Cloudflare: 站在网络边缘
Cloudflare 创办于 2009 年,于 2019 年在纽交所上市。它是一家面向企业技术人员提供内容分发网络(CDN)、网络安全和计算能力的基础设施服务公司,其使命是:help build a better internet(帮助建设一个更好的互联网)。创办 15 年来,Cloudflare 据称已服务了 25% 的网站。而它更为人所知的,可能是因为在访问网站的时候会看到一张检测访问者是否是真人(vs. 机器人)的页面——不得不说,这是一个绝佳的给用户潜意识中铺垫认知的做法。
相信很多读者和我一样,并不了解这家公司的业务模式,也不了解它所面临的竞争环境以及发展历程。和很多科技股一样,这家公司在过去 5 年中股价波动很大,在 Covid 结束后,最近股价达到了一个平台期。上周财报发布后,由于业绩指引偏弱,股价下跌,目前市值大约在 240 亿美元。

创办
2009 年,Cloudflare 的三位联合创始人 Matthew Prince、Michelle Zatlyn 和 Lee Holloway 相识并开始合作。Matthew Prince 和 Lee Holloway 之前曾合作过一个名为 Project Honey Pot 的项目,旨在追踪垃圾邮件的来源。后来,Matthew Prince 休学去哈佛商学院攻读 MBA,在那里遇到了 Michelle Zatlyn。他们开始共同制定商业计划,这就是 Cloudflare 的雏形。从哈佛商学院毕业后,住在加州的 Lee Holloway 加入了他们,三人花了整个夏天完善 Cloudflare 原型。2010 年,他们正式推出了 Cloudflare,目标是帮助构建一个更美好的互联网。
三个创始人的分工角色非常明确:Prince 作为创始人和 CEO 负责公司整体战略和业务发展,Zatlyn 是那个将 Prince 的想法有条不紊地付诸实施的运营者,Holloway 则主管技术和工程。不幸的是,Holloway 在 2015 年前后被诊断出患有额颞叶痴呆症(FTD),这是一种罕见的进行性脑部疾病,无法治愈。FTD 是一种破坏性疾病,会影响大脑的额叶和颞叶,导致性格、行为和认知发生变化。Holloway 的诊断成为他人生的转折点,他开始与健康作斗争,最终于 2016 年从 Cloudflare 辞职。
三个创始人的背景极为多元化。Prince 在访谈中谈到了选择创业伙伴的一些哲学:
- 这让我意识到,我们的目标不是和好友一起创业,而是和拥有不同技能的人一起创业。
- 从 13 年前开始,你知道,确实有人在台下问我们:嘿,我和我的联合创始人很难弄清楚如何分工。我反问:你们之前是朋友吗?回答是:是的,我最好的朋友。我说:你完蛋了。
- 更多元化的团队往往更容易获得成功。
Prince 提到 Holloway 患病的经历时动情地流下了眼泪,而他对 Zatlyn 的赞誉也不绝于口。有趣的是,Prince 说他在哈佛商学院的时候和 Zatlyn 并不熟悉,但在一次课堂案例讨论中注意到了这个女生的独特之处,后来才开始了长期合作。
产品
从 Project Honey Pot 开始,安全就成为了 Cloudflare 基因中不可或缺的一部分。许多媒体和分析报告也将 Cloudflare 归类为「网络安全」公司。事实上,Cloudflare 最初的卖点是能够拦截恶意攻击的 CDN。在同质化竞争和价格战的压力下,安全和性能成为 Cloudflare CDN 的独特附加值。
虽然 Cloudflare CDN 的核心职责是快速为用户提供静态资产,但他们还添加了一些杀手级功能,使产品更具吸引力——其基础是安全性。 Cloudflare 的 CDN 可保护您的网站免受 DDoS 或分布式拒绝服务攻击,在这种攻击中,攻击者试图通过大量请求来摧毁您的网站。
via Technically: What does Cloudflare do?
从 CDN 起家,Cloudflare 与 ISP 建立了深度合作关系:他们将服务器部署到运营商的机房,一方面提高了本地访问性能,另一方面也增强了机房网络的安全性。
因此,他们与世界各地的所有 ISP 合作,在全球 100 多个城市建立主机托管设施。对于 ISP 和 Cloudflare 来说,这是一种奇怪的关系。ISP 允许 Cloudflare 在其设施上部署服务器,可以降低带宽成本。Cloudflare 还可以帮助他们抵御 DDoS 攻击等。因此,一旦 Cloudflare 的服务器部署在这些代码、ISP 和设施上,他们就有了 CDN 业务或核心产品。除此之外,由于这些服务器是可编程的,他们可以在其上构建很多东西。因此,它们只是在同一个基础架构层上,然后,正如他们自己所说的那样,以快速的速度在其上构建。
CDN 就是让他们进入大门的特洛伊木马,因为从 ISP 的角度来看,如果每个商业客户,每个连接到你的普通网民整天都在上网。每次他们要去某个地方,您都必须连接到外部互联网,然后去某个地方再回来。这都是带宽成本,连接成本。如果有 CloudFlare,那就是缓存,比如您主机托管中的数千个网站,这些流量甚至不会出去。它是免费的。基本上,这就是 Netflix 和 Google 的做法,他们都在 ISP 内部放置了一堆机架。当您观看 Netflix 时,您不会在服务器上的开放互联网上观看电影,对吧?一切都在您的 ISP 内部,Cloudflare 就是这样做的。
via Going Deep on Cloudflare and Datadog with Mostly Borrowed Ideas (MBI)
Cloudflare 管理层经常提到其覆盖城市的数量。作为一家提供网络连接基础服务的公司,网络设施的地理分布至关重要。Cloudflare 扮演着互联网基础设施中毛细血管的角色,将安全和性能服务渗透到全球各地用户所在的城市。
以 CDN 和安全为基础,Cloudflare 近年来开始发力边缘计算,推出了 Cloudflare Workers 服务,被视为其增长的第三条曲线。
因此,美国的 CDN 是第一条 S 曲线,但 S 曲线有点趋于平稳,除此之外,它已经相当成熟。第二曲线是安全,这个还处于早期阶段,而且比 CDN 更大。这是一个更大的市场 ... 第三个是 Workers ... 现在他们在世界各地有 275 个地点,到处都是服务器。这些服务器,就像他们仍然拥有大量资源一样。他们发现,与其使用像 CDN 一样只提供 HTTP 请求的愚蠢服务器,不如在服务器上(基本上是 Edge 平台)进行大量计算,而不是必须返回中央服务器来执行应用程序或服务的逻辑,如果可以在边缘执行这些逻辑,对吗?对于某些非常依赖延迟的东西来说,速度会更快,而且,这有点像他们在监管世界中发现的一大用例。欧盟每个国家都有不同的数据法和隐私法。而且这些东西还在不断增加。好吧,如果你可以在本地处理数据,并遵守当地的法律法规等,那么这将是一个巨大的增值。
via Going Deep on Cloudflare and Datadog with Mostly Borrowed Ideas (MBI)
在 CDN、安全和计算三条曲线之上,Cloudflare 推出了 Connectivity Cloud 产品组合,旨在为企业客户提供一体化的网络安全、性能和计算服务。这使其与三大公有云巨头展开了竞争,尽管目前收入规模还相差悬殊。得益于分布式网络拓扑,Cloudflare 的 Workers 提供的计算能力遍布全球数百个城市,支持无服务器应用和数据存储服务。在 Gen AI 时代,Cloudflare 还推出了 Worker AI,让开发者能在其平台上运行自己的 AI 模型,提供分布式算力。
市场和竞争
Cloudflare 采用了典型的 bottom-up 的 go-to-market 策略,通过出色的产品力影响技术人员,再逐步说服 CTO/CIO。公司每季度披露付费客户总数和大客户(ARR>100k)数量。2023年,虽然付费客户总数持续增长,但大客户增速放缓,最近三个季度分别为2558、2756和2878。为加速增长,公司聘请行业老兵Mark Anderson担任营收总裁。
Cloudflare 以优质廉价的服务在开发者社区享有盛誉。注册免费账号即可享受安全防护,还能用 Worker 托管静态页面。Cloudflare 的 1.1.1.1 DNS 服务号称全球最快,免费提供给所有用户,并可过滤恶意软件和色情内容。对一家面向技术人员和企业客户的公司而言,能有如此知名度实属不易。
Prince 曾提到,他会在 Twitter 上搜索用户反馈,并转发给相关团队改进。后来团队甚至开发了监测脚本,及时捕捉用户声音。
开发者体验(DX)是近年的热词。随着技术人员成本上涨,各平台在产品体验上竞相比拼。一个明显变化是,这些平台的UI越来越精美,甚至超越许多 C 端应用。相比 Cloudflare 这样的中生代,Vercel 等新一代平台可能在 DX 上下了更多功夫。
Cloudflare 的老牌竞争对手 Akamai 在 2022 年收购了 Linode。Akamai 成立于 1999 年,比 Cloudflare 早十年;Linode 成立于2003年,提供低成本自助式 VPS 服务,经常在科技博客和播客投放广告,吸引中小型开发者。Akamai认为可借助自身强大的销售能力,加速 Linode 的收入增长;同时补充云计算/存储等产品线,有利于拓展新客户和向老客户交叉销售。
显然,这一思路与 Cloudflare 注重开发者体验,从漏斗上端广泛获取潜在客户的策略不同。
Ben Thompson 写道:
这笔交易凸显了 Akamai 和 Cloudflare 之间的根本区别,并且对后者来说是一个极其看涨的信号。
请记住,Akamai 是一家老牌公司,其业务建立在(现在)互联网内容爆炸式增长的古老技术之上;这意味着该公司的基础设施建立在大量离散的动力不足的节点上,这些节点通过专业化和绝对的地理覆盖范围解决了可扩展性。当然,多年来,Akamai 已经更新了这些节点,并增加了有限的边缘计算功能,但该公司对计算和内容起源是集中的并分层推出的想法有二十年的假设;更重要的是,这些假设实际上已内置于网络中。
与此同时,Cloudflare 受益于十年后的出现,并构建了一种完全不同类型的网络:该公司没有使用专用硬件,而是使用商品硬件(功能也更强大),并创建了 CDN 的所有功能(以及 DDoS 保护)服务,其最大的产品,也是该公司受邀加入世界各地 ISP 的软件)。这个概念被称为软件定义网络,但如果说这个术语低估了 Cloudflare 正在做的事情的话:Workers 计算平台的添加尤其表明它更接近于软件定义的一切。
在 Cloudflare 的世界观中,区分中心和边缘的想法是荒谬的。 Cloudflare 网络中的每台服务器都可以作为 Worker 或 CDN 或 VPN 或安全网关或任何其他可以在软件中编程的服务器运行,也就是说任何东西;该公司的网络就像变形虫一样,能够改变形状,甚至分裂成离散的部分(如地理边界)。换句话说,Akamai 收购 Linode 并不是对 Cloudflare 的做法构成挑战;而是对 Cloudflare 的收购。它正在认输。
更大的竞争对手还是三大公有云厂商:AWS、Azure 和 GCP 都有自己的 CDN、安全和计算产品,Cloudflare 如何竞争呢?
Prince 和 Zatlyn 毕业于 HBS,深谙「颠覆式创新」之道。2021 年,Cloudflare 推出 R2,这是一个低成本的存储服务,在公司 blog 上,Cloudfare 这样介绍这个服务:
我们很高兴地宣布推出 Cloudflare R2 Storage!通过让开发人员能够存储大量非结构化数据,我们正在扩展 Cloudflare 的功能,同时将与典型云存储服务相关的出口带宽费用削减至零。Cloudflare R2 Storage 具有完整的 S3 API 兼容性,可与现有工具和应用程序协同工作。
与 AWS S3 完全兼容,同时出口带宽削减为零——经典案例。

Ben Thompson 分析了其中的经济学,不仅仅适用于 R2,也适用于其它产品:
Cloudflare 能够实现这一目标的原因与 S3 利润如此惊人的原因相同:带宽是一项固定成本,而不是边际成本。举个最简单的例子,如果我有两台计算机通过电缆连接,那么带宽成本就是我为电缆支付的费用;连接后,我可以在任一方向免费传输任意数量的数据。
当然,这并不完全正确:我受到电缆容量的限制;为了支持更多的数据传输,我必须安装更高容量的电缆,或者更多。但是,如果我已经为我最初的核心业务(即保护网站免受分布式拒绝服务攻击并提供内容交付网络)构建了一个全球电缆网络,那么它的价值就是世界各地的 ISP 都为我提供了空间他们放置我的服务器的设施?好吧,那么我就已经拥有大量带宽,其使用边际成本为零,而且顺便说一下,靠近最终用户的位置可以放置一大堆硬盘。
换句话说,这一切都取决于在一切开始的时候,Cloudflare 在 ISP 那里部署的机器,最初它们是为 CDN 和安全服务的,而现在,它们可以享有这张庞大的网络,进而提供更多服务。
AWS 和 Azure 们并不会坐等 Cloudflare 在边缘发起颠覆,它们也推出了各自的边缘计算服务。但 Forrester 的报告 认为,它们与 Cloudflare 相比还处于竞争者的位置上。

财务表现
我对比了 Amazon、Microsoft 和 DataDog(IT 监控服务提供商)的回报。如果你在 Cloudflare IPO 时就买入并持有,到现在(2024 年 5 月初)能够提供 283% 的回报,在几个标的中是最高的。Cloudflare 股价的波动性也是最大的,曾经在 2021 年末达到 200 美元以上的高点,但目前回到了 70 美元附近。最近财报发布后,由于指引疲软,股价还有进一步下跌的风险。

这几支股票的对比实际上代表了两类观点:一边是大型的公有云厂商,一边是过去十年中出现的新玩家。在新玩家阵营中,DataDog 比 Cloudflare 更像一家 SaaS 公司,专注于为企业客户提供各种性能监控、日志分析等工具,详细分析可以去看 MBI Deep Dive,这家公司的亮点在于,它近年来保持研发费用比率大于营销费用比率(占收入的百分比),在收入增速 25%+ 的同时,毛利率也保持在高位(2023 年 80.76%,2022 年 79.31%)。这家公司的策略是不断向老客户提供更多的软件服务,从而不断堆高 RPU 和 ACV(客单价和合同单价),最终在利润结构上越做越好。这家公司在 2023 年扭亏为盈,但仔细看会发现是因为利息收入的贡献,如果减去这部分收入,业务还在微亏。
相比来看,Cloudflare 的 SG&A 费用率(2023 年 63.01%),高于自己的 R&D 费用率(2023 年 27.62%),也高于 DataDog 同期的 36.59%;如果直接比较 dollar value,DataDog 的 SG&A 费用其实略高于 Cloudflare,但前者收入规模更大,摊薄了这些费用。但如果要比较 R&D 费用的话,DataDog 几乎 3 倍于 Cloudflare 的投入。从员工数量来看,截至 2023 年 12 月 31 日,DataDog 有 5200 名员工, Cloudflare 有 3682 名员工,但后者员工数量增加更快,一度以 30%-40% 的速度增加,2023 年增速降到了 14%,在 Q1 2024 年的财报中,Prince 仍然表示会大幅增加销售人员的数量。
亏损和放缓的增长都不是投资人希望看到的。投资人担心,过度扩张的团队会降低销售人效,但无法带来相应的销售增长;也担心在 Worker AI 上投入的 CapEx 会带来毛利率降低;更担心复杂的地缘政治会影响 Cloudflare 的分布式策略。
在财报电话会议上,管理层反复强调在 GTM 特别是企业销售方面的投入,认为地缘政治反而带来了企业对网络安全的更高要求,Prince 在 call 上一连列出了多个近期签约的客户和订单金额,也代表了目前公司的发展重心。
记不清是在哪篇文章中读到过,说 Cloudflare 是一家没有亮点的公司,的确,Cloudflare 的产品线中几乎每一项都能找到诸多竞争对手,其差异化不在于产品本身,而在于它独特的「位置」——这里说的位置就是指其网络节点所处的位置,离客户和用户更近,从而获得更好的性能。以此为基础,逐渐筑高城墙。Prince 曾在访谈中讲到这些边缘节点最终也会产生网络效应,因为每一个节点的加入,都在让其它节点获得更好的连通性。网络效应的价值自然不需多言,但如何衡量这张网络的价值,还有待商榷。
这家创办了 15 年的公司后续会在一个更加不确定的时代因其过去所积累的分布式服务能力获得新的机会。AI 无疑是它投资未来的重要方向,Prince 也提到 Cloudflare 在采购 GPU 并部署到边缘节点上,为客户提供更近也更快的算力。
在 Gen AI 的计算分布上,出现了 3 层结构:
- Hyperscalers
- Edge
- Device
正如最近在模型界涌现出来的不同尺寸的模型一样,对应不同的需求场景,在成本和性能上做取舍。这样,Gen AI 并不是飘在天上的,而是越来越扎根到现实世界中。Cloudflare 沉淀下来的正是在物理上的临近性,有机会在 Gen AI 所带来的这一波浪潮中找到更大的增长机会,但这显然还需要多年的耐心投入才能看到结果。
Appendix & References
References
- Stratechery: Cloudflare’s Disruption, Cloudflare on the Edge, Why Cloudflare Matters, The Absence of Gatekeepers, Promotion Versus Moderation
- Liberty's Highlights: 241: Cloudflare Q4, Streaming Churn, Construction Costs, SpaceX, Deepmind AI + Youtube, Fusion Breakthrough, Bloomberg vs Coal, and Type I & II Fun, Going Deep on Cloudflare and Datadog with Mostly Borrowed Ideas (MBI)
- The Logan Bartlett Show: Interview with Matthew Prince: Untold Stories Behind the $20B Business That Runs the Internet
- MBI Deep Dives: Datadog: Leading the Cloud Infrastructure Dogfight
- Technically: What does Cloudflare do?
- Wired: The Devastating Decline of a Brilliant Young Coder
Q1 24 Earnings Call - Notes on Enterprise Sales
Based on the earnings call, Cloudflare has taken several steps to enhance its enterprise sales capabilities:
- Leadership hiring: Cloudflare hired Mark Anderson as its new President of Revenue to lead the enterprise sales initiatives. Anderson brings extensive experience in enterprise sales from his prior roles. Since joining, Anderson has been focused on assessing and evolving Cloudflare's go-to-market organization.
- Sales team expansion: Cloudflare is accelerating its sales hiring, especially for enterprise-focused roles. The company saw a 56% increase in applicants for strategic account and enterprise sales positions in March after Anderson's appointment. Cloudflare added several senior go-to-market leaders, including a new Chief Partner Officer, Head of Global Sales and Renewals, and multiple regional strategic account sales leaders.
- Sales productivity improvements: Cloudflare has been refining its go-to-market strategies and operations over the past five quarters. This has resulted in double-digit year-over-year improvements in sales productivity in Q1, continuing the positive trend from the trough in early 2023. The company expects this trajectory to continue under Anderson's leadership.
- Large deal wins: Cloudflare highlighted several large enterprise deal wins in Q1 as evidence of its traction in the segment. Key wins included a 3-year $40M contract with a leading technology company, a 4-year $10M contract with a Fortune 100 financial services company, a 3-year $4.5M contract with a large international energy company, and a 1-year $800K contract with a U.S. government agency.
- Partner ecosystem investments: Cloudflare hired a new Chief Partner Officer to scale its enterprise reach through partnerships. The company is collaborating more closely with partners like Accenture, which helped land a major UK public sector deal.
- Product investments: While not strictly a sales initiative, Cloudflare continues to enhance its product portfolio for the enterprise segment. This includes the Cloudflare One suite for SASE and Zero Trust, Workers AI serverless compute platform, Magic WAN, Magic Firewall, etc. The expanding product capabilities make Cloudflare more attractive as a strategic enterprise vendor.
These initiatives are still in the early stages, but Cloudflare is seeing positive signs in its ability to hire experienced talent, win larger enterprise deals, and improve sales productivity. The company plans to continue investing aggressively in enterprise sales under Mark Anderson's leadership to capitalize on its long-term opportunity in the segment.
Interview with Matthew Prince: Untold Stories Behind the $20B Business That Runs the Internet
Mission
- We've started to think about this as how do we make sure that everything in the world can be connected securely, efficiently, reliably, privately, and quickly together, and that's what we're trying to work on every day at Cloudflare.
- At AWS, you're measured on the KPI of what percentage of a customer's data you store and hold on to, and that's just not how we think about ourselves. What we think about is how many endpoints we are connecting together, and data flows through us, but we don't actually hold on to it for the most part very long. We want to connect things together, and so as we've thought through this and tried to say how we can make a simple description of what we do and how it's different than some of the other clouds out there, it is that we're all about connectivity, and fundamentally, that makes us the connectivity cloud.
Founding team
- What that made me realize was that the goal wasn't to start something with your buddies the goal was to start something with people who had a different skill set than you did.
- From 13 years ago and you know we literally had people come up to us off stage asking hey my co-founder and I are having a hard time figuring out how we split things up and I'm like you guys were friends beforehand like yeah my best friend and I'm like you're dead.
- More diverse team win.
Strategy
- I think that the thing that, and this is very much Michelle, but the thing that we did that helped us on both of those fronts was, first of all, to understand that signing up small customers today, or when we were starting, was a critical thing for us to do to build scale and build reps, but it wasn't where the business ended; it just was where it started. But then, secondly, to make sure that we chose the KPIs that we measured ourselves by, which accurately reflected what we thought were the right steps in the business.
- Revenue was not a number that we talked about for the first four years of Cloudflare's life, but we spent a huge amount of time talking about what the cost of processing requests were because we knew that that was something that we had to keep driving down and down and down and down and down in order to make the business viable over the long term.
Team / culture
- We had a rule that said we could never more than double the size of the company or any individual team in 12 months or less, and the rationale for that was that companies are mostly not democracies. We don't take a vote on what programming language we use, what CRM system we use, what customers we go after, or what products we build next. We have people who specialize in each of those things, but the one thing in every company that's democratic is culture. In fact, it's harder to find what culture is other than it's what the majority believes and how they act. If that's the case, the problem is that if you grow faster than doubling in 12 months, that means that inherently your culture will change because you will have more new people who have been there for less than 12 months.
Regulations
- I got to know Eric Schmidt, the former CEO of Google, and I was talking to him when I said, "Eric, how do you think about public policy issues?" And he said, "Matthew, there are no public policy issues; there are only product issues." What I think he meant was if you build a product which is valuable enough, it rises above the fray, and on some levels, it almost becomes unregulatable.
Taking feedbacks from Twitter
- There are always one or two ideas that still come surfacing up from Twitter which we recognize as really good ideas that we could probably build in the period of time that we have, and so I do think that there is still some value in that, and I miss that aspect of it. I have not yet found another platform, and you cannot replicate that with Facebook, Instagram, Snapchat, TikTok, Threads, or any of the different platforms that are out there. It will be interesting, and I think that it would have been much harder for us to get the scale, awareness, and to have the transparency that we did if that platform had not existed. If you are building a company to try and disrupt Cloudflare, it got a little bit harder as Twitter has morphed into something which is a little bit different than what it was as we were coming up, but that does not diminish the importance of following the same path that we did.
Going public
- Our first earnings call in 2015 was a total disaster. Everybody, including our VCs and the Fidelity folks, mocked us, saying, "Thank God that was not when you were actually public." But then we did it every single quarter, and our guidance at first was terrible. We had no idea how to give guidance, so it was way off. But again, we did it over and over and over, and we could do it. So, by the time we actually got to our first public earnings call, which I believe was our 13th one, the machinery was in place, and I think it gave us the confidence to be able to do that. It was less of a big step from one thing to another, and I think we were very prepared.