AI Interview Wiki
主题综述

AI 时代的护城河 · AI Moat 2026

主题综述

更新日志

主流共识

第一点："data moat" 在大多数场景下不存在。a16z 的 Alex Rampell 把这个常识破除得最直接：

"It's almost like gravity. Gravity actually, like one atom actually exerts gravity on you, but you only really see it at like very, very large scale. … when you've seen 4 billion people and like these people are bad, now you can sell each incremental customer … because you've seen more customers and you can get actually better results. But the challenge is that a lot of these moats only really are evident at mega, mega, mega scale."
「它确实有点像引力。引力就像一个原子一样对你施加引力，但你只有在非常大的规模下才能真正看到它。……当你见过 40 亿人，并且知道这些人很糟糕时，现在你可以向每个增量客户出售服务……因为你见过更多的客户，所以你可以获得更好的结果。但挑战在于，很多这些护城河只有在非常非常大的规模下才会真正显现出来。」
Alex Rampell · Why AI Moats Still Matter

第二点：软件本身在做工作——这把"软件的市场"从 IT 预算扩到劳动预算，但也制造了无限竞争。

"The thing that is fundamentally different about this product cycle is that the software itself can actually do the work."
「这个产品周期最根本的不同是：软件本身真的能完成工作。」
Why AI Moats Still Matter · Why AI Moats Still Matter
"The same force creating infinite competition is also creating trillion dollar opportunities in places nobody's looking."
「制造无限竞争的同一股力量，也在没人看的地方创造万亿美元的机会。」
Why AI Moats Still Matter · Why AI Moats Still Matter

第三点："AI 解决了 bootstrap 问题，没解决 retention 问题"——这条 a16z 自己提出来的判断在多个阵营都被回应。

"A really hard thing about building any startup or software is the bootstrap problem. How do you get the first 100, 200 customers? And AI actually solves that problem. It just solves the bootstrap problem. … But what's also clear is that doesn't solve your retention problem if you're a software company."
「建立任何创业公司或软件真正难的地方是 bootstrap 问题——怎么获得最初的 100、200 个客户？AI 确实解决了那个问题。它只解决 bootstrap 问题。……但同样清楚的是，如果你是软件公司，留存问题它解决不了。」
Martin Casado · The State of AI: Growth, Fragmentation

分歧在哪

阵营 A · "传统软件护城河仍然适用，只是要重新挣得"——a16z / Sequoia 主流派

a16z 的 panel 给的最完整版本：

"The defensibility of a software product resides, in my opinion, from owning the end-to-end workflow, from the context in which that it's applied, becoming the system of record, having a network effect, deeply embedding yourself within your customer."
「我认为软件产品的防御性来自：拥有端到端的工作流、被应用的上下文、成为系统记录、网络效应、深度嵌入客户。」
Why AI Moats Still Matter · Why AI Moats Still Matter

但他们也明说"GPT wrapper" 不是 thing：

"We've kind of come to the opinion that there is no AI. There's like a bunch of subspaces that are totally different that all require their own strategy."
「我们的结论是：不存在 'AI'——而是一堆完全不同的子空间，每个都需要各自的策略。」
Martin Casado · The State of AI

Sequoia 的 David Cahn 押一个 a16z panel 没强调的具体子轴——建造本身就是护城河：当所有人同时抢建数据中心时，"能真的把数据中心建出来"这件事的难度被严重低估（他在姊妹主题 ai-capital-cycle 里那条"硬件去商品化 + 风险传递链"的逐字分析是这条的延伸）。

但他也是 Camp A 内部最公开质疑"垄断式护城河"叙事的人——他认为市场高估了 AI 业务的垄断性：跟大科技时代"垄断在没人注意时悄悄建起"不同，这一轮所有人都盯着同一块地，反而更难形成那种隐秘的垄断。

注意——Cahn 这个论点和 a16z 在 State of AI 里展示的数据存在张力：Sarah Wang 的大意是，头部 frontier lab 的收入爬坡不仅超过了历史上最好的 SaaS 公司的早期爬坡，还开始超过 hyperscaler 的早期爬坡——收入正在以空前速度向少数实验室集中。Camp A 内部并不统一。

阵营 B · "Vertical integration is the only real moat"——SpaceX 派 / 硬件操作派

Scott Nolan (General Matter) 给的是最 SpaceX 化的版本：

"Many different sectors where subcontracting is the norm and you're going to subcontract a subsystem to somebody who subcontracts a component and then that component has different inputs and they subcontract that all the way down. … [Cost-plus gives you] layers of subcontractors, dozens deep, and no ability for anyone to do something really novel."
「很多行业里外包是常态——你把一个子系统外包给某人，他再把一个零件外包出去，那个零件又有不同的输入、又外包下去，一路到底。……（cost-plus 体制下）一层层分包商，深达十几层，谁都没法做出真正新颖的东西。」
Scott Nolan · Scott Nolan - SpaceX, Founders Fund
"I think the thing that's most linked to stagnation is probably being a cost plus industry where there's very little incentive for progress, not much incentive to bring the cost structure down."
「我认为最容易导致停滞的大概是 cost-plus 行业——几乎没动力进步，也没什么动力降本。」
Scott Nolan · Scott Nolan - SpaceX, Founders Fund

Evan Spiegel (Snap) 在硬件方向给的是同结论：

"We have a developer ecosystem, so hundreds of thousands of developers who have built millions of lenses on our developer tools. When they build those experiences, they run on our rendering engine, which is really deeply integrated into our operating system. And they see and experience those AR objects through our glasses using our own optical engine, meaning the piece of glass in the glasses and the little projector, we design and build ourselves."
「我们有一个开发者生态——几十万开发者在我们的开发工具上做了上百万个 lens。这些体验跑在我们自己的渲染引擎上，而渲染引擎又真正深度地集成进我们的操作系统。用户透过我们的眼镜、用我们自己的光学引擎去看到并体验这些 AR 物体——也就是眼镜上的那块玻璃、那个小投影仪，都是我们自己设计和制造的。」
Evan Spiegel · Snap CEO Evan Spiegel is Betting on Smart Glasses

David Cahn 帮 Camp B 提供了一句话框架：

"People were thinking very abstractly, sort of in a bit's perspective about AI, but they should be thinking in an atom's perspective about AI."
「人们一直用 bit 的视角抽象地想 AI——其实应该用 atom 的视角想。」
David Cahn · 20VC: Sequoia's David Cahn

阵营 C · "软件层 above commoditized models 是护城河"——Baseten 立场

Tuhin Srivastava 的论点跟 Camp B 在物理方向上对立——他不押"往下"垂直整合，押"往上"软件层（逐字稿末尾的 "modes" 是 "moats" 的听写误差）：

"And to the extent that that is encoded in a model, I think a lot of their business will be at risk, but to the extent that it is encoded in workflows, that is where they will be able to develop modes."
「如果这些信号只编码在模型里，我认为他们的业务将面临很大风险；但如果信号编码在工作流中，他们就能建立起护城河。」
Tuhin Srivastava · Baseten CEO on the AI Inference Crunch

那个存在性问题——独立的应用层到底还能不能在大实验室面前存在——其实是主持人 Sarah Guo 抛给他的（"does the independent application layer get to exist at all versus the labs?"）。Tuhin 的正面回答：

"I think the application layer will exist for a number of reasons. One is because I think this idea that what is valuable to a company is the user signal that they can gather, that only they can gather."
「我认为应用层会存在有几个原因。首先，我认为对于企业来说，唯一有价值的是他们能收集到的独家用户信号。」
Tuhin Srivastava · Baseten CEO on the AI Inference Crunch

阵营 D · "品牌 / 设计 / 客户爱是新护城河"——Cannon-Brookes / Spiegel 路线

Mike Cannon-Brookes 押设计：

"I do think it makes design more valuable. Good design. Scarce resources probably get more valuable and I think good design is quite scarce. … When you have any major technological transition, we're back to an era of fundamental design."
「我确实认为这让设计更值钱。好设计。稀缺资源大概率会更值钱，而我认为好设计相当稀缺。……每逢重大技术变革，我们就回到一个'基础设计'的时代。」
Mike Cannon-Brookes · 20VC: Atlassian CEO

Evan Spiegel 押的是平台级生态系统：

"Pretty early on, we learned that we had to start working on things that were a lot more complicated and hard to copy. Stories are very easy to copy. That's a feature in our app. Of course, we have the patents or whatever, but it's very difficult to enforce software patents."
「很早我们就明白，必须去做那些复杂得多、难以复制的东西。Stories 太好抄了——它只是我们 app 里的一个功能。我们当然有专利之类的，但软件专利极难维权。」
Evan Spiegel · Snapchat CEO on the future of Human connection

David Cahn 把这个观点压回到一个朴素的检验上：

"If you have real customer love and you've built something that people absolutely need, you're going to be able to navigate through any market environment."
「如果你有真实的客户爱、做出了人们绝对需要的东西，你就能穿越任何市场环境。」
David Cahn · 20VC: Sequoia's David Cahn

阵营 E · "Agent-led distribution 正在改写"——Profound 的第三维度

James Cadwallader (Profound) 提出的是上述四个阵营里没人正面接住的一个不同的题目：

"It's not so much that the front door of the internet has changed; it's the person going through the door that has changed."
「不是互联网的前门变了——是穿过门的'人'变了。」
James Cadwallader · From SEO to Agent-Led Growth
"In the old world of SEO, you were building content designed to be picked up by an algorithm but consumed by a human. In this new world, you are building content designed to be discovered and consumed by an agent."
「在旧的 SEO 世界里，你做的内容是被算法抓取、被人类消费。在新世界里，你做的内容是被 agent 发现、被 agent 消费。」
James Cadwallader · From SEO to Agent-Led Growth

如果这个判断成立，Camp D 的"品牌"概念会被重新定义——agent 不会被海报和广告"打动"，但会被 schema、可解析性、引用频次"打动"。这条线在其他阵营里没被认领。

阵营 F · "Outlier founder 属性"——Kalshi 的元论点

Tarek Mansour (Kalshi) 给的是另一种维度——护城河不在产品也不在分发，在谁能突破市场对其的天然抗拒：

"A startup is intrinsically something that the world does not want to exist."
「初创公司本质上是这个世界并不希望存在的东西。」
Tarek Mansour · Lessons from Alfred Lin and Ron Conway
"I think if you want to achieve outlier results, you need some sort of outlier. I'm going to call it imbalance. You need some sort of outlier, unusual thing that is part of your history, something."
「要做出 outlier 的结果，你需要某种 outlier 的东西。我会叫它'不平衡'。某种异于常人、属于你个人历史的东西——某种什么。」
Tarek Mansour · Lessons from Alfred Lin and Ron Conway

这条不是"什么是护城河"的回答，是"在 AI 时代谁能挖出护城河"的回答。它跟其他五个阵营是元层级关系，不是平行关系。

都没说透的

我的看法

判断（不是事实）：这场争论的关键不是"哪一种护城河是真的"，而是护城河的层级开始按客户类型分裂——B2B 企业客户的护城河大概率是 Camp A + Camp C 的组合（workflow + system of record + retention），消费类 AI 产品的护城河大概率是 Camp D（品牌、设计、生态）+ Camp E（agent-friendly 分发）的组合，硬件 / 物理基础设施的护城河仍是 Camp B（垂直整合）。所谓"AI 护城河"作为单一概念正在解体——这本身就是这一轮变化的重要信号。

我对这个判断的把握：中等。最强的支撑是 a16z 已经自己说出了"there is no AI, there's like a bunch of subspaces"——这等于承认护城河不是单一概念；最弱的环节是 Camp E（agent-led distribution）——我把它当成会成立的趋势，但实证还不够支撑判断它会重要到改写 B2C 护城河结构。

还想知道什么

取材

继续读
线索
阅读线索：agent 不到 9 nines 可靠，那应该卖给企业还是消费者
线索
阅读线索：如果 coding 是 commodity，AI startup 的护城河长在哪儿
线索
阅读线索：推理成本一年跌 10×，AI app 该按什么定价
线索
摘报
线索
摘报
线索
摘报