Author: Charlie, Partner at Generative Ventures

In the past two years, Stripe's acquisition spree has dazzled the entire Fintech industry.

It acquired stablecoin infrastructure company Bridge and crypto wallet development platform Privy.At the beginning of the year, it completed the acquisition of usage-based billing company Metronome. After summer, Stripe and Advent International jointly made an acquisition offer of more than $53 billion to PayPal. A few days later, it was reported that it was in talks to acquire OpenRouter, an AI model aggregation platform with a potential valuation of $10 billion.

Individually, each transaction seems to belong to a different sector. Metronome is billing, Bridge is stablecoin, Privy is wallet, PayPal is consumer payments, and OpenRouter is AI infrastructure. However, when put together, it becomes clear that Stripe didn't just buy five companies, but rather five different forms of "entry points into economic activity." How businesses calculate revenue, how developers invoke models, how agents purchase services on behalf of users, how consumers identify themselves and authorize payments, and ultimately, through which assets and networks funds are settled—Strip is connecting these previously fragmented links piece by piece. In the past, Stripe's ambition was to increase the GDP of the internet. Now, it seems to believe that simply processing payments after GDP is generated is not enough. It hopes to enter the transaction formation process earlier, and even participate in defining what constitutes a "transaction" in the future internet. From Metronome to Orb: Payment Companies Vie for "Pricing Power" Stripe's acquisition of Metronome and Adyen's acquisition of Orb can almost be seen as two moves in the same battle. In January 2026, Stripe completed its acquisition of Metronome. Less than six months later, in June, Adyen announced its acquisition of Orb for approximately $335 million in cash. Both companies are targeting usage-based billing, which is infrastructure that charges based on actual customer usage, number of calls, data consumption, or business results. These types of products used to seem like just a sub-module within a SaaS financial system. Traditional SaaS pricing was relatively simple: a company would buy 100 accounts, each paying a fixed monthly fee. Stripe Billing, Zuora, or other subscription management tools only needed to manage plans, renewals, discounts, and invoices. AI has changed this logic. When an enterprise customer uses an AI product, the cost may depend on the input tokens, output tokens, model type, inference time, GPU consumption, cache hit rate, number of tool calls, and even whether the task was successfully completed. The cost structure for the same customer can differ across dates, features, and models. Therefore, billing is no longer about generating an invoice at the end of the month, but has become a real-time data processing system. It first collects hundreds of millions of usage events, then transforms these events into billable items based on commercial contracts. It handles tiered pricing, minimum spending requirements, prepaid limits, committed usage, overage charges, and customer-defined terms, while ensuring that sales, finance, product, and the customer all agree that the final figures are accurate. Metronome and Orb address precisely this complexity. The common logic behind the two acquisitions is not difficult to understand. Payment addresses how money is received, while Billing addresses exactly how much should be received. Without the former, companies cannot complete transactions; without the latter, companies cannot even accurately describe the transaction. However, Stripe and Adyen's starting points are not entirely the same. Stripe's acquisition of Metronome is more like strengthening its already very large Billing product line. Stripe's original Billing products initially revolved around the subscription economy, but as AI companies became one of Stripe's most important new customer groups, the traditional subscription architecture gradually became unable to cover real-time, high-frequency, and multi-dimensional usage billing. Instead of continuously adding complex features to its existing system, Stripe directly acquired an architecture redesigned from the perspective of usage events. Adyen's acquisition of Orb, however, has a greater strategic significance. As I mentioned in my previous article about Adyen's acquisition of Orb, this was a rare external acquisition in Adyen's history. In the past, Adyen had an almost obsessive commitment to organic growth (opposing M&A), choosing to build most of its core payment platform, acquiring capabilities, risk control, and global localized network in-house. The fact that it's now willing to pay $335 million to Orb demonstrates that usage-based billing is no longer a secondary feature that can be slowly scheduled, but rather a crucial entry point for payment platforms entering the AI ​​era. The two companies are not just competing for billing revenue, but for the "right to set prices." When a payment company can only see the final payment amount, it is the executor at the end of the transaction chain. When it simultaneously controls usage events, pricing rules, customer contracts, and payment results, it begins to understand how a business actually generates revenue. Whoever controls this layer can more easily move upwards into revenue analysis, contract management, and financial automation, and downwards into payments, taxes, treasury management, and credit. This is also why the acquisitions of Metronome and Orb are more important than they appear. They are not two similar billing tools, but rather the outposts for Stripe and Adyen in their battle for the operating system of enterprise business models. OpenRouter: Why the AI ​​Relay Station Suddenly Worth $10 Billion Of these acquisition rumors, OpenRouter is the most likely to be discussed, and the one most worthy of further exploration. OpenRouter currently offers over 400 models and connects requests to more than 70 model or inference providers. Developers can top up their accounts in one place, switch models using the same API, and automatically select routes based on price, speed, context length, availability, and specific tasks. This intermediary model initially attracted attention for its ease of access. Some models don't support local bank cards, some services have regional restrictions, and some small teams are unwilling to manage a dozen or so API accounts separately. Therefore, the intermediary takes on the roles of account opening, topping up, currency conversion, calls, and billing aggregation. However, OpenRouter's value goes beyond just "top-up." A mature AI transit hub actually plays four roles simultaneously: a model marketplace, an API gateway, a smart router, and a clearing platform. First, it reduces the cost of model switching. Previously, after an AI application selected OpenAI, migrating to Anthropic or an open-source model required modifying the interface, testing the output, and handling the rate limiting rules of different vendors. A unified API turns models into relatively standardized commodities, allowing developers to quickly switch based on performance and cost. Second, it handles vendor routing. The same model may be provided by multiple inference service providers, with differences in price, latency, geographical coverage, and stability. OpenRouter can automatically switch when a provider experiences a failure or rate limit, essentially performing real-time scheduling between models and the computing power market. Third, it aggregates demand. Individual developers find it difficult to negotiate prices and capacity with large model companies, but by aggregating the token consumption of a large number of developers, the platform gains bargaining power similar to a cloud computing reseller or payment aggregator. Finally, it controls transaction data. What applications are growing, which models developers are migrating from, how many new calls a price drop will bring, and which tasks prioritize quality over cost—this data may be closer to the real needs of the AI ​​industry than public benchmarks. The biggest difference between OpenRouter and traditional API Marketplaces lies in the extremely real-time and substitutable nature of model calls. A single request might cost only a few cents or even less, but billions of requests occur daily. It's not an occasional software purchase, but a micro-resource allocation performed with each token generation. Therefore, the core business model of the transit hub is: It doesn't own the most expensive models, nor does it bear the huge capital expenditure for training the basic models. Instead, it establishes a unified interface between the highly fragmented supply and demand sides and charges a small fee based on traffic. Does this sound familiar? This is almost exactly the business model of early payment aggregation platforms. Stripe doesn't issue Visa or Mastercard cards, nor does it operate with every bank. It encapsulates complex payment networks, acquiring institutions, local payment methods, risk control, and merchant systems into an API, then charges a fee for each transaction. OpenRouter doesn't train GPT, Claude, or Gemini. It encapsulates model providers, GPU inference nodes, pricing, routing, and developer billing into an API, then charges a fee for each model consumed. OpenRouter CEO Alex Atallah even likens his company to Stripe in the world of AI models. Now, Stripe seems poised to turn this analogy into an acquisition. Why Does Stripe Need an AI Intermediary? To understand the significance of OpenRouter for Stripe, we must first understand the revenue structure of AI products. Today, many AI companies face the same problem: their revenue and costs are not fixed. For each task a user completes, the application invokes one or more models. Each invocation incurs token costs, but how the application charges customers can be through subscriptions, pay-as-you-go, points, task results, or a hybrid package. Enterprises must first obtain usage data from OpenRouter or model vendors, then calculate the amount due through Metronome, and finally complete payment through Stripe. In other words, OpenRouter, Metronome, and Stripe occupy different positions on the same chain: OpenRouter determines which model to use for an AI request and how much it costs; Metronome translates the call volume and contract rules into customer prices; and Stripe ultimately collects payment and completes the settlement. Once integrated, Stripe has the opportunity to control the entire closed loop from inference costs to commercial revenue. For example, an AI customer service company might promise to charge businesses based on the "number of issues successfully resolved." To complete a customer service task, the backend needs to invoke three different models, use search tools, and access external databases. OpenRouter can record the actual consumption of the underlying models and tools. Metronome can calculate the final bill based on success rate, usage, and enterprise contracts. Stripe can deduct payments from enterprise customers, process taxes, and handle cross-border settlements. Bridge allows global customers or AI agents to complete near-real-time fund transfers using stablecoins. At this point, Stripe is no longer just processing that final credit card payment. It begins to manage the entire economic process of an AI service, from production costs, resource allocation, price formation to payment settlement. There's also a longer-term strategic significance here. In the future, the entity initiating the purchase may not necessarily be a person; it could also be an Agent. An AI Agent, in order to complete a task, will automatically invoke models, purchase data, rent computing power, access paid APIs, and pay other Agents. These transactions are smaller in amount, more frequent, and won't tolerate the complex Checkout pages of the credit card era. Payment must be tied to the call itself: the request is approved, the service is executed immediately, and the fee is settled based on actual consumption. OpenRouter is already at the entry point for this kind of machine-generated transaction. Stripe, on the other hand, has capabilities in billing, payment credentials, risk control, taxation, and stablecoin settlement. Combined, a single API request can become a complete commercial transaction. This is why OpenRouter is far more significant to Stripe than simply acquiring a rapidly growing AI company. Stripe doesn't need to compete with OpenAI or Anthropic in training its base models. Model layer training is capital-intensive, technologically rapidly evolving, and leadership positions are unstable. The best model today may not still be the best six months from now. A smarter approach is to control the routing and settlement layers that all models must pass through. Regardless of whether OpenAI, Anthropic, Google, or some yet-to-be-released open-source model ultimately prevails, as long as developers still need to compare models, control costs, and ensure stable calls, OpenRouter will benefit from industry growth. This is similar to Stripe never betting on which bank consumers will ultimately use. It chooses to stand between banks, bank cards, merchants, and consumers, earning revenue from the infrastructure that drives the growth of internet commerce. PayPal's board's "disapproval" is more like a bid. PayPal is the most traditional and easily understood asset in this deal. Stripe and Advent International proposed a price of $60.50 per share, valuing the company at over $53 billion, a premium of approximately 28% over the share price before the announcement. The two bidders planned to each hold a 50% stake and had secured approximately $50 billion in bank financing commitments. PayPal's board subsequently deemed the offer insufficient and expressed concerns about regulatory and financing complexities. However, it's important to distinguish between "deeming the offer insufficient" and "formally rejecting the deal." As of July 27, the latest reliable reports indicate that PayPal has not yet formally responded to the proposal. The board is comparing two paths: accepting a certainty premium offered by an outside buyer, or continuing to bet on the new management's restructuring plan to restore growth. Therefore, this is more like the beginning of negotiations than the end. The board was certainly unwilling to accept $60.50 in the first round. PayPal's stock price had exceeded that level at the end of 2025, and it had generated approximately $5.6 billion in cash flow over the past twelve months. If it accepted the offer immediately, the board would find it difficult to explain to long-term shareholders why a company with 400 million users, Venmo, Braintree, a global payment license, and a vast merchant network was worth only a fraction of its market capitalization at the height of the pandemic. But PayPal's problems are equally real. Its brand Checkout is being squeezed by Apple Pay and Google Pay, Venmo has a large user base but has not been fully monetized, and Braintree contributes a large amount of transaction volume but is a processing business with relatively limited profit margins. PayPal has many assets, but these assets haven't formed a sufficiently compelling growth loop. Stripe is eyeing this mismatch of "strong assets, weak organization." Stripe excels in products, developers, and merchants, but lacks a truly global consumer identity and wallet network. PayPal and Venmo can fill this gap. PayPal has a consumer entry point, but lacks Stripe's growth momentum within AI companies, the developer ecosystem, and modern internet platforms. True synergy isn't simply about adding up the payment volumes of the two companies; it's about connecting PayPal's consumer identity with Stripe's merchant, billing, and agentic commerce infrastructure. However, two major obstacles remain. The first is price. The market widely expects Stripe and Advent to have room to raise their bids. Bidders have already raised approximately $17 billion in equity and $50 billion in debt financing, and PayPal's board clearly knows they can still increase their offer. PayPal's earnings report on July 28th will be key to the next phase of the negotiation: if core Checkout continues to deteriorate, the board will face increasing pressure to sell; if performance improves significantly, Stripe may have to offer a higher price. The second factor is regulation. The merged group could process approximately $3.7 trillion in payments annually, and regulators will certainly scrutinize the concentration of the online payments, merchant acquiring, and wallet markets. Reports indicate that Braintree is a potential divestiture asset because it has the most significant overlap with Stripe's payment processing business. Therefore, the most likely next step in this deal is not for PayPal to immediately agree or completely reject it, but rather for Stripe and Advent to raise their offers while providing clearer solutions regarding asset divestiture, financing structure, and regulatory commitments. Stripe is assembling a new operating system for the internet economy. By putting PayPal, OpenRouter, and Metronome together, Stripe's strategic path becomes clear. OpenRouter handles resource allocation. Metronome handles metering and pricing. Stripe handles payment and revenue management. PayPal and Venmo handle consumer identity, wallets, and authorization. Bridge and Privy handle stablecoin settlement and on-chain accounts. These correspond to several core questions in the future digital economy: Who is buying? What resources are being purchased? How much is actually being used? How much should be paid? What credentials are used for authorization? Through which network is the final settlement of funds? Traditional payment companies only answer the last question. Stripe is now attempting to answer all of them. This also means that Stripe's competitors are changing. In the traditional payment market, its competitors are Adyen, Checkout.com, Worldpay, and PayPal. In the era of AI economy and Agentic Commerce, its potential competitors will include OpenAI, Google, Amazon, Apple, Coinbase, and any platform capable of simultaneously controlling identity, computing resources, business access, and fund flows. Stripe's advantage is neutrality. It doesn't manufacture phones, operate a search engine, or need to protect any underlying model. It can connect different models, different wallets, different payment methods, and different settlement networks. Its risks also stem from the same place. With each additional layer of infrastructure it controls, Stripe becomes less like a simple, asset-light payment technology company and more like a super platform that needs to coordinate consumer business, enterprise software, financial licenses, AI routing, and blockchain networks. The real difficulty in acquiring PayPal wasn't signing the $53 billion check, but whether Stripe could manage the massive legacy systems and organization without disrupting PayPal's user and merchant relationships. The real risk of acquiring OpenRouter isn't whether there's demand for AI intermediaries, but rather whether large model companies will restrict intermediaries, and whether enterprise clients will ultimately choose to contract directly with model suppliers. However, at least from a strategic perspective, Stripe's transactions are not haphazard. The business model of the internet is shifting from buying software seats to buying real-time usage; from human clicks on checkout to agents automatically initiating transactions; from single model calls to dynamic routing between multiple models and computing power providers; and from bulk credit card settlements to instant settlements between stablecoins and machines. Each of Stripe's acquisitions preemptively addresses a key shift. Ten years ago, Stripe condensed a complex payment network into a few lines of code. In its next phase, it aims to streamline model calls, usage-based billing, identity authorization, and fund settlement into a single interface. At that point, Stripe will no longer be handling payments from internet companies; it will be dealing with the internet economy itself.