争将变成:
谁能够成为 AI 能力被发现、被调用、被定价、被分发、被变现的那个入口。
超级入口之争,才刚刚开始。