70亿美元。
美国时间8月16日，支付巨头Stripe最终敲定协议，以超过70亿美元的价格收购AI模型路由平台OpenRouter。这个数字是什么概念？仅仅三个月前——2026年5月——OpenRouter刚以13亿美元的估值完成B轮融资。三个月，估值翻了5.4倍。
更有意思的是买家的身份。Stripe，这家过去十几年专注做在线支付基础设施的公司，为什么要花70亿美元买一个很多人眼中的"API中转站"？
因为OpenRouter做的事情，表面上是"让开发者用一个API调用所有模型"，本质上是AI时代的流量分发权——当你每次请求都要经过它来决定走GPT、走Claude、走Gemini还是走开源模型的时候，它就成了AI世界的交通信号灯。而交通信号灯的价值，在任何时代都不亚于道路本身。
支付巨头杀进AI基础设施，这场战争才刚刚开始。
OpenRouter到底值不值70亿？
先搞清楚一个问题：OpenRouter是做什么的？
简单说，它是一个AI模型的统一网关。开发者不需要分别对接OpenAI、Anthropic、Google、开源模型等几十个API，只需要接OpenRouter这一个接口，就能调用超过400种AI模型。平台自动处理模型路由、故障切换、负载均衡、成本优化——哪个模型便宜用哪个，哪个模型快用哪个，哪个模型挂了自动切到备用。
这个模式听起来是不是很熟悉？对，它就是云计算时代的多云管理平台，或者支付领域的Stripe本身——Stripe当年做的事情，就是让商户不用分别对接微信、支付宝、信用卡、Apple Pay，接一个Stripe搞定所有支付渠道。现在OpenRouter在AI领域做的是一模一样的事情。
那它为什么值70亿？三个原因。
第一个原因是增长速度。今年5月估值13亿，8月就卖到70亿——这不是正常的SaaS增长曲线，这是指数级爆发。能让Stripe在三个月内愿意多付50多亿美元，说明OpenRouter的业务数据在这三个月里发生了质的飞跃。开发者用脚投票，越来越多的AI应用把模型路由层交给OpenRouter，而不是自己对接每一家。
第二个原因是战略位置。模型路由是AI栈里一个非常微妙的位置——它在模型提供商和应用开发商之间，是所有AI请求的必经之路。谁控制了这一层，谁就拥有了：定价权（可以跟模型厂商谈批量折扣）、数据权（能看到所有请求的类型和分布）、分发权（新模型上线能不能快速获得流量，它说了算）。这跟搜索引擎在互联网时代的地位、应用商店在移动时代的地位是同一个逻辑。
"当年所有人都觉得做支付通道没技术含量，直到Stripe做成了950亿美元的公司。现在很多人觉得模型路由就是个API转发，直到它卖到了70亿。"—— 一位硅谷投资人的评论
第三个原因是网络效应。OpenRouter接入的模型越多，对开发者的吸引力就越大；用OpenRouter的开发者越多，模型厂商就越愿意主动接入。这个正向飞轮一旦转起来，后来者很难追赶——就像支付领域，商户不会因为另一个支付通道费率低一点就轻易切换，因为迁移成本太高了。
当然，70亿也不是没有争议。有人觉得三个月翻5.4倍太疯狂，这是AI泡沫的又一铁证。但你回头看Stripe自己的历史——2010年代初，也有很多人说"网上支付通道能值几个钱"，后来Stripe估值最高冲到了950亿美元。基础设施的价值，总是在被低估的时候爆发。
支付巨头为什么要买AI公司？
更值得玩味的问题是：为什么是Stripe？为什么是一家支付公司来买OpenRouter？
表面上看，支付和AI模型路由是两个完全不相关的业务。但如果你往深想一层，就会发现这两者的底层逻辑是一模一样的。
Stripe的核心能力是什么？不是"处理信用卡支付"这个动作本身——这个技术很多公司都能做。Stripe真正的核心能力是抽象复杂性：把各个国家、各个支付方式、各个银行、各个监管要求的极端复杂性，抽象成一个简单的API。开发者只需要几行代码，就能接入全球支付能力，不需要关心背后的复杂性。
OpenRouter做的是什么？一模一样的事情：把各个AI模型、各个API接口、各个定价策略、各个速率限制的极端复杂性，抽象成一个简单的API。开发者只需要几行代码，就能调用全球所有主流AI模型，不需要关心背后的复杂性。
这不是跨界并购。这是Stripe在自己最擅长的赛道上，向相邻领域的自然延伸。支付是"钱的路由"，OpenRouter是"AI请求的路由"——Stripe买的不是一个新业务，是同一种能力在新时代的应用。
而且这里面还有更深的协同效应。
第一，客户重叠。Stripe的客户是谁？是全球数百万的互联网公司和开发者——这些人，恰恰也是OpenRouter的目标客户。你已经在用Stripe处理支付了，现在Stripe顺便帮你处理AI模型路由，这是顺理成章的交叉销售。
第二，数据协同。Stripe知道每一笔交易的金额、商品、用户画像；OpenRouter知道每一次AI调用的类型、成本、效果。这两者结合起来，能做的事情太多了——AI驱动的反欺诈、智能客服、个性化推荐、动态定价……支付数据+AI能力，这是任何纯AI公司都不具备的优势。
第三，商业模式升级。支付是一个交易抽成的生意，规模大但利润率有限。AI基础设施是一个按调用量收费的生意，增长快、毛利高、粘性强。Stripe通过收购OpenRouter，直接把自己从"支付公司"升级成了"商业AI基础设施公司"——这个故事在资本市场上的估值空间，比纯支付大得多。
"Stripe的野心从来不是做最好的支付公司。它的野心是做互联网的基础设施层——支付是第一个切口，AI是第二个。"—— 一位科技行业分析师
你看，这就解释了为什么Stripe愿意出70亿。这不是买一个收入多少倍PS的SaaS公司，这是买一张下一代互联网基础设施的入场券。当年错过了云计算，错过了移动互联网入口的人，不会再错过AI基础设施这波浪潮。
AI中间层战争正式打响
Stripe收购OpenRouter，更大的意义在于：它标志着AI行业的竞争，正在从"模型层"向"中间层"蔓延。
过去三年，AI行业的战争主要发生在模型层——OpenAI、Anthropic、Google、百度、阿里、字节，大家拼的是谁的模型更强、谁的参数更多、谁的跑分更高。模型层是AI时代的"芯片"，确实重要。
但随着模型能力越来越趋同——GPT-5.6、Claude Opus、Gemini 3、Qwen3.8、DeepSeek V4在很多场景下的差距正在缩小——真正决定用户体验和商业价值的，正在从"模型本身"转向"怎么用模型"。这中间包括了：模型路由、提示工程、Agent编排、工具调用、成本优化、监控可观测性……这些被统称为"AI中间层"。
中间层的战争，现在才刚刚开始。而OpenRouter只是第一枪。
我们可以预见，接下来会有更多的并购和整合发生在这个领域：
云厂商不会坐视不管。AWS、Azure、Google Cloud本身就在做模型路由和统一API的事情——Amazon Bedrock、Azure AI Studio、Google Vertex AI本质上都是官方版的OpenRouter。现在Stripe这个第三方杀进来，云厂商肯定会加速自己的中间层布局，甚至可能收购更多垂直领域的AI工具公司。
大模型厂商会防御性布局。OpenRouter这种中立第三方的存在，对模型厂商来说是双刃剑——好处是带来流量，坏处是削弱了用户粘性。你用OpenRouter，今天用GPT明天可以无缝切Claude，模型厂商的护城河就变浅了。可以预见，OpenAI、Anthropic们会推出更多"绑定式"的功能和优惠，试图把用户留在自己的生态里，不被中间层"截胡"。
创业公司还有机会。OpenRouter被收购不代表中间层战争结束，反而说明这个赛道被验证了。在更垂直的领域——比如专门做Agent编排的、专门做AI监控的、专门做成本优化的——还会有新的创业公司跑出来，甚至可能成长为下一个OpenRouter。
当然，OpenRouter被Stripe收购之后，也不是没有挑战。最大的问题是中立性。被收购之前，OpenRouter是完全中立的——哪个模型好就推荐哪个，不会偏向任何一家。但被Stripe收购之后，其他支付公司、其他云厂商会不会担心Stripe"偏心"？模型厂商会不会担心自己在OpenRouter上的权重被操纵？这种信任问题，是Stripe需要认真解决的。
谁掌握了路由，谁就掌握了AI的咽喉
让我们把视角拉得更大一点。
互联网的历史，某种程度上就是一部"路由权"争夺的历史。
最早的互联网，路由权掌握在ISP手里——你连哪个运营商的网，就走哪个运营商的线路。后来搜索引擎出现了，路由权转移到了Google手里——你搜什么、看到什么结果，Google决定。再后来移动时代，应用商店掌握了路由权——你能下载什么App、什么App能被推荐，苹果和Google决定。再后来社交媒体时代，Facebook和Twitter掌握了信息流的路由权。
每一次路由权的转移，都伴随着巨大的财富转移和权力更迭。掌握路由的公司，无一例外都成为了那个时代的巨头。
AI时代的路由权是什么？就是当你说"帮我订张机票""帮我写份报告""帮我分析一下这份数据"的时候，这个请求会被交给哪个AI、用哪个模型、走哪个工作流——这个决定权，就是AI时代的路由权。
现在，这个权力还处于分散状态。你用ChatGPT，路由权在OpenAI手里；你用Claude，路由权在Anthropic手里；你自己开发应用接API，路由权在你自己手里。但随着AI应用越来越复杂、需要调用的模型越来越多、成本优化越来越重要，越来越多的开发者会把路由权交出去——交给OpenRouter这样的专业平台。
Stripe看到了这一点。所以它花70亿美元，买下了AI时代第一个规模化的路由平台。它赌的不是OpenRouter今天有多少收入、多少客户，它赌的是：未来五年，绝大多数AI请求都会经过一个路由层，而它要成为那个路由层。
"互联网的每一个时代，最终赢的都不是造路的人，是站在路口收过路费的人。Stripe现在要做AI时代的路口收费员。"—— 一位互联网老兵的观察
当然，这场战争才刚刚开始。OpenRouter现在是领先者，但远不是终局。云厂商有天然的基础设施优势，大模型厂商有生态绑定优势，创业公司有垂直深耕的机会——谁能最终掌握AI的路由权，现在还言之过早。
但有一点是确定的：70亿美元的收购价，已经给AI中间层这个赛道定了价。它不再是一个"小工具""API转发器"，它是一个价值数百亿美元的基础设施赛道。接下来会有更多的钱、更多的公司、更多的人才涌入这个领域。
当所有人都在盯着大模型你追我赶的时候，Stripe悄悄在模型的上游买下了一个收费站。这步棋，很高明。
明天见。
$7 billion.
On August 16 US time, payments giant Stripe finalized a deal to acquire AI model routing platform OpenRouter for over $7 billion. Let that number sink in: just three months ago, in May 2026, OpenRouter closed its Series B at a $1.3 billion valuation. Three months. A 5.4x return.
The more interesting question is the buyer. Why would Stripe, a company that spent the last 15 years building online payment infrastructure, drop $7B on what many dismiss as an "API middleman"?
Because what OpenRouter does on the surface—"let developers call every AI model through one API"—is in practice traffic distribution power for the AI era. When every request passes through its gates to decide whether it goes to GPT, Claude, Gemini, or an open-source model, it becomes the traffic light of the AI world. And traffic lights, in any era, are worth no less than the roads themselves.
The payments giant just stormed AI infrastructure. The war is only beginning.
Is OpenRouter Actually Worth $7 Billion?
First, let's answer a basic question: what does OpenRouter do?
Simply put, it's a unified gateway for AI models. Instead of integrating dozens of separate APIs from OpenAI, Anthropic, Google, and open-source model providers, developers integrate once with OpenRouter and get access to over 400 AI models. The platform handles model routing, failover, load balancing, and cost optimization automatically—use whichever model is cheapest, fastest, or most reliable, with automatic backup if one goes down.
Does this model sound familiar? It should. It's the multi-cloud management platform of the cloud computing era, or Stripe itself in the payments space—what Stripe did for payments was let merchants skip integrating WeChat Pay, Alipay, credit cards, and Apple Pay separately; one Stripe integration handled every channel. OpenRouter is doing the exact same thing for AI.
So why $7 billion? Three reasons.
First: growth velocity. $1.3B in May, $7B in August—this isn't normal SaaS growth; this is exponential explosion. For Stripe to willingly pay an extra $5+ billion in three months, OpenRouter's business metrics must have undergone a qualitative leap in that window. Developers voted with their feet; more and more AI applications are handing their model routing layer to OpenRouter instead of integrating every provider themselves.
Second: strategic position. Model routing occupies an incredibly delicate spot in the AI stack—between model providers and application developers, on the critical path of every AI request. Whoever controls this layer holds pricing power (volume discounts with model vendors), data power (visibility into every request's type and distribution), and distribution power (deciding which new models get instant traffic). This is the same logic that made search engines the gatekeepers of the web era, and app stores the gatekeepers of mobile.
"Everyone thought payment processing was trivial tech until Stripe built a $95B company. Now many people think model routing is just API forwarding—until it sells for $7B." — A Silicon Valley Investor
Third: network effects. The more models OpenRouter supports, the more attractive it is to developers. The more developers use OpenRouter, the more model providers will proactively integrate. Once this flywheel spins, it's incredibly hard for newcomers to catch up—just like in payments, merchants don't switch processors over a slightly lower rate because the migration cost is too high.
Of course, $7B isn't without controversy. Some argue that a 5.4x jump in three months is madness, yet another data point for the AI bubble. But look at Stripe's own history—in the early 2010s, plenty of people said "how much can an online payment gateway be worth?" before Stripe's valuation peaked at $95 billion. Infrastructure value always erupts when it's most underestimated.
Why Is a Payments Giant Buying an AI Company?
The more intriguing question: why Stripe? Why would a payments company acquire OpenRouter?
On the surface, payments and AI model routing seem like completely unrelated businesses. But think one layer deeper, and you'll realize their underlying logic is identical.
What is Stripe's core competency? It's not "processing credit card payments" itself—plenty of companies can do that. Stripe's real superpower is abstracting complexity: taking the extreme complexity of different countries, payment methods, banks, and regulatory requirements, and abstracting it all into one simple API. A few lines of code give developers global payment capabilities without needing to understand the underlying mess.
What does OpenRouter do? The exact same thing: abstracting the extreme complexity of different AI models, API interfaces, pricing strategies, and rate limits into one simple API. A few lines of code let developers call every major AI model on the planet without worrying about what's underneath.
This isn't a cross-sector acquisition. It's Stripe extending naturally into an adjacent domain using the capability it does best. Payments is "routing money"; OpenRouter is "routing AI requests"—Stripe isn't buying a new business. It's buying the same capability applied to a new era.
And there are deeper synergies still.
First: customer overlap. Who are Stripe's customers? Millions of internet companies and developers worldwide—precisely OpenRouter's target audience. If you already use Stripe for payments, Stripe handling your AI model routing too is a natural cross-sell.
Second: data synergy. Stripe knows the amount, product, and user profile of every transaction. OpenRouter knows the type, cost, and outcome of every AI call. The combination unlocks so much—AI-driven fraud detection, intelligent customer service, personalized recommendations, dynamic pricing... Payments data plus AI capability is an advantage no pure-play AI company has.
Third: business model upgrade. Payments is a transaction-take business with massive scale but limited margins. AI infrastructure is a usage-based business with fast growth, high gross margins, and strong stickiness. By acquiring OpenRouter, Stripe upgrades itself from a "payments company" to a "commercial AI infrastructure company"—a narrative that commands a much larger valuation multiple in capital markets than pure payments.
"Stripe's ambition was never to be the best payments company. It's to be the infrastructure layer of the internet—payments was the first wedge; AI is the second." — A Tech Industry Analyst
You see? That explains why Stripe is willing to pay $7 billion. It's not buying a SaaS company at some revenue multiple—it's buying a ticket to the next generation of internet infrastructure. Those who missed cloud computing, who missed mobile internet entry points, won't miss the AI infrastructure wave.
The AI Middle Layer War Officially Begins
Stripe's acquisition of OpenRouter has a larger significance: it marks AI industry competition spreading from the "model layer" to the "middle layer."
For the past three years, the AI war was primarily fought at the model layer—OpenAI, Anthropic, Google, Baidu, Alibaba, ByteDance, everyone competing on whose model was strongest, who had more parameters, who topped benchmarks. The model layer is the "chip" of the AI era—certainly critical.
But as model capabilities converge—the gap between GPT-5.6, Claude Opus, Gemini 3, Qwen3.8, and DeepSeek V4 narrowing in many scenarios—what truly determines user experience and business value is shifting from "the model itself" to "how you use the model." This includes model routing, prompt engineering, agent orchestration, tool use, cost optimization, monitoring and observability... collectively known as the "AI middle layer."
The middle layer war is only just beginning. And OpenRouter is just the first shot.
We can expect more M&A and consolidation in this space:
Cloud providers won't sit idly by. AWS, Azure, and Google Cloud already build model routing and unified APIs themselves—Amazon Bedrock, Azure AI Studio, and Google Vertex AI are essentially official versions of OpenRouter. With Stripe entering as a third party, cloud providers will accelerate their own middle-layer strategies and may acquire more vertical AI tooling companies.
Foundation model companies will play defense. Neutral third parties like OpenRouter are a double-edged sword for model providers—they bring traffic, but they also erode lock-in. If you use OpenRouter, you can seamlessly switch from GPT to Claude tomorrow, and model vendors' moats shallow. Expect OpenAI, Anthropic, and others to launch more "bundled" features and incentives to keep users in their ecosystems, preventing the middle layer from "hijacking" the relationship.
Startups still have opportunities. OpenRouter being acquired doesn't mean the middle layer war is over—it validates the category. In more vertical slices—agent orchestration, AI monitoring, cost optimization—new startups will emerge, possibly growing into the next OpenRouter.
Of course, OpenRouter under Stripe faces challenges too. The biggest is neutrality. Before the acquisition, OpenRouter was completely neutral—it recommended whichever model was best, favoring no one. But after being acquired by Stripe, will other payments companies and cloud providers worry about Stripe "tilting the scale"? Will model vendors worry about their weighting on OpenRouter being manipulated? That trust question is something Stripe needs to address carefully.
Who Controls Routing Controls AI's Throat
Let's zoom out.
The history of the internet is, in a sense, a history of "routing power" struggles.
In the early internet, routing power sat with ISPs—whichever carrier's network you connected to determined your path. Then search engines arrived, and routing power shifted to Google—what you searched for and what results you saw, Google decided. Then in the mobile era, app stores controlled routing—what apps you could download and what got recommended, Apple and Google decided. Then in the social media era, Facebook and Twitter controlled news feed routing.
Every transfer of routing power was accompanied by massive wealth transfers and power shifts. The companies that controlled routing invariably became the giants of their era.
What is routing power in the AI era? When you say "book me a flight," "write me a report," "analyze this data"—which AI handles this request, which model it uses, which workflow it follows—that decision is routing power in the AI era.
Right now, this power is still fragmented. Use ChatGPT, and routing sits with OpenAI. Use Claude, and it sits with Anthropic. Build your own app against APIs, and you hold routing yourself. But as AI applications grow more complex, call more models, and cost optimization becomes critical, more developers will hand routing power over—to specialized platforms like OpenRouter.
Stripe saw this. So it spent $7 billion to buy the first scaled routing platform of the AI era. It isn't betting on OpenRouter's revenue or customer count today—it's betting that over the next five years, the vast majority of AI requests will pass through a routing layer, and it intends to be that layer.
"In every era of the internet, the ultimate winners aren't the people building roads—they're the people standing at the intersections collecting tolls. Stripe wants to be the AI era's toll collector." — An Internet Veteran's Observation
Of course, this war is just beginning. OpenRouter is the current leader, but it's far from the endgame. Cloud providers have natural infrastructure advantages. Foundation model providers have ecosystem lock-in. Startups have vertical depth. Who ultimately controls AI routing is still anyone's guess.
But one thing is certain: the $7 billion price tag has set the bar for the AI middle layer. It's no longer a "small tool" or "API forwarder"—it's a tens-of-billions infrastructure category. More money, more companies, more talent will flood into this space.
While everyone was watching foundation models leapfrog each other, Stripe quietly bought a tollbooth upstream of the models. That's a hell of a chess move.
See you tomorrow.
