
Prediction: AI Compute Exchanges Are Being Rebuilt—The “App Store Moment” for Model Providers Is Arriving
Bilingual Article|English First, Chinese Version Below
English readers can continue directly. 中文读者请向下滑动至 中文版。
On July 24, 2026, The Wall Street Journal reportedly revealed that payment giant Stripe was in talks to acquire AI model aggregation platform OpenRouter at a valuation approaching $10 billion.
Only a little more than two months earlier, OpenRouter had reportedly been valued at approximately $1.3 billion. If the proposed transaction moves forward at the reported price, its valuation would have increased nearly sevenfold within one quarter.
That raises an obvious question:
Why would a company that trains no frontier models, owns no major GPU clusters, and is often described as a “model reseller” be worth almost $10 billion?
The answer lies in a structural shift across the AI industry:
Distribution is becoming as valuable as production—and may eventually become even more powerful.
For the first stage of the generative AI boom, attention concentrated on whoever could train the strongest model.
The next stage will focus increasingly on whoever controls:
how developers discover models;
how requests are routed;
how prices are compared;
how usage is billed;
how suppliers are settled;
how models are packaged into complete services.
AI infrastructure is approaching its own “App Store moment.”
Model providers will continue building the engines, but third-party platforms are beginning to control the roads, tollgates, traffic rules, and payment systems around them.
Stripe × OpenRouter may be only the beginning.
1. What Is OpenRouter? One Entrance to More Than 400 Models
OpenRouter’s core product is remarkably simple:
It brings the APIs of more than 400 AI models into one standardized interface.
Without an aggregation platform, a developer may need to separately:
register accounts with OpenAI, Anthropic, Google, DeepSeek, Kimi, and other providers;
negotiate different commercial terms;
manage multiple API keys;
adapt incompatible request formats;
monitor separate rate limits;
reconcile fragmented invoices;
build provider-specific failover systems;
track model upgrades and deprecations.
With OpenRouter, many of those tasks can be consolidated into one integration.
A developer can connect once and gain access to a broad model catalog that may include flagship, balanced, lightweight, multimodal, reasoning, and coding models.
The platform also provides routing capabilities that can consider:
task complexity;
price;
response latency;
model availability;
context-window requirements;
tool support;
provider reliability.
If one provider becomes unavailable or reaches a rate limit, the request can be redirected to a compatible alternative.
For small and medium-sized development teams, the main value is not merely lower API pricing.
It is the ability to avoid building and maintaining an entire multi-model infrastructure layer independently.
One API can replace dozens of integrations.
2. OpenRouter’s Growth Shows Where the Market Is Moving
According to the figures cited in the original material, OpenRouter processes approximately 25 trillion tokens per week, equivalent to roughly 100 trillion tokens per month.
Its global user base reportedly exceeds eight million, while annualized revenue has reached approximately $50 million after rising fivefold within six months.
These numbers matter because they reveal how quickly developers are moving away from direct, single-provider access.
Instead of making a permanent decision such as:
Our entire product will run on this one model.
More teams are adopting a different approach:
Connect to a model gateway first, then route each task to the model that best matches its cost, speed, and capability requirements.
This changes the center of user loyalty.
Developers may no longer build their infrastructure around one model vendor.
They build around the gateway.
When a new flagship model launches, the platform can add it.
When an older model becomes too expensive or unreliable, the platform can reduce its traffic.
The developer’s application remains unchanged.
The model becomes replaceable.
The gateway becomes persistent.
3. Why Would Stripe Buy OpenRouter?
Stripe is already one of the most important financial infrastructures in the internet economy.
It supports services such as:
subscriptions;
API usage billing;
credit purchases;
enterprise invoicing;
international payments;
supplier payouts.
Many major AI companies use Stripe or similar infrastructure to monetize subscriptions and consumption-based products.
Stripe already has a powerful position in the question:
How should AI usage be billed and paid for?
Acquiring OpenRouter would allow it to move one step earlier in the chain:
Which model should the user pay for in the first place?
The combined system could connect:
Model discovery;
Request routing;
Usage measurement;
Customer billing;
Payment processing;
Supplier settlement.
Previously, Stripe helped AI companies collect money after a model had already been selected.
With a routing platform, Stripe could also participate in deciding which model receives the request, how many tokens are consumed, and how revenue is distributed.
This is no longer only payment infrastructure.
It begins to resemble a financial operating system for AI consumption.
4. Stripe Is Connecting the Entire Usage-Based Billing Chain
The potential OpenRouter acquisition is more meaningful when viewed alongside Stripe’s earlier expansion into metered billing.
In December 2025, Stripe reportedly spent approximately $1 billion acquiring Metronome, a platform specializing in usage-based billing for API and AI companies.
Metronome helps businesses calculate charges according to:
requests;
tokens;
compute usage;
storage;
feature consumption;
subscription tiers.
OpenRouter, by contrast, determines where AI requests are sent.
Put the two pieces together:
OpenRouter selects the model;
Metronome measures consumption;
Stripe charges the customer;
Stripe can distribute revenue to suppliers.
The strategic direction becomes clear:
Model selection, metering, billing, payment, and settlement can be connected inside one commercial system.
As AI agents begin conducting more online transactions autonomously, this position may become even more important.
An agent may need to decide:
which model should complete a task;
how much intelligence the task requires;
how many tokens it can afford;
which provider offers the best route;
whether the result justifies the price.
Stripe does not merely want to help humans pay AI companies.
It may be preparing to help AI systems make and settle purchasing decisions themselves.
5. The Real Product Is the Decision Layer
OpenRouter’s value is often misunderstood because it does not own the underlying model intelligence.
But the platform controls a different and increasingly important asset:
The decision layer between AI demand and AI supply.
That layer determines:
which model receives the request;
whether the request should be upgraded or downgraded;
which provider becomes the backup;
how price and performance are compared;
how much traffic each supplier receives.
A model company may spend billions training a flagship system.
But if an aggregation platform controls the developer relationship, it can decide whether that flagship model receives 50% of the traffic—or 5%.
This gives the channel three major forms of power.
Traffic Allocation
Routing logic determines which supplier earns the revenue.
A small adjustment to default routing can redirect billions or trillions of tokens.
Price Discovery
When hundreds of models appear behind one API, price comparison becomes automatic.
Large brand premiums become harder to preserve when a cheaper model can complete the same task reliably.
User Stickiness
Developers become loyal to the interface, billing system, routing logic, and reliability guarantees.
The underlying models can change without requiring the application to be rebuilt.
The user relationship belongs to the gateway.
6. Intelligent Routing Is Weakening the Power of Model Brands
In a direct API relationship, the developer explicitly selects a model.
The model name appears in the code, invoice, documentation, and product positioning.
Under automatic routing, that relationship changes.
A platform may assign:
simple extraction to a lightweight model;
routine coding to a balanced model;
complex reasoning to GPT-5.6 Sol or Claude Opus 5;
failed requests to an alternative provider.
The developer may no longer know—or care—which model completed every intermediate step.
The only questions that matter are:
Did the task finish?
Was the result accurate?
Was the response fast enough?
Was the total cost acceptable?
As soon as the gateway can provide stable answers to these questions, the model brand becomes less visible.
The platform becomes the product.
Models become execution components inside the product.
This creates an uncomfortable possibility for model providers:
The better the routing layer becomes, the easier it is to replace one model with another.
The premium migrates upward—from the model to the distribution and orchestration platform.
7. Unexhub’s Alternative Route: From Aggregation to Trading
Stripe and OpenRouter represent one route toward controlling the AI distribution layer:
A payment company expands upward into model routing.
Unexhub represents another:
A compute and token trading platform expands upward into routing, pricing, settlement, and execution.
Unexhub’s goal is not limited to collecting APIs into one catalog.
It aims to create a market in which model and computing-power suppliers can compete for developer and enterprise demand.
The proposed structure includes:
multi-model aggregation;
standardized supplier access;
API bidding;
dynamic price discovery;
millisecond-level routing;
pay-per-token consumption;
real-time or near-real-time settlement;
intelligent task distribution.
In a traditional aggregation platform, the developer chooses from a fixed price list.
In a trading-oriented platform, prices can respond to current supply and demand.
When demand is high and capacity is scarce:
prices may rise;
suppliers earn stronger returns;
additional resources are encouraged to enter the market.
When demand is lower and capacity is abundant:
prices may fall;
developers receive more affordable access;
suppliers can monetize idle capacity.
This is the difference between aggregating products and creating a market.
8. Real-Time Settlement Solves a Problem Routing Alone Cannot Solve
Routing can determine where a request goes.
But a functioning market also needs an efficient way to move money back to suppliers.
Under traditional platform arrangements, small and medium-sized suppliers may wait one week, several weeks, or even a month to receive revenue.
During that period, they must continue paying for:
GPUs;
electricity;
servers;
bandwidth;
data-center resources;
engineering and maintenance.
Long settlement periods create financing pressure.
A supplier may have available capacity and customer demand, yet still be unable to expand because revenue returns too slowly.
Unexhub aims to support real-time or near-real-time clearing:
Once verified usage is recorded, the corresponding supplier revenue can be calculated and returned more quickly.
The value is not simply that suppliers get paid sooner.
Faster settlement can also:
improve cash flow;
reduce financing needs;
support capacity expansion;
increase willingness to accept traffic;
improve platform supply.
Liquidity becomes part of market efficiency.
9. Early Zero-Commission Access Helps Expand the Supplier Network
For qualifying early suppliers, Unexhub plans to offer platform commission rates as low as 0% during the applicable incentive period.
That means eligible suppliers may retain the full distributable revenue generated by their services.
For compute and model providers operating in a high-depreciation industry, this can be more meaningful than an ordinary promotional discount.
A difference of several commission points may affect whether a GPU cluster operates profitably.
The incentive can help suppliers:
recover infrastructure investment faster;
test market demand with lower commercial risk;
monetize unused capacity;
reach enterprise customers;
reduce sales and distribution expenses;
reinvest more revenue in service quality.
The purpose of the incentive is not merely to make the platform appear cheaper.
It is to attract a broader, more diverse supply network.
The more suppliers the platform has, the stronger its price discovery and routing become.
10. Developers Gain Pay-Per-Token Access and Market-Based Prices
Unexhub is designed to connect multiple levels of AI supply, including major model ecosystems such as:
Gemini;
ChatGPT;
Claude;
DeepSeek;
GLM;
other mainstream model services.
It can also include capacity and specialized models from:
small and medium-sized AI companies;
independent development teams;
vertical-model providers;
regional compute operators;
specialized suppliers.
Developers can pay according to actual token usage instead of maintaining separate accounts and prepaid balances across several vendors.
Under eligible routes and market conditions, selected model calls may be available at prices as low as approximately 30% of prevailing public market rates.
The benefit is not limited to the visible token price.
A unified platform can also reduce:
multi-provider integration costs;
credential-management overhead;
invoice reconciliation;
manual model comparison;
failover development;
routing maintenance;
provider monitoring.
Millisecond-level routing can combine price, performance, latency, and availability to identify the lowest-cost route capable of completing the task reliably.
That is more valuable than simply selecting the cheapest model in isolation.
11. The “App Store Moment” for Model Providers
The reported Stripe × OpenRouter deal is important because AI model providers are approaching the same structural moment that mobile developers experienced with app stores.
Before app stores, software companies had to build or negotiate their own distribution channels.
They needed:
payment systems;
installation mechanisms;
customer acquisition;
update infrastructure;
licensing;
regional distribution.
Apple’s App Store centralized these functions.
Developers could focus more heavily on creating applications, while the platform handled discovery, distribution, payments, and updates.
AI model providers are now moving toward a similar structure.
Instead of independently building:
global developer channels;
API billing;
contract systems;
routing;
failover;
enterprise settlement;
demand generation;
they can list models on aggregation and trading platforms.
The platform brings developers and enterprise customers.
The model provider supplies capability.
The exchange handles traffic, pricing, billing, and distribution.
This reduces the barrier for smaller and specialized model suppliers.
A vertical model does not need to become a global brand before finding users.
It needs to perform well enough to earn traffic inside the marketplace.
12. Listing May Become the Default Choice
OpenRouter reportedly aggregates more than 400 models.
Other aggregation platforms have also connected hundreds of model services.
Unexhub is building a deeper trading infrastructure around model supply, token consumption, pricing, routing, and settlement.
As these platforms accumulate more developers and enterprise users, model providers may face a new reality:
Either list on the marketplace, or risk becoming invisible to demand.
This does not necessarily weaken model companies.
It can expand their accessible market.
A smaller provider can instantly reach users it could never have acquired alone.
But the trade-off is clear:
the provider gains distribution;
the platform gains channel power.
This is exactly what happened in earlier digital markets.
Application developers gained access to billions of smartphone users, but app stores gained control over discovery, ranking, payments, and fees.
AI model distribution is moving toward a similar balance of power.
13. From Model Marketplace to Intelligent Utility Exchange
The long-term opportunity is larger than listing and routing individual models.
A platform can eventually combine multiple models and tools into complete outcomes.
For example:
one model decomposes the task;
another reads long documents;
a low-cost model performs batch extraction;
a coding model builds the application;
a flagship model reviews the final result.
The customer no longer needs to buy direct access to one model.
The customer purchases an outcome:
a completed research report;
a migrated codebase;
an automated financial workflow;
a customer-service system;
a deployable application.
At that stage, the commercial product changes from:
Tokens from a named model.
To:
A completed task delivered under defined cost, quality, and SLA conditions.
The marketplace becomes an intelligent utility exchange.
The platform does not merely distribute models.
It organizes them into productive capacity.
14. Who Controls the Future AI Channel?
Stripe/OpenRouter and Unexhub begin from different positions.
Stripe begins with payments and usage-based billing.
OpenRouter begins with API aggregation and routing.
Unexhub begins with token distribution, compute trading, dynamic pricing, and supplier settlement.
Yet all three are moving toward the same strategic layer:
The power to determine which resource serves which task, at what price, and under which commercial terms.
That layer controls:
traffic distribution;
model revenue;
developer costs;
supplier margins;
pricing signals;
user relationships;
execution quality.
The company with the strongest model still matters.
But the company controlling model selection may influence the entire market around it.
Conclusion: Model Providers Build the Engines, but Platforms Are Building the Highway
Tokens are becoming a basic unit of AI production and consumption.
The platform controlling how tokens circulate can influence which models receive demand, how prices are discovered, and where revenue flows.
Stripe’s reported $10 billion interest in OpenRouter is therefore not simply a bet on an API reseller.
It is a bid to occupy the position between every AI request and its destination.
OpenRouter offers routing.
Metronome offers usage measurement.
Stripe offers payment and settlement.
Together, they could form a closed commercial loop around AI consumption.
Unexhub represents a parallel evolution: transforming model access into a more open trading system through dynamic bidding, real-time settlement, early zero-commission incentives, millisecond routing, and pay-per-token consumption.
The reported acquisition may mark the beginning of the AI industry’s “App Store moment.”
Model providers will continue building the engines.
But distribution platforms are building the highway, controlling the exits, measuring the traffic, and collecting the tolls.
The next AI infrastructure winners may not be determined only by benchmark scores or parameter counts.
They may be determined by a simpler question:
Who controls the path between AI demand and AI supply?
That is the deeper story behind the restructuring of compute exchanges—and why Stripe × OpenRouter may be only the beginning.
中文版
预判:算力交易中心正在重构——模型供应商的“App Store 时刻”即将到来,Stripe × OpenRouter 只是开始
2026 年 7 月 24 日,据《华尔街日报》报道,支付巨头 Stripe 正与 AI 模型聚合平台 OpenRouter 洽谈收购事宜,潜在交易估值接近 100 亿美元。
仅仅两个多月前,OpenRouter 的估值据称还只有约 13 亿美元。如果交易最终按照报道中的价格推进,其估值将在一个季度内增长近 7 倍。
一个不训练前沿模型、不拥有大型 GPU 集群,甚至经常被外界称为“模型转售商”的中间层,为什么能够值接近 100 亿美元?
答案隐藏在 AI 产业链正在发生的一次结构性变化中:
渠道价值正在快速接近生产价值,未来甚至可能反过来超越生产者。
生成式 AI 爆发的第一个阶段,市场最关注的是谁能训练出能力最强的模型。
进入下一个阶段,竞争焦点则会逐渐转向:
开发者通过什么入口发现模型;
请求如何被自动路由;
模型价格如何被比较;
Token 用量如何计费;
供应商如何获得结算;
模型如何被封装成完整服务。
AI 基础设施正在接近自己的“App Store 时刻”。
模型供应商继续制造引擎,而第三方平台开始建设道路、收费站、交通规则和支付系统。
Stripe × OpenRouter 可能只是一个开始。
一、OpenRouter 是什么?一个入口连接 400 多款模型
OpenRouter 的核心产品非常容易理解:
将全球 400 多款 AI 模型的 API 整合进一个标准化接口。
如果没有聚合平台,开发者可能需要分别完成:
注册 OpenAI、Anthropic、Google、DeepSeek、Kimi 等厂商账户;
单独谈判商业条款;
管理多组 API Key;
适配不同请求格式;
监控各家的速率限制;
核对分散账单;
为不同供应商开发备用线路;
跟踪模型升级与下线。
通过 OpenRouter,很多工作可以被合并到一套接入中。
开发者只需完成一次集成,就能访问包括旗舰、均衡、轻量、多模态、推理和编程模型在内的广泛模型目录。
平台还可以根据以下因素进行智能路由:
任务难度;
价格;
响应延迟;
模型可用性;
上下文窗口;
工具支持;
供应商可靠性。
当某家厂商发生故障、触发限流或性能下降时,请求可以切换到兼容的备用线路。
对于中小开发团队而言,最重要的价值并不只是 API 价格更低。
而是不用独立建设和维护一整套多模型基础设施。
一套 API,可以替代数十套接入。
二、OpenRouter 的增长,揭示了市场正在向哪里移动
根据原文所引数据,OpenRouter 每周处理约 25 万亿 Token,相当于每月约 100 万亿 Token。
其全球用户规模据称已超过 800 万,年化收入达到约 5000 万美元,半年内增长约 5 倍。
这些数据之所以重要,是因为它们反映出开发者正在逐渐离开“单一模型直连”的模式。
过去,团队可能会作出一个长期决定:
我们的整个产品都使用这一款模型。
现在,越来越多开发者采用另一种方式:
先接入统一模型网关,再根据任务成本、速度和能力动态选择模型。
这会改变用户黏性的归属。
开发者不再围绕一家模型厂商建设全部基础设施。
他们开始围绕入口建设。
新旗舰模型发布后,平台可以直接增加。
旧模型价格上涨或稳定性下降后,平台可以减少它的流量。
开发者的应用代码不需要发生明显变化。
模型可以替换。
网关则持续存在。
三、Stripe 为什么要买 OpenRouter?
Stripe 已经是互联网经济中最重要的金融基础设施之一。
它支持的业务包括:
订阅收费;
API 按量计费;
额度购买;
企业账单;
跨境支付;
供应商分账。
大量头部 AI 企业需要 Stripe 或类似基础设施,实现订阅和用量型产品的商业化。
Stripe 已经在“AI 应该怎样收费和支付”这个问题上占据重要位置。
获得 OpenRouter 后,它就可以向交易链的前端继续移动:
用户首先应该为哪款模型付费?
两者结合后,可以串联以下环节:
模型发现;
请求路由;
用量计量;
用户计费;
支付处理;
供应商结算。
过去,Stripe 在模型已经被选定之后,帮助 AI 公司完成收费。
拥有路由平台后,Stripe 还可能参与决定请求进入哪款模型、消耗多少 Token,以及收入如何分配。
这就不再只是支付基础设施。
而是一套面向 AI 消费的金融操作系统。
四、Stripe 正在打通完整的按量计费链路
把 OpenRouter 的潜在收购,与 Stripe 此前对按量计费领域的布局放在一起看,战略会更加清晰。
2025 年 12 月,Stripe 据称以约 10 亿美元收购了按量计费平台 Metronome。
Metronome 可以帮助 API 和 AI 企业,根据以下维度计算费用:
请求数量;
Token 用量;
算力消耗;
存储;
功能使用;
订阅档位。
OpenRouter 则负责决定请求被发送到哪里。
把两块能力拼起来:
OpenRouter 选择模型;
Metronome 统计用量;
Stripe 向客户收费;
Stripe 将收益分配给供应商。
Stripe 的真实意图也因此变得清晰:
模型选择、用量计量、账单生成、支付处理与供应商结算,可以被装进同一套商业系统。
随着 AI 智能体开始自动完成越来越多线上交易,这一位置会变得更加重要。
未来,一个智能体可能需要自主决定:
应该调用哪款模型;
任务需要多强的能力;
最多可以消耗多少 Token;
哪个供应商的线路更合适;
当前结果是否值得相应价格。
Stripe 不只想帮助人类向 AI 公司付款。
它还可能为 AI 系统自主采购与结算提供基础设施。
五、真正的产品,是中间的“决策层”
很多人低估 OpenRouter,是因为它并不拥有底层模型能力。
但它控制着另一项越来越重要的资产:
AI 需求与 AI 供给之间的决策层。
这一层决定:
请求最终进入哪款模型;
是否需要升档或降档;
哪家供应商成为备用线路;
价格与性能怎样比较;
每家模型可以获得多少流量。
模型公司可能投入数十亿美元训练旗舰系统。
但如果聚合平台掌握开发者关系,它就可以决定这款旗舰模型获得 50% 的流量,还是仅获得 5%。
渠道由此获得三种重要权力。
流量分配权
路由逻辑决定哪家供应商获得收入。
一次默认规则调整,就可能转移数十亿甚至数万亿 Token。
价格发现权
当数百款模型进入同一接口后,价格比较会自动发生。
如果更便宜的模型能够稳定完成同一任务,大额品牌溢价就越来越难维持。
用户黏性
开发者会逐渐依赖统一接口、账单系统、路由逻辑和稳定性保障。
底层模型即使变化,应用也不必重新开发。
用户关系留在平台。
模型成为平台背后可以替换的供应方。
六、智能路由正在削弱模型品牌的控制力
在直接调用模式中,开发者会明确选择模型。
模型名称会出现在代码、账单、文档和产品宣传中。
自动路由普及后,这种关系开始改变。
平台可以把:
简单抽取交给轻量模型;
日常编程交给均衡模型;
复杂推理交给 GPT-5.6 Sol 或 Claude Opus 5;
失败请求切换到另一家供应商。
开发者可能不再知道,也不再关心每一个中间步骤究竟由哪款模型完成。
真正重要的是:
任务是否完成;
结果是否准确;
响应是否足够快;
总成本是否合理。
一旦网关能够稳定满足这些需求,平台就会成为用户真正信任的产品。
模型则成为平台里的多个执行引擎之一。
这给模型供应商带来一个现实问题:
路由层越成熟,底层模型就越容易被替换。
品牌溢价会逐渐从模型层转移到分发与编排平台。
七、Unexhub 的另一条路线:从聚合走向交易
Stripe 与 OpenRouter 代表了一条掌握 AI 分发层的路径:
支付公司向上进入模型路由层。
Unexhub 代表另一条路径:
算力与 Token 交易平台向上延伸到路由、定价、结算和智能执行。
Unexhub 的目标并不只是把 API 放进一个目录,而是建立一个模型与算力供应商围绕开发者和企业需求展开竞争的市场。
其规划体系包括:
多模型聚合;
标准化供应商接入;
API 动态竞价;
实时价格发现;
毫秒级路由;
按 Token 实际用量付费;
实时或近实时清结算;
智能任务分配。
传统聚合平台展示的是厂商固定价目表。
交易型平台则希望让价格根据实时供需变化。
当需求旺盛、可用容量不足时:
价格可以上升;
供应商获得更高回报;
更多资源受到激励进入市场。
当需求较低、供给充足时:
价格可以下降;
开发者获得更低成本;
供应商也能变现闲置资源。
这就是“聚合产品”与“建立市场”的区别。
八、实时结算,解决路由无法单独解决的问题
路由可以决定请求去哪里。
但一个真正运行的市场,还必须让资金高效回到供应商手中。
在传统平台中,中小供应商可能需要等待一周、数周,甚至一个月才能获得收入。
与此同时,他们仍需持续支付:
GPU 成本;
电费;
服务器费用;
带宽;
数据中心资源;
工程与维护成本。
漫长账期会制造融资压力。
供应商即使拥有空闲容量和客户需求,也可能因为资金回流过慢,无法继续扩大供给。
Unexhub 的目标是推进实时或近实时清结算:
有效调用完成计量后,对应供应商收益可以更快计算与回流。
真正的价值不只是让供应商更早收钱。
快速结算还可以:
改善现金流;
降低融资需求;
支持容量扩张;
提高承接流量的意愿;
丰富平台整体供给。
流动性本身,就是市场效率的一部分。
九、早期 0 佣金,帮助平台扩大供应网络
针对符合条件的早期供应商,Unexhub 计划在适用激励期内提供最低 0% 平台佣金。
这意味着符合条件的供应商可以保留服务产生的全部可分配收入。
对于高折旧的算力和模型服务行业来说,这项权益比普通促销折扣更有实际意义。
几个百分点的佣金差异,可能直接影响一组 GPU 能否盈利。
早期 0 佣金可以帮助供应商:
更快回收基础设施投资;
以更低商业风险测试市场需求;
变现闲置资源;
触达企业客户;
降低销售与分发成本;
将更多收入重新投入服务质量。
这项激励并不只是为了让平台价格看起来更便宜。
更重要的目的,是吸引更广泛、更多样的供给进入市场。
供应商越多,平台的价格发现与路由能力就越强。
十、开发者获得按 Token 实付与市场化价格
Unexhub 计划连接不同层级的 AI 供应,包括主流模型生态,例如:
Gemini;
ChatGPT;
Claude;
DeepSeek;
GLM;
其他主流模型服务。
平台还可以接入:
中小 AI 公司;
独立开发团队;
垂直模型供应商;
区域算力运营方;
专业服务提供商。
开发者可以根据实际 Token 用量付费,不必同时维护多家厂商账户和预充值余额。
在符合条件的线路和市场环境下,部分模型调用价格最低可至市场公开价约 3 折。
实际价值并不只体现在可见的 Token 单价。
统一平台还可以降低:
多供应商接入成本;
密钥管理压力;
分散账单核对;
模型人工比较;
故障切换开发;
路由维护;
供应商监控。
毫秒级路由可以综合价格、能力、延迟和可用性,为任务寻找能够稳定完成工作的最低成本路线。
这比孤立地选择最便宜的模型更有价值。
十一、模型供应商的“App Store 时刻”
Stripe × OpenRouter 的潜在交易之所以重要,是因为 AI 模型供应商正在接近移动应用开发者曾经历过的结构性时刻。
在应用商店出现之前,软件公司需要自行建设或谈判分发渠道。
它们要解决:
支付系统;
软件安装;
用户获取;
更新机制;
授权;
地区分发。
App Store 将这些功能集中起来。
开发者可以把更多精力放在产品本身,而平台负责发现、分发、支付和更新。
AI 模型供应商如今也在走向类似结构。
它们不再需要独立建设:
全球开发者渠道;
API 计费系统;
合同体系;
路由;
故障切换;
企业结算;
市场获客。
模型可以直接进入聚合或交易平台。
平台带来开发者与企业客户。
模型供应商提供能力。
交易中心负责流量、定价、计费和分发。
这会降低中小模型与垂直模型供应商进入市场的门槛。
一款垂直模型不必先成为全球知名品牌,才能获得用户。
它只需要表现足够好,就有机会在市场里获得流量。
十二、“上架”可能成为模型供应商的默认选择
OpenRouter 已经聚合了 400 多款模型。
其他模型聚合平台也已经接入数百款服务。
Unexhub 则进一步围绕模型供给、Token 消费、竞价、路由与结算建设交易基础设施。
当这些平台汇聚足够多的开发者和企业用户后,模型供应商可能面对一个新现实:
要么上架进入市场,要么逐渐失去被用户发现的机会。
这并不必然削弱模型公司。
平台可以为它们打开更大的市场。
一家中小供应商可以迅速触达过去无法独立获取的客户。
但交换条件同样明显:
供应商获得分发;
平台获得渠道权力。
此前的数字市场早已出现过同样的变化。
应用开发者获得了触达数十亿手机用户的机会,App Store 则获得发现、排名、支付和抽佣权。
AI 模型分发正在走向类似的权力平衡。
十三、从模型商店,走向智能效用交易所
长期机会并不只是将单个模型上架并进行路由。
平台最终可以把多款模型和工具组合成完整成果。
例如:
一款模型负责拆解任务;
一款模型读取长文档;
低成本模型完成批量抽取;
编程模型开发应用;
旗舰模型最终审核结果。
客户不再需要购买某款模型的直接访问权。
客户购买的是结果:
一份完整研究报告;
一个迁移完成的代码库;
一套自动化财务流程;
一个客服系统;
一款可以部署的应用。
商业产品会从:
某款模型提供的 Token。
转变为:
在约定成本、质量和 SLA 条件下完成的任务。
模型市场也将进一步演变成智能效用交易所。
平台不再只是分发模型。
它开始把模型组织成生产力。
十四、谁将控制未来的 AI 渠道?
Stripe/OpenRouter 与 Unexhub 的起点不同。
Stripe 从支付与按量计费出发。
OpenRouter 从 API 聚合和路由出发。
Unexhub 从 Token 分发、算力交易、动态定价和供应商结算出发。
但它们都在向同一个战略层靠近:
决定哪种资源应该服务哪项任务、价格是多少,以及按照什么商业条件完成。
这一层可以影响:
流量分配;
模型收入;
开发者成本;
供应商利润;
市场价格;
用户关系;
执行质量。
拥有最强模型依然重要。
但控制模型选择的平台,同样可以影响模型周围的整个市场。
结语:模型供应商制造引擎,平台正在修建高速公路
Token 正在成为 AI 生产与消费的基础单位。
控制 Token 如何流通的平台,可以影响哪款模型获得需求、价格如何形成,以及收入流向哪里。
因此,Stripe 对 OpenRouter 接近 100 亿美元的潜在兴趣,并不只是对一家 API 转售平台的投资。
它是在争夺站在每一次 AI 请求与最终目的地之间的位置。
OpenRouter 提供路由。
Metronome 提供用量计量。
Stripe 提供支付与结算。
三者结合后,可以形成围绕 AI 消费的完整商业闭环。
Unexhub 则代表另一种同步发生的演化:通过动态竞价、实时结算、早期 0 佣金激励、毫秒级路由和按 Token 实付,把模型调用进一步变成一个更加开放的交易市场。
这笔潜在收购,可能标志着 AI 行业“App Store 时刻”的开始。
模型供应商仍然负责制造引擎。
但分发平台正在修建高速公路,控制出口、统计流量,并设置收费方式。
下一阶段 AI 基础设施的赢家,可能不只由模型跑分和参数规模决定。
而会由一个更简单的问题决定:
谁控制着 AI 需求与 AI 供给之间的通道?
这才是算力交易中心重构背后的深层逻辑,也是 Stripe × OpenRouter 可能只是开始的真正原因。