The 30% Moat Is Being Re-Measured by AI Agents
For 17 years, 30% has been one of the most familiar numbers in the app economy.
In 2008, Apple established the App Store's famous 30% commission. Google followed, and the industry largely accepted the model. Developers learned to calculate their margins, pricing, and growth around that number.
In 2026, however, the 30% line is beginning to move.
Not primarily because of antitrust lawsuits or developer protests, but because AI Agents are redefining what an "app store" actually is.
Apple has recently lowered its standard commission in China to 25%, while its small-business program can reduce the rate to 12%. Google cut its in-app purchase commission from 30% to 20% in March, with developers participating in its app-experience program potentially paying 15%.
Both are moving downward—but neither has fundamentally changed the game.
The more disruptive moves are coming from Microsoft and Anthropic.
Microsoft's 85/15: A New Pricing Anchor
At Microsoft Build 2026, Microsoft introduced the Windows Agent Store, with developers receiving 85% of transaction revenue while the platform takes 15%.
Compared with the 70/30 model that Apple and Google have normalized for nearly two decades, Microsoft's 85/15 structure is more than a minor adjustment. It represents a different economic proposition for developers.
Microsoft does not need to rely primarily on commissions. Its broader strategy is tied to its massive Microsoft 365 ecosystem and Agent 365 subscriptions.
The real advantage is distribution.
Agents can potentially be integrated directly into interfaces users already open every day—Windows, Copilot, Teams, Outlook, and other Microsoft environments.
Developers therefore don't necessarily need to build another destination users must actively search for.
They can build capabilities that appear where users already work.
That changes the distribution equation from:
Users find applications
to:
Agents find users.
And that may ultimately be much more important than the difference between a 15% and 20% commission.
The More Aggressive Experiment: Anthropic's 0% Commission
If Microsoft's 15% already looks aggressive, Anthropic has gone one step further.
In March 2026, Anthropic introduced the Claude Marketplace, explicitly choosing not to charge transaction commissions.
The idea is straightforward: enterprise customers can redirect portions of their existing annual commitments toward third-party tools and services available through the marketplace.
The model resembles the enterprise marketplace strategies used by major cloud platforms, but with one major difference:
Anthropic removes the platform transaction fee.
That is not necessarily charity.
It is ecosystem strategy.
By giving up transaction revenue, Anthropic can make its ecosystem more attractive to developers and enterprise buyers. When developers don't have to sacrifice part of every transaction to distribution fees, their decision increasingly comes down to three questions:
Which platform provides the best development tools?
Which platform has the most valuable users?
Which platform makes it easiest to build and distribute Agents?
The commission itself becomes only one part of the equation.
Developer Economics Have Changed
During the mobile-app era, platform selection was relatively straightforward.
iOS offered strong monetization but higher platform fees. Android offered enormous reach but greater fragmentation.
AI Agents introduce at least three different dimensions.
1. Commission Rate
The headline numbers are becoming increasingly diverse:
Apple: 25% standard rate in China, with 12% for eligible small developers
Google: 20%, with certain programs reaching 15%
Microsoft: 15%
Anthropic: 0%
Token marketplaces: potentially dynamic pricing rather than a fixed platform commission
The difference between 0% and 25% can translate directly into a meaningful margin difference.
But commission alone doesn't determine profitability.
2. Distribution Efficiency
A low commission is meaningless if the platform cannot deliver users.
Apple has a highly monetized consumer ecosystem.
Google has enormous global reach.
Microsoft has access to a massive enterprise software installed base.
Anthropic has an enterprise customer ecosystem and increasingly sophisticated AI tooling.
This leads to a crucial principle:
The cheapest distribution channel is not necessarily the most profitable one.
The real calculation is net value after distribution costs.
3. Lock-In Cost
The form of lock-in is changing, but the underlying issue remains.
Traditional mobile applications are deeply tied to their operating systems.
Microsoft's open-source Agent Framework can potentially reduce some migration barriers.
Anthropic's marketplace eliminates transaction fees, but applications may still become deeply integrated with Claude-specific capabilities.
The question is no longer simply:
"How much does the platform charge?"
It is:
"How expensive would it be to leave?"
Unexhub's Third Path: Turning "Channel Fees" into "Market Pricing"
There is another way to look at the problem.
What exactly is a platform commission?
At its core, it is a fee charged for access to distribution.
Unexhub approaches the problem from a different layer.
Rather than building another traditional application store, Unexhub positions itself as a Token aggregation and distribution computing-power trading center.
The underlying idea is to turn Token consumption into a market-driven resource.
API prices can respond dynamically to supply and demand. Developers may access lower prices during periods of excess capacity, while suppliers can capture better returns when demand increases.
For suppliers, the model also emphasizes:
Early-stage 0% commission
Real-time settlement
One-click channel access
Dynamic API bidding
Millisecond-level routing
This creates a fundamentally different economic structure from the App Store model.
Apple and Microsoft are competing over:
Who gets to distribute the Agent?
Unexhub is addressing another question:
What does it cost to run the Agent?
That distinction matters.
If Tokens become increasingly market-priced and tradable, computational costs no longer have to be treated as a fixed overhead.
Developers can instead optimize around:
model selection,
supply availability,
time of usage,
Token pricing,
and routing efficiency.
In other words, computing cost becomes an optimization variable rather than a fixed bill.
The Value Proposition for Developers and Suppliers
This is where the model becomes particularly interesting.
Unexhub brings together AI suppliers across different scales, including major model ecosystems such as Gemini, ChatGPT, Claude, DeepSeek, and GLM, alongside quality models provided by individual developers and small-to-medium-sized companies.
For developers and enterprise users, the value is not simply "more models."
It is the ability to compare and route across multiple sources through one access layer, rather than maintaining separate accounts, API keys, billing systems, and integrations for every provider.
Under Unexhub's current positioning, developers can pay according to actual Token consumption, with access prices starting as low as 30% of market pricing, depending on available supply and trading conditions.
That means the potential savings become especially meaningful for high-frequency applications and Agent workloads.
For suppliers, the economics are equally important: one integration can open an additional distribution channel, while real-time settlement helps shorten the gap between providing compute resources and receiving revenue.
The result is a two-sided marketplace:
Developers pursue lower Token costs. Suppliers pursue higher utilization and faster settlement. The market connects the two.
That is fundamentally different from simply adding another 30% commission layer.
A New Multi-Polar Distribution Landscape
The AI Agent economy in 2026 is moving away from the old "Apple vs. Google" structure.
The emerging landscape looks more like:
Apple → 25%
Google → 20%
Microsoft → 15%
Anthropic → 0%
Token marketplaces → dynamic market pricing
Each represents a different economic philosophy.
Apple and Google: Defending the Existing Ecosystem
Their strategy is gradual optimization.
Lower commissions, stronger developer incentives, and continued control over massive consumer ecosystems.
Microsoft: Attacking Through Distribution
Microsoft is using lower commissions and enterprise distribution to attract developers.
The key weapon isn't simply 15%.
It is access to users who are already inside the Microsoft ecosystem.
Anthropic: Buying Ecosystem Loyalty
Anthropic's 0% commission model removes one of the most obvious barriers between developers and enterprise customers.
The bet is simple:
Give up transaction fees today to strengthen the ecosystem tomorrow.
Unexhub: Rebuilding the Cost Layer
Unexhub is playing a different game.
Instead of competing primarily over where Agents are discovered, it focuses on how their underlying AI computation is sourced and priced.
Tokens become the economic unit.
Suppliers compete for demand.
Developers compare prices.
Routing responds to market conditions.
And computing resources become increasingly liquid.
So Where Should Agent Developers Go?
The answer depends on the business model.
If your target is consumers
Apple and Google remain difficult to ignore.
Their commissions are relatively high, but their user bases and established payment ecosystems remain powerful distribution advantages.
If your target is enterprise users
Microsoft's Agent ecosystem and Anthropic's Claude Marketplace deserve serious consideration.
The difference between 15% and 0% can materially affect margins, particularly for high-value transactions.
If you want to minimize platform dependence
A Token marketplace represents another option.
Instead of paying primarily for access to a distribution channel, developers can optimize the underlying AI consumption cost itself.
That is a fundamentally different approach.
💡 The most important change may not be that the 30% commission is falling.
The bigger change is that developers are gaining more ways to avoid treating platform fees as an unavoidable cost of doing business.
They can negotiate distribution.
They can choose different ecosystems.
They can run open-source models.
And increasingly, they can optimize the Token layer underneath their applications.
The 30% Line Is No Longer a Fixed Boundary
Seventeen years ago, Apple drew a line at 30%.
In 2026, that line is being challenged from several directions at once.
Microsoft is pulling it down.
Anthropic is pushing it toward zero.
Open-source models are creating alternatives to platform dependency.
Token marketplaces are attempting to bypass the traditional commission structure altogether.
The important story is therefore not simply that 30% is becoming 25%, 15%, or 0%.
The deeper story is that the economic structure of AI applications is being rewritten.
For developers, the question is no longer simply:
"Which app store should I choose?"
It is becoming:
"Who controls my users, who controls my models, who controls my compute, and who sets the price of every Token I consume?"
That is where the next AI platform battle will be fought.
中文版|中文内容
30% 的护城河,正在被 AI Agent 重新丈量
过去 17 年,30% 一直是应用经济中最熟悉的数字之一。
2008 年,苹果为 App Store 确立了著名的 30% 抽成模式。随后谷歌跟进,整个行业逐渐接受了这一规则。开发者开始围绕这个数字计算利润、制定价格,并寻找生存空间。
但到了 2026 年,30% 这条线开始松动。
这并不主要是因为反垄断诉讼,也不只是开发者抗议,而是因为一个更根本的变化正在发生:
AI Agent 正在重新定义什么叫"应用商店"。
苹果近期将中国区标准佣金下调至 25%,符合条件的小型开发者优惠档可以进一步降至 12%。谷歌则在 3 月将应用内购买的佣金从 30% 下调至 20%,参与相关计划的开发者甚至可以进一步降至 15%。
两家巨头都在降价。
但真正的变化,可能还不在这里。
更激进的玩家正在出现:微软与 Anthropic。
微软 85/15:一个新的价格锚点
在 2026 年 Microsoft Build 大会上,微软推出 Windows Agent Store,开发者获得 85% 的交易收入,平台抽取 15%。
与苹果、谷歌沿用了近 20 年的 70/30 模式相比,微软的 85/15 并不是简单的小幅优惠,而是一种全新的开发者经济模型。
微软并不需要单纯依靠应用抽成赚钱。
它真正依赖的是庞大的 Microsoft 365 生态以及 Agent 相关订阅服务。
更重要的是——分发能力。
Agent 可以直接进入用户每天使用的 Windows、Copilot、Teams、Outlook 等工作环境。
开发者不一定还需要开发一个用户必须主动搜索、下载和打开的"应用"。
他们可以直接把能力植入用户已经在使用的工作界面。
这意味着 AI Agent 的分发逻辑正在发生变化:
过去:用户寻找应用。
变成:
未来:Agent 主动寻找用户。
这可能比 15% 和 20% 之间的差异更加重要。
更激进的实验:Anthropic 的 0% 抽佣
如果说微软的 15% 已经足够激进,那么 Anthropic 直接把数字推向了 0%。
2026 年 3 月,Anthropic 推出 Claude Marketplace,明确采用不收取交易佣金的模式。
其核心思路是:企业客户可以将原本用于 Anthropic 服务的年度承诺预算,用于购买 Marketplace 中的第三方工具和服务。
这种模式与 AWS、Azure 等企业云 Marketplace 有一定相似之处,但最大的区别在于:
Anthropic 直接取消了交易佣金。
这并不意味着它在"做慈善"。
实际上,这是一种生态战略。
通过放弃交易抽成,Anthropic 可以降低开发者进入生态的成本,同时增强企业用户与第三方工具之间的连接。
当开发者不再需要为每一笔交易支付平台费用时,他们最终真正需要考虑的就变成了三个问题:
哪个平台的开发工具最好?
哪个平台的用户最有价值?
哪个平台最容易让 Agent 被发现和使用?
佣金比例反而不再是唯一指标。
开发者的选择逻辑已经变了
移动互联网时代,开发者选择平台的逻辑相对简单。
iOS 用户付费能力强,但平台费用较高;Android 用户规模庞大,但设备更加碎片化。
AI Agent 时代,这套逻辑正在被拆成至少三个维度。
1. 抽成比例
现在市场上的代表性数字已经越来越多:
苹果:中国区标准档 25%,小型开发者可至 12%
谷歌:20%,部分计划可至 15%
微软:15%
Anthropic:0%
Token 交易市场:更多采用动态市场定价,而不是固定的平台抽成
从 0% 到 25%,看起来只是一个数字变化,但对于高交易额业务而言,它直接关系到利润空间。
不过,佣金并不是决定盈利能力的唯一因素。
2. 分发效率
一个平台即使只抽 0%,如果没有用户,也没有意义。
苹果拥有强大的消费级生态。
谷歌拥有庞大的全球用户规模。
微软拥有巨大的企业软件生态。
Anthropic 则拥有企业客户与 AI 开发工具生态。
所以真正应该计算的不是:
"平台抽多少?"
而是:
"扣除分发成本之后,我最终获得多少价值?"
3. 锁定成本
平台锁定的形式正在变化,但问题本身并没有消失。
传统移动应用高度依赖操作系统。
微软的开源 Agent Framework 可以在一定程度上降低迁移门槛。
Anthropic 的 Marketplace 虽然不抽佣,但 Agent 一旦深度依赖 Claude 的能力,也可能形成新的技术绑定。
所以今天真正应该问的问题已经不是:
"平台收费多少?"
而是:
"如果有一天我要离开,需要付出多大代价?"
Unexhub 的第三条路径:把"渠道费"变成"市场定价"
那么,平台佣金到底是什么?
从本质上看,它就是平台针对用户触达与分发能力收取的一笔费用。
而 Unexhub 选择了完全不同的切入点。
它不是再建设一个传统意义上的应用商店,而是定位为:
Token 聚合与分发的算力交易中心。
核心逻辑,是让 Token 消耗本身逐渐成为一个可以市场化定价的资源。
API 价格可以随着供需关系动态变化。
开发者在算力供给充足、需求较低的时段,有机会获得更低价格;供应商则可以在需求旺盛时提高资源利用率和收益。
对于供应商而言,Unexhub 强调的核心权益包括:
早期入驻 0 佣金
实时结算
一键接入渠道
API 动态竞价
毫秒级路由
这与传统 App Store 的商业逻辑完全不同。
苹果和微软争夺的是:
"Agent 应该分发给谁?"
而 Unexhub 更关注:
"Agent 运行的算力到底应该以什么价格获得?"
这两个问题并不冲突,但处于不同层级。
如果 Token 最终成为一种市场化、可交易的算力资源,那么开发者就不必再把算力成本完全视为固定支出。
相反,他们可以围绕以下变量进行优化:
模型选择
供应商
Token 价格
使用时间
供需关系
路由效率
也就是说:
算力成本正在从固定账单,变成可以持续优化的变量。
对开发者和供应商而言,价值到底在哪里?
这也是 Unexhub 模式真正值得关注的地方。
Unexhub 汇聚不同规模的 AI 供应商,包括 Gemini、ChatGPT、Claude、DeepSeek、GLM 等主流模型生态,同时也支持个人开发者以及中小企业提供的优质模型资源。
对于开发者和企业用户来说,价值并不只是"模型更多"。
更重要的是:
通过一个统一入口,对不同供应商、不同模型和不同价格进行比较与路由。
开发者无需分别维护大量账号、API Key、充值体系和接口集成。
按照 Unexhub 当前的市场定位,开发者可以按照实际 Token 使用量付费,在具体供应与交易条件允许的情况下,最低可享受市场价格约 3 折的调用成本。
对于高频调用、批量处理和 AI Agent 工作流而言,这种成本优化空间会更加明显。
而供应商获得的则是另一种价值:
一次接入,即可获得更多分发渠道。
同时,通过实时结算,可以缩短从提供算力资源到获得收入之间的资金周期。
最终形成一个双边市场:
开发者寻找更低的 Token 成本,供应商寻找更高的资源利用率与更快的资金回笼,而市场负责连接双方。
这已经不是简单增加一个 30% 或 15% 的渠道抽成。
而是在尝试重新定义 AI 应用背后的成本结构。
一个新的多极化分发格局正在形成
2026 年的 AI Agent 渠道竞争,正在逐渐摆脱过去单纯的"苹果 vs 谷歌"模式。
新的市场结构可能更接近:
苹果 → 25%
谷歌 → 20%
微软 → 15%
Anthropic → 0%
Token 交易市场 → 动态市场定价
每一种模式背后,都代表不同的商业哲学。
苹果与谷歌:守住既有生态
它们采取的是渐进式优化。
通过降低佣金、增加开发者激励,同时继续利用庞大的消费级生态维持优势。
微软:用分发能力发动进攻
微软真正的武器并不只是 15%。
更重要的是:
它拥有一个已经存在的企业用户生态。
Anthropic:用 0% 换生态忠诚
Anthropic 的逻辑更加直接:
今天放弃交易收入,换取未来的开发者与企业生态。
Unexhub:重新定义成本层
Unexhub 玩的是另一套游戏。
它不是单纯竞争 Agent 应该在哪里被发现,而是关注:
Agent 背后的 AI 算力到底应该从哪里获得,以及应该以什么价格获得。
当 Token 成为经济计量单位之后:
供应商可以竞争需求。
开发者可以比较价格。
路由可以根据市场实时调整。
算力资源也就逐渐具备更强的流动性。
Agent 开发者到底应该选择哪里?
答案取决于你的商业模式。
如果目标用户是 C 端消费者
苹果和谷歌依然很难绕开。
虽然 25% 和 20% 的佣金并不低,但它们成熟的支付体系、用户规模和消费生态依然具有明显优势。
如果目标用户是企业客户
微软的 Agent 生态和 Anthropic 的 Claude Marketplace 值得重点关注。
15% 与 0% 的差距,对于高价值交易而言并不是一个小数字。
如果希望降低平台依赖
Token 交易市场则提供了另一条路径。
开发者不再只是寻找"哪个平台抽成更低",而是直接优化:
底层 AI Token 的实际消耗成本。
这是一种完全不同的思路。
💡 真正重要的变化,也许并不是 30% 正在变成 25%、15% 或 0%。
更重要的是,开发者正在获得越来越多的选择,不必再把平台费用视为 AI 应用商业模式中不可避免的固定成本。
可以选择不同的分发渠道。
可以选择不同的模型供应商。
可以使用开源模型。
也可以进一步优化应用底层的 Token 成本。
未来真正需要优化的,不只是应用售价,而是整个 AI 应用的成本结构。
30% 已经不再是一条固定的边界
17 年前,苹果用 30% 画下了一条线。
2026 年,这条线正在同时受到多个方向的冲击。
微软正在把它拉低。
Anthropic 正在把它推向 0%。
开源模型正在提供平台之外的选择。
Token 交易市场则试图绕开传统的固定渠道抽成。
因此,真正值得关注的已经不是:
30% 变成了 25%、15% 还是 0%。
更深层的变化是:
AI 应用的经济结构正在被重新设计。
对于开发者而言,未来真正的问题也不再只是:
"我应该把 Agent 上架到哪个应用商店?"
而是:
"谁控制我的用户?谁控制我的模型?谁控制我的算力?谁决定我每一次 Token 调用的价格?"
这,可能才是下一轮 AI Agent 平台竞争真正的战场。