Why did Stripe spend $10 billion to buy OpenRouter? Is it building the Visa of the AI era?

Hello everyone, welcome to Money in Motion. This is a podcast focused on how money flows. We are not discussing which concepts are the hottest, but rather how money is being moved, settled, and trusted in the real world.

In this episode, we have Jordan, co-founder and CEO of AIsa, joining us to discuss the underlying logic behind Stripe's recent acquisition of OpenRouter. We will explore the token model, routing business, API resource trading, agent marketplace, and the position Stripe truly wants to occupy in the AI era. We want to invite a frontline practitioner standing at this intersection to help us break it down.

Jordan defines his company AIsa as "Alibaba for AI Agents": it is not merely a payment project and does not wish to be compressed into the narrative of Agentic Payment. AIsa is more focused on where agents need to go to purchase data, tools, models, computing power, and other API resources when they transition from "thinking" to "acting," and how these resources can be packaged into skills that agents can directly call upon.

Guest: Jordan (Co-founder & CEO of AIsa)

Hosts: Will & Yuki (Co-founders of Money in Motion)

The views expressed in this article are those of the guest and do not represent the views of the guest's organization or this media outlet.

1. What exactly did Stripe buy?

As soon as this deal was announced, almost everyone interpreted it as "a payment company entering AI." Jordan's first reaction was to clarify the roles involved—because once the roles are clearly defined, you will find that what this money buys is quite different from what people think.

1.1 Clarifying the roles

Host: A payment company spends $10 billion to buy its own customer. What do you think of this deal?

Jordan: There are several definitions that need to be clarified first.

First, what exactly does Stripe do? Stripe is a payment processor, primarily focused on merchants, particularly checkout, and it has a very complete suite of supporting services.

Second, what is OpenRouter? In the Chinese context, the intuitive first word is definitely "transit station"—aggregation, routing. But from the perspective of payments, OpenRouter is also merchant-facing: the model calls from over 400 model vendors are a service in themselves, a resource that developers and businesses need to procure.

So the question arises: Stripe is already its payment processor, why spend another $10 billion to buy it?

First of all, this price is certainly not calculated based on secondary market valuation standards. OpenRouter's disclosed ARR is only $50 million, and $10 billion corresponds to nearly 200 times PS, which is an extremely high valuation. If you benchmark Manus—it was acquired when its ARR just surpassed $100 million, and the corresponding multiple was only in the teens to twenties. So, 200 times PS is not derived from a valuation model.

From my perspective, it resembles two things: one is defensive, and the other is strategic synergy.

1.2 The God’s-eye view of tokens

Jordan: If you break down the value that OpenRouter brings, it is definitely not just a model transit.

It is a blind testing ground before many large models are released, a place for ranking across various dimensions. There is another layer that few people see—it is currently the only place in the world that can see model call data from a God’s-eye view.

What models did I call to complete a task? How were the parameters of those models configured? How was the fallback set? Which models perform best in which task scenarios? You cannot see this data anywhere else.

On another level, from a developer's perspective: which development teams have the highest and fastest growth in call volume on OpenRouter? This is the data that many investment institutions are eager to acquire.

Stripe is now one of the leading super unicorns in the payment field, alongside Visa and Mastercard. If it gives up this opportunity and hands this perspective over to others, it would be a significant strategic loss for them—the cost of making up for it later would be extremely high.

When Meta spent $19 billion to acquire WhatsApp, the team was only a few dozen people. Looking back today, that $19 billion is worth far more than the price at that time in terms of Meta's overall strategic layout.

I believe this is more of a defensive move.

1.3 Its relationship with agentic payment is not that significant

Host: Looking at the previous round makes it clearer. The $1.3 billion valuation on May 26 was led by CapitalG, with Nvidia participating. Google's logic is: by investing in a traffic distribution window, you gain a God’s-eye view; Nvidia's logic is more straightforward: the more developed the distribution platform, the better the chips sell. Both make sense, but they are related to token distribution itself.

Stripe is different. It does not distribute tokens on this chain. Its earlier position was to collect payments—when it started making AI-related products in 2022 and 2023, the vast majority of startups used Stripe, including the collection and processing of payments for all major models, which are basically all on Stripe. It has already established its position.

So the real question may be: is its next step to upgrade from "collecting payments for AI companies" to "controlling AI consumption"—using routing, measurement, and settlement as a complete infrastructure? If tokens become a unit of economic consumption for machines, then mastering the routing and settlement of tokens could simultaneously master the Visa and Cloudflare of the AI era.

Jordan: To answer this, we first need to define agentic payment. In my understanding, true agentic payment is when agents can autonomously initiate payments—what is called A2A is still very early.

Stripe's only move in agentic payment is the launch of MPP (Machine Payment Protocol). However, the volume on it is actually not large; it has not reached that stage.

So first, this deal is not significantly related to agentic payment. Second, it is already the payment processor for OpenRouter.

In my view, this is more of a move on the merchant side. Stripe's core is always to serve merchants, and the token service for large models will be the largest merchant category in the future.

If you look back at how Stripe grew, it is a typical to Developer product, growing alongside that batch of unicorns from YC, and is now the first choice for almost all startups.

From this perspective, what it wants is the entry point, the payment processing on the merchant side of the AI economy. And in the merchant side of the AI economy, so far, the consumption of tokens is definitely the largest, followed by data and tools. It first captures this layer, which is the core motivation.

1.4 From $159 billion to $1 trillion

Host: Now looking at its actions over the past two years—acquiring Bridge, Privy, Metronome, and possibly negotiating for OpenRouter, while also wanting to buy PayPal. From your perspective, what exactly does it want to do?

Jordan: My view is also limited, from the perspective of payments: it has lost the real opportunity to create a card organization.

Visa and Mastercard are the core underlying rails for all money movement in the world. Stripe has already missed the chance to become a window for such a network.

If you look at all its actions, aren’t they all centered around this?

First, Tempo, its own public chain. Starting from the merchant side, facing developers, and then expanding into the stablecoin ecosystem, where its core—in my view—is still hoping to bypass card organizations and integrate the settlement process into its own ecosystem and public chain.

Bridge acquired compliance licenses and compliance channels, doing on-ramp and off-ramp. Privy is a digital currency wallet side, a relatively large entry point, also considered a user-side entry. Recently acquired Metronome does billing, and in fact, it has become the largest billing service provider for many large AI companies—including OpenAI—relying on it for internal pricing design. Now bringing OpenRouter in can significantly amplify Metronome's service volume.

PayPal is the user-side entry in Web2. I remember seeing its data, with over 400 million C-end wallets—something Stripe does not have. Its Link product is still far from PayPal's ecosystem. With its own public chain, obtaining the ticket to the stablecoin ecosystem, and acquiring the largest wallet system in Web2, its entire ecosystem would be quite complete.

Stripe is currently valued at $159 billion. Although it cannot create the payment rail that Visa took over sixty years to build, once it gathers all these ecosystems, it can aim for $500 billion to $1 trillion, which is not a dream. It is certainly benchmarking against Visa and Mastercard—each of which is currently valued at around $600 to $700 billion.

As for why it is so active in AI—because AI represents the largest flow of funds. Today, on the merchant side, payment collection, aside from e-commerce and OTA, is AI. It captures AI companies, and every AI company has to pay large model vendors. So, buying OpenRouter first, this logic is not hard to deduce.

Host: Its recent actions have been very numerous, and a large part should also be to create momentum in the capital market—AI is so hot right now; if you don’t ride that wave, what else can you ride? So those focusing on agentic payment need to distinguish: its engagement with AI is different from the point of agentic payment.

2. The business logic of model transit stations

If the strategic value of this position is so great, why did OpenRouter choose to sell at this point? The host raised a question from an entrepreneur's perspective: isn’t achieving shelling to the extreme a moat in itself? Jordan replaced that term and provided a rather pessimistic answer.

2.1 Is shelling to the extreme a good business?

Host: I've been thinking about something lately: for entrepreneurs in the AI era, reaching the extreme of shelling out might itself be a good business.

Stripe and OpenRouter are essentially very similar. Stripe doesn’t own a bank network; it packages all these things into a stable interface for everyone to use. OpenRouter also doesn’t train models; it provides all the underlying capabilities and a unified interface. To some extent, Stripe provides a payment API, while OpenRouter provides a model API.

Now everyone says that AI entrepreneurs have a hard time—whatever you do, you fear that a major model update will wipe out your project. Does that mean that reaching the level of infrastructure in shelling out might be a moat?

Jordan: Let me tweak the term "shelling out" a bit.

Stripe has done a lot of processing based on card organizations and all these banking channels, making it extremely simple for real developers to integrate it. I call this process building blocks. Especially in the overseas and North American startup ecosystem, particularly in the SaaS ecosystem, this is a very typical building block process.

But we can’t simply conclude that you can build with any block or step into any ecosystem.

2.2 Three Things to Get Right

Jordan: A big opportunity always has several factors.

First, the market must be large enough. Tokens are definitely a sufficiently large market.

Second, timing must be precise. In the first two or three years after OpenRouter started, no one thought this thing was remarkable—at that time, there weren’t so many large models, nor was there a differentiation between Chinese and American models, open-source and closed-source ecosystems; such a big opportunity was not visible, yet they had already started.

You should ask Alex if he saw today’s landscape back then? I don’t think so. I believe he was more continuing the many insights he accumulated from his NFT trading platform OpenSea—just building something on the side, maybe there would be an opportunity later. I don’t think he had that strong penetrating power and visionary insight. He hit a very good timing. Think about it, how many versions of models have iterated from 2023 until now? Almost every couple of weeks there’s a new version. When he did this, it was a completely different era. As a result, he actually became the first brand in model routing and aggregation.

Third, what ecosystem you are in. Can OpenRouter succeed in North America, and can it succeed in the Chinese ecosystem? I don’t think so.

2.3 China Won't Produce an OpenRouter

Jordan: Recently, many people have been comparing Silicon-based liquidity that is submitting for listing, but you will find many differences at the underlying level.

From the infrastructure perspective, overseas, NVIDIA basically dominates at the GPU level; but domestically, because many are open-source models, there will be a large number of different chip manufacturers. Furthermore, there’s also a differentiation between open-source and closed-source, where the deployment cost of open-source models is relatively higher.

The underlying infrastructure is very different, so it’s hard to say that China can produce a project of the scale of OpenRouter.

2.4 A Business with Low Barriers and Thin Margins

Jordan: Back to building blocks itself—the barriers are actually very thin.

This is something that hasn’t been discussed much: why did OpenRouter choose to seek a sale at this time?

Some joke that Alex made a precise turn to AI at the peak of NFTs, and now he’s making a precise sale just before AI reaches its peak. Jokes aside, you can indeed make a judgment.

The business model of OpenRouter has at least thousands of companies doing it globally. The threshold is low, the barriers are low, and the margins are low.

Its revenue comes from two parts. One part is the recharge fee, about 5%—but of that, about 2.9% goes to Stripe, which then distributes it to card organizations, leaving only a tiny fraction for itself. The other part is BYOK, where you bring your own key, and after exceeding a certain amount, you have to pay a 5% fee, which is also a very thin business.

Editor's Note: BYOK (Bring Your Own Key) refers to developers using their own model vendor keys, with fees paid directly to the model vendor, not passing through OpenRouter; OpenRouter charges a nominal fee based on "what it would have cost." The official pricing page shows that the monthly list price for inference within the free limit is $25,000, with a 5% charge for excess, and the enterprise version has a free limit of $200,000. Money doesn’t pass through but is charged based on nominal transaction volume, which corresponds to the assessment fee model of card organizations in the payment industry, rather than the acquiring model.

Even if it passes the costs through, the revenue is only $50 million—while its daily token consumption is about 40 to 60 trillion. This doesn’t count as a good business.

Looking at the landscape: every cloud vendor can create its own OpenRouter. Amazon has Bedrock, Google has Vertex, Microsoft has AI Studio, all attacking from different angles.

So at this time, while someone is interested in this asset, he made a precise judgment to sell it off first.

Three, The Token Model is Logistics in the AI Economy

If the threshold for model transfer is so low, where is the real business for such platforms? Jordan gave an analogy to e-commerce: models are not goods; models are logistics.

3.1 Agents Don’t Come with Model Names

Host: Besides models, your platform has skills, APIs, and other services. What do you think the real services on such platforms will be? Surely it’s not just the tokens of large models.

Jordan: This question is particularly well put.

Many people open our website, see models, and naturally understand us as a model transfer and aggregation platform. During the last round of financing, some investors asked: what’s the difference between you and OpenRouter?

I used an example at that time. Amazon is the world’s largest e-commerce marketplace; it has logistics; JD has logistics, Alibaba also has logistics. But would you say Amazon is a logistics company? Is Alibaba a logistics company? No. Completing the transfer of goods is the most basic capability.

Similarly, in the AI economy, models are gradually being commoditized, and most people judge that the price of tokens will drop tenfold or even more—it’s just a commodity.

You can think of models as logistics in the e-commerce economy: they are the logistics in the AI economy, the most basic infrastructure.

So for AIsa, what we value is not to create a model aggregation capability.

Our starting point is different: the customers served on the AIsa platform are inherently agents—not people, not consumers, not a certain enterprise. From day one, we built this product for agents.

So when an agent comes to a platform, what do they want to procure?

They come with a clear intention. Their first reaction when calling resources is not "I want to call a certain large model," but "I found that AIsa can provide direct API interfaces similar to SimilarWeb and Semrush for website traffic analysis." An MCP can access all resources on the platform, and skills are configured according to scenarios. Models are the last layer by default: if you want to generate images, we have image models; if you want videos, we have video models.

Many of the first 10,000 agent users on the platform came from the OpenClaw (little lobster) ecosystem—they came because we packaged the Twitter API into a skill, completed all OAuth authentication, and turned it into on-demand calls. Users directly tell the agent: "Help me generate a tweet based on this topic." The agent generates the article, matches images, and even videos; as for which model was called at the underlying level, that’s a default action.

This is the logic of how agents procure services on AIsa: they come with clear intentions, call underlying APIs through skills, and the model layer is completed by default. This is the biggest difference.

3.2 The Wall Won’t Suddenly Collapse One Day

Host: I encountered a specific problem when I was creating SEO-related skills. I went to buy a Semrush API account on Xianyu and used a crawling method, which was very inefficient. As an individual, you can’t buy APIs aimed at the business side.

So I’m curious: for players like SimilarWeb and Semrush, their profit model has always been aimed at charging enterprises for many years. If they start supporting agent payments, is this a supplement or a detriment to their existing business? At least officially, they haven’t opened it up for agents to use directly. How did you convince them?

Jordan: First of all, we didn’t try to convince them.

These SaaS vendors transforming into data vendors is an inevitable result after being educated by the market. The collapse of the wall won’t happen suddenly one day; it involves many small actions. These data vendors and digital resource providers have already realized this issue.

If you usually pay attention to this field, you should often see a viewpoint: in the future, the number of agents on the internet will definitely exceed that of humans.

When agents become real economic entities—the narrative of creating wallets for agents comes from this. But just creating a wallet for agents, so what? Giving them private keys and authorization is easy. The real problem to solve is whether these service providers are willing.

Isn’t this a conflict of interest? They have to move their vested interests. But we will first assume they will move. Because if the wall has already started to collapse, if you don’t take action, you will completely fall behind this era. If Semrush doesn’t act, and SimilarWeb has started providing services to agents, will you keep up? This is always the result of competition.

Moreover, many enterprises have already begun adopting AI workflows. If you need to monitor competitor websites’ traffic, why should a person keep an eye on it? It’s definitely handed over to agents. When agents take these actions, shouldn’t they open API permissions for them? It’s very logical.

3.3 The Most Valuable Part of SaaS is Data, Not UI

Jordan: If you let agents use browsers to click around, it won’t work. This market has already given the answer—look at how SaaS companies have plummeted over the past year.

The core reason is that SaaS is human-made. The essence of SaaS has two layers: the underlying database and the upper layer, which is the UI. The UI provides you with various buttons; you click and generate a 30-day sales report. But for agents, why not just access your database directly? I don’t need your UI layer.

Therefore, to get to the truth—in the age of agents, the most valuable part of a SaaS company is your data and the know-how embedded in your clients' workflows. So why not open this part up as an interface for agents to use?

Thus, in the early stages of this track, the real competitive advantage lies in the ability to leverage the resources of these service providers, rather than how flashy your product is. Agents come to you because you have the resources they need. First, resources they cannot manage themselves; second, resources that are more cost-effective or advantageous than doing it themselves. The logic is roughly like this.

4. AI Agents as Alibaba

The shelves are set up, and suppliers are gradually coming on board. The next question is: what will this marketplace ultimately look like? And when all buyers are agents, who will handle disputes?

4.1 From Marketplace to Credit

Host: Can you share a long-term vision—what does AIsa ultimately want to achieve?

Jordan: It’s simple; as I mentioned at the beginning: Alibaba for AI agents. In terms that are easier to understand overseas, it’s the Amazon everything store for AI agents.

Everything you want is here; we have payment gateways and a clearing and settlement system; next, isn’t it just the “Alipay” part that we can also provide for agents?

Because once there is a flow of goods—previously called logistics, now goods flow—it already involves the flow of money; and with the flow of money comes credit. That’s our next step. This is why Ant Financial exists: you will definitely provide lending and banking for agents.

Some people are already doing this. A couple of days ago, I saw a company whose founder was an early co-founder of Circle. After several iterations, they are now a compliance bank for agents—accounts need to undergo KYC and KYB, with FDIC insurance, usable by both merchants and individuals.

So if you ask about the business model, we are here to do this.

4.2 Code is Law: Disputes in the Agent World Will Be Much Simpler

Host: Let’s delve deeper into the Alibaba analogy. Alibaba provides not just products; it also offers supply and demand, search, payment, credit fulfillment, as well as dispute resolution and platform governance.

In the future, when agents use “Alibaba,” and buyers are autonomous decision-making agents, who exactly are the sellers? What happens if issues arise? I believe the need for “humans” today largely stems from the need for someone to take the blame. Once we move to agents, what issues will arise in terms of disputes and after-sales service, and how will platform governance be affected?

Jordan: We already have this.

From a product perspective, the first layer is the marketplace, connecting supply and demand: one side is agents, and the other side is the resource providers that agents need—data resources, tool resources, GPU, storage, computing power, all of which are considered native resources they require.

As for chargebacks, a part that cannot be avoided in traditional payments, we believe in the world of agents, it will be much simpler than in the human world.

Think about it, what problems do e-commerce merchants deal with daily? I don’t like this product; I regret buying it; the logistics delivered it damaged; my mood changed, and I want a different color.

But for agents, it’s a world of code. Code is law.

I clearly describe in a markdown file what my interface is, what the latency is, and what the quality is; you can even initiate a small call first to check the parameters; once confirmed, you accept it, and the transaction is complete. Agents won’t say, “Sorry, I’m in a bad mood today; I want to change.” No way.

In the agent marketplace, such human chargebacks and disputes arising from carbon-based life forms' emotions will be significantly reduced.

Of course, new issues will certainly arise: illusions, malicious actions, and harmful bots. These will need new mechanisms and technologies to address.

We have introduced a large number of algorithms on the platform. Besides giving agents an identity—this is a topic of much discussion in the industry—the more critical aspect is how to determine if it is a legitimate agent. When a new agent comes on board, you have no way to assess its intentions and origins unless you conduct a “KYA” for each agent. But if the platform has accumulated enough data, and this agent has initiated multiple calls, you can fully track its behavior. This is the value of the platform.

Thus, we started early on to introduce algorithms to learn these behaviors and give them overall scoring—laying the groundwork for future credit and even banking.

5. Where is the Real Money in Agentic Payment?

With shelves set up and credit established, how does money flow? This is the toughest part—it's not a prediction; it's a judgment from the front lines. And the first number it gives is exactly the opposite of market expectations.

5.1 Fiat Currency Still Accounts for 95% to 99%

Host: Previously, a report from Visa and another company stated that there were many transactions on x402, but after excluding the volume, there was actually only a small amount of real volume.

Everyone talks about agentic payment, but there are actually many ways: one is on-chain stablecoin transfers, another is using cards—card organizations are also promoting this, and there’s also wallets where users recharge and then spend. How do you see the future payment paths?

Jordan: First of all, the fiat currency world still occupies 95% to 99% of the volume.

According to Money in Motion, the flow of money is the true flow of money.

At the same time, we also believe that x402, MPP, including Visa’s VIC and Mastercard’s Agent Pay—we are early members and are very open and proactive in supporting the maturation of these future agentic payment protocols.

However, since the end of last year, when Base started charging facilitators, the real volume of on-chain transactions has shown a significant decline.

At that time, we were genuinely doing this—pushing users to adjust Twitter API, financial data API, putting transactions on-chain one by one. When V2 was launched, we also did batch settlement; those tens of millions of transactions were real transactions.

But we found that the agents we are currently serving are pure Web2 AI companies and AI startup teams, and they do not have a strong demand for on-chain settlement.

I still believe in that grand vision—future agents will carry wallets and settle transactions on a per-transaction basis through the 402 protocol. I believe in this vision. However, in the current implementation path, it may not necessarily align with the most commercial rules. We are not saying that every transaction must be on-chain.

But we can also see that the adoption rate of stablecoins is certainly much faster than that of the fiat currency world. Therefore, we strongly support various ecosystems. For example, Circle—one of the largest compliant stablecoin operators—when it launched nano payments, we were its technical partner and key launch partner, fully supporting the entire launch process. Now, if you look at the agent digital marketplace on Circle, the vast majority of resources there are provided by AIsa.

We cannot just sit there and wait—outside the x402 ecosystem, real volume can definitely be generated.

Host: There is a lot of noise in the market right now; you have shared frontline insights. Currently, many agent wallet scenarios do not seem to have emerged from real demand.

Jordan: We should focus on creating things of real value. Given the significant benefits of AI, it’s better to concentrate on creating real value first.

5.2 The Opportunity for Stablecoins Lies with Those Without Credit Cards

Host: From your perspective, will AI and crypto intersect in the future? In what way will they intersect? Or is it possible that they may never intersect?

To add some context: a couple of days ago, Circle's Jeremy published an article discussing the agentic economy. He mentioned that currently, buying tokens is to produce work; if in the future, we directly buy skills or exclusive agents to produce work, could a new labor market emerge? Skills and exclusive agents are inherently globalized, and globalization may fit better on-chain.

Jordan: Jeremy is absolutely right to make such a judgment from his standpoint.

I also believe that stablecoins are a tremendous creation—I previously worked on a wallet aimed at B-end cross-border payments using stablecoins, completely unrelated to speculation. Similarly, AI is irreversible; in the future, everyone and every business will have their own agent.

Under this premise, you will find: you are in Hong Kong, I am in Singapore, in Silicon Valley, and we naturally feel that AI has already played a significant role in daily life. However, there are still many emerging markets and underdeveloped markets where people and businesses cannot afford models at the level of Opus.

This raises a problem. You see, why do the top AI large model manufacturers in China become our investors? Because they also see this big opportunity. A large number of Chinese large models are going overseas, and Chinese tokens are going overseas; this is a given fact. Enterprises and individuals in developing countries are unlikely to use those few large model manufacturers in Silicon Valley—if I run a small shop on the street and need an agent to operate, why would I use Opus? The most suitable for them would definitely be the large models from China.

So, what payment methods will these institutions, individuals, and small and medium enterprises use to procure AI services? The vast majority of them do not have credit cards. To reach them with AI services, a payment system is necessary. We believe that the optimal solution may be stablecoins.

The answer is evident.

So in my view, the biggest opportunity for the combination of AI and crypto—specifically stablecoins—lies in individuals and small to medium-sized enterprises in emerging markets accessing the AI services they need through stablecoins. Achieving this step would already be quite remarkable.

5.3 However, AI Adoption is Still Low

Host: Currently, the adoption rate of AI itself is still low; the number of people worldwide who are willing to pay for models may not exceed 10%.

Jordan: Right, you see OpenAI claims to have 900 million weekly active users, but the vast majority are just using it for free to ask a question. To truly integrate it into your life and help you run a small to medium-sized enterprise is basically impossible—it's unaffordable.

Host: On the individual side, there is indeed change; at least domestically, both children and the elderly have started using Doubao. But the deployment of agents on the enterprise side is still very early.

Jordan: We are also seeing a large number of FDE-type organizations emerging to help enterprises implement AI, indicating that people are indeed recognizing this opportunity. Looking ahead 12 months, it will be a different landscape.

Host: Thank you, Jordan. We started this episode with a transaction that has not yet been officially confirmed, but the focus did not remain on the transaction itself. Jordan's assessment is that Stripe is not buying agentic payment; it is acquiring the entry point for AI economic entities on the merchant side—the only place globally where we can see how models are being invoked. And behind all its actions is the same goal: to bring every layer outside that network into its own hands.

However, this episode ends on another note. When an entry point is valued at $10 billion, most people and businesses in the world still do not have a credit card that can be used to purchase AI.

In EP.03, Professor Jia Hang mentioned that the next global payment network needs new user value, new profit distribution, and new governance mechanisms; in the previous episode, Rufei asked whether agents need to make money and what they would do with it. This episode, Jordan's answer is that agents do not need a wallet, but rather a place where they can buy things and a mechanism to judge whether it is worth trusting. The real testing ground for this may not be in Silicon Valley.

------ END ------

Compiled version｜Money in Motion EP.05 Dialogue with Jordan: Why Stripe Spent $10 Billion to Buy OpenRouter and Insights on Agentic Payment. This article is a transcription from spoken to written format, presented in the original dialogue style, with some content edited and merged. The views expressed in this article represent only the guest's personal opinions and do not reflect those of the guest's organization or this media outlet.

For Money in Motion, we are always concerned not with a single hot topic but with how the financial system is being rewritten—where the money comes from, the paths it takes to whom, and how new technologies and rules are changing it. If you are also interested in payments, stablecoins, AI Agents, and the next generation of financial infrastructure, feel free to follow Money in Motion on Xiaoyuzhou, YouTube, and Substack. We will continue to discuss how money will flow in the next episode.