11 Years, $11 Billion, and Then What? | A Conversation with Airwallex's Kai Wu: The AI Era and the Road to $100 Billion

👦🏻 Podcast Interview: Koji

🥷 Edited by: Crossing

🧑‍🎨 Layout: Zeoooo

🚥 I love listening to founder stories — the successes, the failures, the dramatic reversals, the ones still searching for an ending. I always find something resonant and inspiring in them.

Earlier this year, Crossing invited many of the new generation of AI entrepreneurs, who mostly shared stories of setting out from zero — ambitious beginnings full of promise. But going forward, I want to invite more founders who've already crossed the mountain, to hear them talk about the journey that got them here.

➤ This week's guest on Crossing is Kai Wu, Chief Revenue Officer at Airwallex, which just closed a new funding round last month at an $11 billion valuation. He shared with us the key decisions that took this Chinese-founded company from Melbourne to $11 billion over eleven years: turning down Stripe's acquisition offer, gritting through four years of losses to build out a global network first, and insisting on hiring local managers in every market worldwide.

These decisions struck me because they all circle the same question every founder eventually faces: When money and vision, bet and risk, are laid out before you simultaneously — which do you choose?

If you're working on something with no short-term payoff but long-term conviction, this episode will give you some concrete strength.

Beyond the story, we also talked plenty about AI. Having entered the decacorn club, Airwallex's advantages now stack onto a generational opportunity: AI is pushing global finance toward an inflection point. So this episode focuses on how AI is transforming fintech:

• Airwallex now serves major LLM companies: Moonshot AI, MiniMax, Zhipu AI, DeepSeek, and others. How are new demands — multi-model switching, tiered pricing, peak/off-peak dual rates, real-time dynamic billing — forcing financial infrastructure to evolve?

• Will the ten separate financial SaaS tools once needed for expense control, accounting, revenue recognition, reconciliation, and billing be replaced by a single Agent interface?

• After ChatGPT Finance's launch, how will the competitive landscape for "financial super-intelligent entry points" evolve?

• What exactly are Airwallex's AI financial products — Kai, AgentOS, T+0, Airi?

• Airwallex has acquired Leapfin and OpenPay, with more deals in progress. What's the core strategic logic driving this active M&A push — what key assets are they really after?

Listen on WeChat:

Listen on Xiaoyuzhou:

🎬 The video podcast is now live on Koji's WeChat Channels, Xiaohongshu, Bilibili, YouTube, and other platforms

Rapid-Fire Q&A

👦🏻 Koji

Let's start with the classic rapid-fire round to help everyone get to know you quickly. How old are you?

👨🏻‍💻 Kai Wu

👦🏻 Koji

Where did you go to school?

👨🏻‍💻 Kai Wu

Fudan University.

👦🏻 Koji

MBTI and zodiac sign?

👨🏻‍💻 Kai Wu

INTJ, Pisces.

👦🏻 Koji

Describe Airwallex in one sentence.

👨🏻‍💻 Kai Wu

We're a financial and payments platform serving global businesses.

👦🏻 Koji

Your title is CRO — that seems to be showing up more and more at Silicon Valley companies. What does the CRO role actually entail?

👨🏻‍💻 Kai Wu

You could say my main job is driving company revenue.

Serving Moonshot AI, MiniMax, Zhipu AI

From multi-model support to tiered rates to peak/off-peak pricing, AI companies have turned a simple invoice into a real-time, dynamic system.

👦🏻 Koji

Your customers include many of today's major LLM companies going global. What's distinctive about serving them?

👨🏻‍💻 Kai Wu

The whole LLM wave really only took off in the last 18 to 24 months. For us, it's been a process of learning and developing new products from scratch — which happens to give us a latecomer's advantage.

Traditional software companies have relatively mature billing and payment models. Their approach is straightforward: set a monthly plan, bill monthly, very simple.

But LLMs are highly dynamic and real-time. First, all LLM companies typically operate multiple models with different performance levels and different pricing. So for a billing and payments platform like us, we need to do real-time billing based on usage data the LLM companies provide their customers.

Add another layer: LLM usage is typically tiered. So it's a relatively complex billing model — real-time, with rates constantly shifting.

👦🏻 Koji

Including DeepSeek, and different pricing for peak versus off-peak hours?

👨🏻‍💻 Kai Wu

Exactly. All of these logic layers need to be built into our billing and payments platform for the billing to be accurate.

👦🏻 Koji

Why does this calculation fall to you rather than the model companies themselves?

👨🏻‍💻 Kai Wu

Good question. It's most direct for us to do it, since we're the ones issuing the final bill.

First, this isn't a core competency for LLM companies. Second, even if they calculated it themselves, they'd still need to send it to us to charge the customer. So we might as well calculate it directly and complete the collection.

If they calculated it and sent it to us, there'd be reconciliation issues in between — an unnecessary extra step.

👦🏻 Koji

So the pressure on you must be pretty intense.

Undercharge, and the model company might ask you to cover the difference. Overcharge, and users get furious — possibly blaming the model company, who then comes back to you.

👨🏻‍💻 Kai Wu

That's exactly right. Since we're the ones sending the bill to the customer, the responsibility is very heavy.

How a Cup of Coffee Became a Global Payments Network

In 2015, several University of Melbourne classmates running a café discovered: buying beans wasn't hard, but getting the money to South America and Africa was.

👦🏻 Koji

We know Airwallex is an impressive company that reached an $11 billion valuation in a short time.

But for many people, you remain somewhat mysterious. Today I'd love for you to walk us through Airwallex's founding story. From zero to now, what were the main stages? What was the starting point?

👨🏻‍💻 Kai Wu

We were founded in 2015 in Melbourne, Australia, by Jack and his co-founders.

The initial problem was actually quite simple: moving money globally. At the time, they had opened a café in Melbourne. Melbourne has an incredible coffee culture, but they don't grow coffee beans locally — all beans are imported from around the world, including South America and Africa.

The payment infrastructure in those regions was quite underdeveloped. When importing coffee beans, they ran into several practical problems.

First, funds arrived very slowly. One of our founders happened to share a name, in pinyin, with someone on a sanctions list, so SWIFT transfers kept getting held up.

Second, it was extremely expensive. The sample beans themselves didn't cost much, but SWIFT transfers charged a flat fee regardless of amount — maybe $20 to $50 per transaction. The original motivation was simply solving these pain points.

👦🏻 Koji

What was their background? If they were just running a café or coffee business, how did they have the capability to build something as technically demanding and real-time as a global payments network?

👨🏻‍💻 Kai Wu

This brings us to our CEO, Jack. They were all University of Melbourne classmates who met in school. The café was just a side project they ran together out of shared interest.

Jack had worked as an engineer at a bank, developing foreign exchange algorithms and trading systems. Another founder, Xijing, had been a serial entrepreneur in technology and had worked at Google X.

When people with these technical backgrounds came together, and cloud computing was becoming increasingly mature, they started wondering whether technology could solve this problem.

The global payments network at the time was essentially running on 1970s mainframe architecture. Take SWIFT: its fundamental problem is that money and information travel separately. SWIFT is essentially just a messaging system — it only transmits information. Banks receive the message, then manually move the funds. So it's highly inefficient.

In the cloud computing era, rebuilding the logic of money movement through a cloud-based payments network was entirely technically feasible.

At the same time, real-time clearing networks were emerging around the world. What we needed was a way to connect these networks and match them with data, so that capital flows and information flows could move in sync.

So in 2015, several companies including Wise and Revolut also got their start around that time. This convinced us that rebuilding the logic of money movement networks was viable.

👦🏻 Koji

So you saw both unmet demand and maturing infrastructure?

👨🏻‍💻 Jack Zhang

Right, especially the local real-time small-value clearing networks in various countries.

The rise of mobile payments at the time was largely driven by the rapid development of payment tools like Alipay in China, which prompted many countries to build their own real-time small-value payment systems.

Once money reaches the local level, local payment is actually very fast. We just needed to solve that cross-border segment of global movement to achieve end-to-end, door-to-door, efficient, and real-time global money movement.

👦🏻 Koji

At that stage, you weren't the only ones who saw this opportunity — many peers were exploring it together.

Going forward, how did Airwallex not only seize this opportunity but gradually grow bigger and stronger? Were there competitive stories along the way?

👨🏻‍💻 Jack Zhang

I think the key difference was vision. Many cross-border money transfer companies succeeded, but in retrospect, we benefited enormously from our own vision.

There were two points in our vision that were especially important.

First, we wanted to build a global network, not a regional one. If you only built local payments in Southeast Asia, you might have turned profitable early. But we chose to build a global network, and once that network was in place, our addressable customer base would be extraordinarily broad.

Today our payment network supports real-time fund arrival in over 90 countries.

Losing Money for Four Years to Build the Global Network First

If we had only done Southeast Asia, we might have been profitable long ago; from Day One, Airwallex chose global coverage instead.

👦🏻 Koji

So at that stage, the company went through a fairly long period of losses?

👨🏻‍💻 Jack Zhang

Yes, we lost money for quite a long time. We were building this underlying network — we only started seeing a little revenue in 2018, and it wasn't relatively complete until 2019.

So for the first three to four years, all our energy went into product. For a startup, focus is extremely important when resources are limited.

We didn't spend our energy building upper-layer applications at the time. Instead, we patiently worked on integrating with banking partners, obtaining local compliance licenses, and connecting to local clearing networks. The first vision was to resolutely cover the globe.

👦🏻 Koji

How did you resist the temptation of profitability back then? At that point, many competitors' thinking was probably: do Southeast Asia well and deep first, become profitable, and then expand outward once you have a capital base.

But you chose to absorb losses and build the global network from 0 to 1.

👨🏻‍💻 Jack Zhang

Actually, making this strategic choice was very easy, because you could reach the conclusion through data.

The world's largest and most important payment corridors are intercontinental, not confined to any single region. Do the math and it's clear: to achieve sufficient market scale, you must have global coverage. If you're an observer with data, this conclusion comes easily.

But in execution, there was enormous uncertainty. The severe lag in profitability meant you had to have extraordinary fundraising ability and investor relations capability; at the same time, your team needed to withstand pressure and stand shoulder-to-shoulder with you through that long exploration period.

I think many companies of the same era couldn't predict this. If you only built a local network in the Middle East, Southeast Asia, or South America, the market scale you could reach obviously couldn't compare to a global network. Some local-network companies developed reasonably well at the time, but their eventual outcomes were very similar — they basically chose to sell themselves to giants like Visa or PayPal, becoming supplements to those giants' local networks.

If you stop network expansion too early and instead develop upper-layer applications on your existing small patch of territory, monetization is indeed fast in the short term, but it makes the road ahead narrower and narrower.

Investors Thought This "Global Bank" Vision Was Crazy

👦🏻 Koji

Understood. So vision determines the endgame. You mentioned resolutely building a global network from Day One. What about the second part of the vision?

👨🏻‍💻 Jack Zhang

The other part was, we realized that only doing money transfer is a relatively poor business, because you become too infrastructure-level — only a very limited set of B2B customers can directly use your product.

We found that to truly retain B2B customer flow, we essentially needed to provide a global transaction banking product. This meant the capabilities we needed went far beyond a money transfer network. You had to build applications on top, providing extremely rich products on both the receiving and paying ends to meet customers' end-to-end needs.

For example, on the paying side, only supporting transfers is far from enough. Customers need corporate cards for various daily expenses; they need connections to various e-wallets for payroll distribution.

On the receiving side, customers can't rely solely on bank transfers — they also need merchant acquiring services, completing consumer payments through payment gateways. You have to build all these end-to-end products.

At the time, many of our investors thought this was completely crazy. They believed startups must focus and couldn't possibly build so many products simultaneously. But our team's characteristic was being sufficiently hardworking and having extremely strong execution, and eventually we built all these products.

Having a global banking vision is crucial — it gives you the confidence to truly compete on the same track as traditional banks.

If your product line isn't complete, you have no competitiveness. Today, our transaction banking product matrix is already very complete, spanning bank accounts, cross-border money transfer, foreign exchange, card payments, merchant acquiring, high-yield accounts, and credit products — everything.

Essentially, we've rebuilt a new type of global bank in the cloud.

This process was extremely long. It wasn't until 2021 that we truly found Product-Market Fit on this global banking track.

And during 2018 to 2021, we were actually mainly focused on payments alone, which made us acutely feel the limitations of the ceiling.

👦🏻 Koji

This sounds somewhat contrary to a lot of classic startup theory. The usual theory advises that startups should first go deep in one extremely narrow vertical, then expand.

But you laid out your strategy so comprehensively from the start — this must have posed enormous challenges to organization and execution?

👨🏻‍💻 Jack Zhang

Whether you're building a very vertical feature or a full product line, the first principles come down to whether you can meet customers' baseline needs.

Actually, when building the full product line, we still followed the approach of doing as much subtraction as possible on each individual path, cutting features down to a minimally viable state — this is identical to how other startups build MVPs.

But if some core mainline capabilities are cut and customers simply can't complete a closed-loop experience, then that compromise won't work.

Our starting point has always been customers' actual needs, and the revenue ceiling corresponding to those needs.

👦🏻 Koji

Going back to the milestones you mentioned. In 2018, the global payment network was basically built and began serving customers. What types of customers did you mainly serve then?

👨🏻‍💻 Jack Zhang

We had many online education platform customers who needed to pay salaries to teachers distributed around the world; many sharing economy platforms that needed to pay landlords or service providers in various locations.

Also included were some overseas education service agencies that needed to help students pay tuition to schools abroad. So for the more than two years after 2018, this global money transfer network was our primary product platform.

👦🏻 Koji

By 2021, Airwallex had transformed into a more comprehensive cloud-based online bank?

👨🏻‍💻 Jack Zhang

Yes.

👦🏻 Koji

After that, what other key development stages were there?

👨🏻‍💻 Jack Zhang

Before the AI wave erupted, frankly we didn't encounter too many extremely exciting new product opportunities — though of course, the global banking business logic itself could still go deeper. At this stage, we were mainly filling out our capital products.

Asset management and value-add are very traditional and core banking capabilities. For example, high-yield accounts — when customers' funds are idle, how do we generate returns for them while ensuring extremely low risk and high credit ratings.

At the same time, by providing flexible credit lines and credit products to meet enterprises' working capital needs. We were mainly gradually completing these product lines.

👦🏻 Koji

When you define yourselves as a global bank, the competitors you're facing are no longer other fintech companies, but those traditional physical banks?

👨🏻‍💻 Jack Zhang

Right.

👦🏻 Koji

At that point, what was the core selling point that could convince customers to choose you over those long-established traditional banks?

👨🏻‍💻 Jack Zhang

Mainly two points.

First, the scale of network coverage. Today it's very hard to find any traditional bank globally that covers the underlying clearing network as broadly as we do. Traditional banks mainly rely on establishing physical branches locally to expand their network outward, which makes construction and operating costs extremely high and operational efficiency relatively low.

Second, user experience. Whether it's our web interface, mobile app, or our API — these are things banks simply don't have. Many platform-type customers need flexible, autonomous, controllable, and low-cost payment methods, and in this regard they actually don't have many choices.

The flexibility and controllability can manifest in two ways:

First is seamless integration with enterprise financial systems. For example, a company might generate hundreds of payable bills each month in its ERP system. If paying through traditional bank online banking, finance staff need to manually enter each one, verify amounts, and confirm payment — extremely inefficient.

We've completed standard API integrations with mainstream ERP systems. Customers only need to import bills from their ERP into our platform with one click, select the items they want to pay, and execute batch payments instantly.

👦🏻 Koji

So it seems these pain points and needs have always been clearly visible. The key is what kind of organization and execution capability it takes to actually deliver a good product.

👨🏻‍💻 Wu Kai

Exactly. Many traditional banks run on core architectures shaped in the 1950s and 1960s, with code written in the 1980s. But in the cloud-native era, writing code and calling APIs is the most efficient way for systems to collaborate.

APIs are the shortest path for systems to communicate through code, so we've insisted on API-first from day one.

For modern tech companies, integrating with our APIs is extremely straightforward. We just need to ensure flexible interface design, expose sufficiently rich technical parameters, and return underlying control to the enterprise's own engineering team. They can handle the remaining application-level development efficiently according to their own business logic.

👦🏻 Koji

I've recently been following some new-generation digital banks built specifically for startups, like Mercury. Is the underlying logic of what they're doing similar to yours?

👨🏻‍💻 Wu Kai

There are indeed some similarities. But there's one critical difference between us and Mercury: Mercury isn't an API-first platform. It primarily serves smaller, early-stage startups that typically don't want to invest additional engineering resources in financial automation or technical integration.

However, in the AI era, this logic has shifted somewhat.

Financial automation, from engineering team to a single operations person

In the past, only mid-to-large enterprises could justify building engineering teams for financial automation. AI is now making this capability affordable for SMBs too.

👦🏻 Koji

What new changes has the AI era brought?

👨🏻‍💻 Wu Kai

Previously, only mid-to-large enterprises had the motivation and budget for financial system automation, because it required assembling a dedicated engineering team to integrate APIs — a high barrier to entry.

But in the AI era, that R&D threshold has been dramatically lowered. System integration and automated workflows that once demanded an engineering team can now be handled by a single financially-savvy operations person using AI.

This means financial automation has become radically accessible. Small and medium enterprises can now enjoy this technological dividend with minimal barriers.

👦🏻 Koji

Against this backdrop, Airwallex has recently released some AI + finance products. Can you elaborate?

👨🏻‍💻 Wu Kai

We currently have two main AI product lines.

The first is called Kai, a native AI finance assistant embedded directly in our web-based system. It operates as an agent to automatically optimize customers' financial workflows.

It can analyze your company's financial and reimbursement policies; optimize your liquidity by intelligently suggesting when to park idle funds in high-yield accounts, when to execute batch payments, and how to achieve optimal cash flow allocation.

It can also automatically build approval and expense control workflows based on your company's budget management policies. Additionally, if a payment gets stuck or errors out, Kai can perform deep root-cause diagnosis.

The second is Agent OS. If a customer's financial operations staff have written their own automation scripts or mini-programs, we provide dedicated command-line tools and API-based MCP interfaces through Agent OS.

This allows customers' own AI agents to directly invoke all of Airwallex's underlying global clearing and payment capabilities, enabling highly autonomous financial automation.

👦🏻 Koji

Understood. So the latter is mainly targeted at larger, more technically capable customers?

👨🏻‍💻 Wu Kai

Not necessarily. Many startups today have extremely high technical acuity. It really depends on whether the company's finance personnel themselves have product and engineering thinking.

If finance staff have product thinking, they'll be quite eager to use Agent OS to design their own bespoke automated financial workflows.

👦🏻 Koji

So as an agent, Kai currently interacts with users primarily through a natural language conversational interface?

👨🏻‍💻 Wu Kai

Yes, it's entirely a very natural natural-language conversational interface.

👦🏻 Koji

How long has Kai been live? What typical use cases or interesting user data have you observed?

👨🏻‍💻 Wu Kai

Kai has been live for about a month and is gradually rolling out to all customers.

From what we've observed so far, user retention and stickiness in the second month are quite strong. Once customers get accustomed to this natural language interaction, it's hard for them to tolerate going back to the tedious "click-click-click" of complex system backends.

For example, creating a cross-border payment used to require filling out extensive forms; now you just send Kai a simple instruction.

If you need to build a multi-level approval reimbursement workflow, simply upload your company's PDF reimbursement policy to Kai and it automatically generates a complete workflow for your confirmation. Once you start using it, the efficiency gains are generational.

Ten financial SaaS tools will consolidate into one agent entry point

👦🏻 Koji

This experience does sound excellent. Do you think that in the future, whether for C-end Alipay or various traditional commercial banks, this kind of AI finance assistant will become an industry standard?

👨🏻‍💻 Wu Kai

For any financial institution with certain R&D and technical capabilities, this is absolutely a must-have.

Because the lift it gives to user interaction experience is just too direct and too obvious.

👦🏻 Koji

Traditional commercial banks' mobile banking apps often have hundreds or thousands of functions, making them incredibly inefficient to navigate. Using natural language as the entry point is indeed the most natural solution.

However, from your perspective, what does having this AI interaction capability mean for Airwallex's own development? Is it merely a front-end interface upgrade?

👨🏻‍💻 Wu Kai

On the surface it's an interface change, but this change generates many new capabilities.

Airwallex has already built out solid underlying global banking infrastructure — accounts, clearing capabilities. On top of this, enterprises previously had to purchase and integrate various independent vertical software solutions to handle financial processes like fund reconciliation, accounts payable and receivable, revenue recognition, and billing. These used to be handled step by step by different financial software, consuming significant time and effort.

But now, customers no longer care about the complex intermediate processes — they just want the final result. As AI agents' intelligence in specialized domains increasingly approaches human levels, the agent doesn't need to navigate traditional software menus at all. Just feed it the business logic, and it runs through the workflow itself.

This saves us enormously from developing all kinds of cumbersome upper-layer software. That's why we're convinced the future financial software ecosystem will be fundamentally reshaped.

Previously, the market was flooded with dozens or hundreds of vertical SaaS products targeting niche segments like expense control, accounting, and revenue recognition. In the future, the underlying capabilities of these software products may all be absorbed into a unified agent-based financial operating system.

Looking five years ahead, these dozen or so previously independent financial SaaS segments may consolidate into one.

Originally a company's finance team might need five people, each proficient in ten different financial software tools; in the future, perhaps just three people using a unified AI agent system can complete all work with high quality and efficiency.

👦🏻 Koji

So you're actually pessimistic about the future of current vertical financial SaaS?

👨🏻‍💻 Wu Kai

Traditional financial SaaS must actively transform, or risk being eliminated by the era.

Moreover, they now face a problem that wasn't prominent before but is fatal in the AI era: data fragmentation.

Previously, each software handled its own function — expense control SaaS couldn't access billing data, billing SaaS couldn't see reconciliation data. This wasn't a major issue.

But in the AI era, because single vertical tools cannot obtain complete financial chain data, the AI agents built on top of them lack comprehensive contextual input. Such localized AI is inevitably constrained, even unintelligent, unable to complete complex cross-domain automation tasks.

👦🏻 Koji

Understood. But for these vertical SaaS players, breaking data silos and integrating downward toward the fund layer would require enormously difficult and costly reform.

👨🏻‍💻 Wu Kai

Indeed, any transformation they attempt will demand tremendous effort.

👦🏻 Koji

Beyond Kai and Agent OS, I've noticed Airwallex has two other products — T:0 and Airi. What specific problems do they solve?

👨🏻‍💻 Wu Kai

T:0 is an AI-native startup financial management platform. The name comes from the intention that from day zero of any startup's founding, they can have an AI-powered mini-CFO or automated finance operations team to help founders manage finances.

Tasks that previously required two or three finance employees — bookkeeping, reconciliation, financial report generation, compliance policy screening, and even more strategically valuable cash flow forecasting and revenue/expense trend prediction — T:0 can independently and high-quality automate. This product is currently launched primarily in North America, and has received excellent feedback in Silicon Valley's startup circles.

The other product, Airi, starts from one-click checkout. After a consumer completes their first payment on a merchant site using our payment acceptance service, their next purchase at any partner merchant requires no re-entry of cumbersome card or personal information — just one-click express checkout.

But our vision for Airi extends far beyond "one-click checkout." We expect it to eventually evolve into an AI agent wallet for individual users.

In the future, as intelligent commerce develops, consumers will extensively authorize their AI agents to compare prices, select products, and place orders autonomously. In this process, AI agents will need a trustworthy, native AI financial wallet to automatically complete fund transfers and settlement.

👦🏻 Koji

So its ultimate vision is to support future Agent-to-Agent novel payment and clearing?

👨🏻‍💻 Wu Kai

Exactly.

Data, talent, traffic: the three cards of financial entry points

👦🏻 Koji

Speaking of SaaS, I also noticed Airwallex recently completed two industry acquisitions — Leapfin and OpenPay. Could you share the strategic thinking behind these two deals?

👨🏻‍💻 Wu Kai

Put simply, in the AI era, our M&A logic is very clear: we only look at two things — data and talent.

First, data. In finance, there exists massive amounts of highly industry-specific proprietary data that will never be publicly exposed on the internet. Without this high-quality data, you cannot cultivate AI financial agents that perform well in specialized domains.

Take our acquisition of Leapfin, a revenue-recognition SaaS, as an example. After entering this space, we discovered that revenue-recognition logic in finance is extraordinarily complex.

A traditional manufacturing enterprise, a SaaS company that keeps books on an accrual basis, and a warehousing and logistics service provider — each follows completely different revenue-recognition rules.

These granular industry-specific operational and reconciliation rules are nearly impossible to find on the public internet. Yet these vertical SaaS tools have accumulated years of real-world business cases from numerous clients. Having our AI agents learn from these authentic financial operations is the most efficient path to building advanced financial capabilities.

Second, talent. When we acquire these outstanding startups, their founders' individual capabilities are multiplied by AI productivity tools. In the past, product managers could only conceptualize on documents; now, with code-generation tools, they can rapidly build functional prototypes themselves.

We value these top-tier industry entrepreneurs highly — enough to pay a reasonable premium well above what we would have paid in the past.

👦🏻 Koji

I recently received a new feature push while using ChatGPT called Finance, prompting me to link my personal bank account. After linking, I could send natural-language commands directly in ChatGPT's chat window — check my statements, analyze spending changes, even help me draft a savings plan. The logic is very similar to your Kai assistant.

I noticed it can't currently connect to Airwallex. How do you evaluate large model platforms directly cutting into financial interactions? Is there any possibility of ChatGPT integrating with Airwallex in the future?

👨🏻‍💻 Wu Kai

This is an extraordinarily broad and strategically consequential question — in fact, leading enterprises across every industry globally are weighing it carefully.

From the current state of play, the actual experience of large language models directly providing financial connectivity and querying remains early-stage.

On one hand, because they can only connect to one or two accounts at a single point, they cannot reconstruct a comprehensive view of an enterprise's overall capital chain.

On the other hand, the reason no major financial institution or fintech giant has easily opened up fully to them involves deep considerations around who controls the underlying data and traffic entry points.

While there's no simple, definitive answer, I can draw an analogy from the e-commerce industry's trajectory:

Large model vendors are naturally eager to cut into intelligent commerce. But observe: major e-commerce platforms including Amazon and Walmart do not choose to fully share their core underlying data with large models, because data is their most fundamental moat. Those who choose to experimentally collaborate with large models tend to be platforms like Shopify — providers of underlying commercial tools that do not themselves own closed front-end traffic loops.

In this博弈, large model platforms' biggest card is traffic. But in the AI era, don't enterprises with vertical business strength want to forge their own products into "super entry points" within their industries?

If underlying large models ultimately become commoditized general-purpose goods, then why can't enterprises that possess industry-specific data and business workflows build their own vertical super-intelligent entry points, rather than handing that entry point over to someone else? This is a long-term strategic choice.

Imagine: currently users might spend twenty minutes daily on ChatGPT, but if we truly enter the AGI era, users will spend hours each day interacting with specific agents and having them complete real tasks — the commercial value of vertical business entry points becomes immense.

So all parties remain in a prolonged phase of weighing and observing. Large model giants need data from various business platforms to enrich their services, while business platforms covet the massive potential user base that large models could bring.

At this stage, algorithmic intelligence is not the hardest moat to build — the direction of traffic flow, data security, and long-term commercial闭环 are.

Rejecting Stripe's Acquisition, and Rejecting Making Money Too Soon

At the Series B, Stripe wanted to acquire Airwallex; the hard part wasn't execution, but choosing between money and vision.

👦🏻 Koji

Looking back at Airwallex's journey, were there moments when you found decision-making extraordinarily difficult?

👨🏻‍💻 Wu Kai

Often, the hard part isn't execution — it's the act of choosing itself. Especially when you must resist the temptation of enormous short-term commercial gain to stay true to a long-term vision.

For instance, after the company's Series B round, payments giant Stripe had proposed an outright acquisition of us. Had we accepted, it would have been a very respectable commercial exit — but we ultimately declined.

Similarly, there were strategic trade-offs in the company's early days between pursuing short-term breakeven versus insisting on laying the full long-term infrastructure.

I previously served as Airwallex's CFO for over three years. In 2018, I had built out a financial model projecting the company's future. One of our early lead investors looked at it and said point-blank that the projections were absurd, impossible to achieve.

But when we reviewed it in 2024, we found that actual business growth over the preceding years had tracked almost precisely along the trajectory laid out in that 2018 model, with deviation of only about one and a half quarters' worth of growth.

When you hold a clear long-term vision in your mind, and through rigorous data analysis become convinced of the market's scale and direction, while maintaining extreme discipline and tenacity in execution details — the rest comes down to whether your execution capability and perseverance can sustain you through the pressure to reach the finish line.

Looking back, in 2018 we had virtually no substantive revenue on our books. To sketch out a grand model targeting hundreds of millions in ARR during that "cold start" phase, and then track so closely to it afterward — there was certainly an element of luck, but conviction in the vision was what carried us through.

After $1.3B ARR: Contending for the Financial Super-Entry Point

👦🏻 Koji

Your latest ARR has reached $1.3 billion?

👨🏻‍💻 Wu Kai

Yes, $1.3 billion.

👦🏻 Koji

When you had almost nothing back then, what framework or logic did you use to model such a long-cycle financial projection?

👨🏻‍💻 Wu Kai

Indeed, at that zero-starting-point stage, you are entirely vision-driven.

The core anchor of the model is actually quite simple — breaking down the customer profile.

You work through it concretely: Who will be our target customers? What is their current transaction volume? What share of their future capital flows could run on our system? Based on corresponding monetization rates, how much revenue can we capture? Across different country markets, what resources do we need to deploy for expansion?

Once you break down this series of basic facts and data, the core skeleton of the model naturally comes together.

👦🏻 Koji

We've recently noticed through public channels that Airwallex's overall payment volume is racing ahead at double or even faster growth rates. What's driving such astonishing momentum?

👨🏻‍💻 Wu Kai

There are two core drivers.

First, while entirely new overseas large markets are becoming scarcer, our penetration rate and market share within existing core markets continue to climb rapidly.

Second, and most importantly — cross-utilization of our product matrix. Many of our customers initially onboarded using only our most basic cross-border collection accounts; after a year or two, they gradually adopt fund transfers, FX, acquiring, card payments, and our upper-layer automated financial SaaS, credit, and other capital products.

This drives the per-customer value contributed on our platform to grow at roughly 20% to 30% annually.

👦🏻 Koji

In the press release for this new $11 billion funding round, Airwallex co-founder and CEO Jack Zhang specifically noted that the current moment sits at a "critically important historical inflection point" in global financial development.

How do you understand this statement? Why is the present moment so special and important?

👨🏻‍💻 Wu Kai

This underlying transformation is ultimately being led by AI.

As mentioned earlier, under AI's impact, traditional finance's vast landscape of siloed point-solution vertical software faces reorganization and displacement. In this wave of reshuffling, new cross-category "financial intelligent super entry points" will inevitably emerge.

Currently the industry and academia are vigorously debating: Will future intelligent entry points be fully absorbed by general-purpose super-models, or will vertical super entry points with strong business moats still emerge in specific vertical tracks like finance, healthcare, and industrials?

Our judgment is that, at least in financial services, vertical "super entry points" will possess irreplaceable unique value.

Airwallex is making comprehensive strategic布局 with precisely this goal.

👦🏻 Koji

If the macro trend is destined to give birth to this financial super-intelligent super entry point, which other peers do you think have already secured this ultimate ticket?

👨🏻‍💻 Wu Kai

Stripe is certainly among them.

Stripe's product matrix expansion has been extraordinarily rapid in recent years. In their first decade, they were likely very disciplined, focusing solely on doing one thing extremely well: the online payment gateway.

But in the past three to four years, their product lines have been expanding aggressively horizontally. This aligns with the logic we just discussed: if you want to become the future super entry point for financial intelligence, relying on a single hit product is absolutely insufficient. Because AI intelligence itself must be full-scenario and highly generalized. So Stripe is definitely a core contender.

Another is Europe's Revolut. While their business model differs from ours, their core strength lies in having over ten million extremely active, highly sticky C-end users, giving them a natural moat on the traffic side. Beyond these two, other players are likely confined to certain regional pockets.

👦🏻 Koji

At a time when capital markets are relatively rational, what are the most critical underlying business metrics supporting Airwallex's $11 billion valuation in this round? Can you share?

👨🏻‍💻 Wu Kai

First, we have very solid business fundamentals on the revenue side: we have $1.3 billion in ARR.

👦🏻 Koji

8x PS.

👨🏻‍💻 Wu Kai

In the current market environment, an 8x PS multiple isn't high for a tech company with $1.3 billion in ARR that's still growing at 90% year-over-year.

At the same time, we're in the fintech industry, which has extremely high gross margins — consistently around 70%, on par with the world's top software companies. After accounting for administrative and personnel costs, our profitability is also very solid, so the fundamentals are more than sufficient to support the valuation.

Moreover, our core consideration in choosing to close another funding round at this particular moment is that it allows us to more confidently and aggressively acquire high-quality data assets in the industry, and to attract the best talent.

Having ample capital significantly accelerates this process for us.

👦🏻 Koji

This sounds like Airwallex will be making quite a few frequent and bold M&A moves in the market in its next phase?

👨🏻‍💻 Wu Kai

Yes, we are indeed maintaining extremely high alertness in evaluating various M&A opportunities. The core criteria are locking in top-tier talent and unique industry data.

The hardest part of globalization isn't choosing markets — it's choosing the right locals

👦🏻 Koji

As a company that had globalization written into its DNA from day one, and successfully took extremely complex transaction finance business into compliance across dozens of countries, what practical lessons can you share with Chinese companies and entrepreneurs going global today?

👨🏻‍💻 Wu Kai

The most critical lesson: unwavering commitment to localization.

You must always believe that only local talent who truly understand the culture, rules, and business logic can deliver experiences that best match local customer expectations.

To put this principle into practice, we started recruiting many local GMs in 2019. It was a process of repeated failures and retries — fail, try again, fail again, try again — until we found the right people.

The reason is, when you haven't yet built strong brand recognition locally, the initial candidates you attract are highly uneven in quality. It takes time to sift through them. In the US market, we went through multiple GM changes, and we're still adjusting even now.

At one point we had a head of US commercial operations who was nominally based in San Francisco, but actually lived in San Jose nearby. Once, Jack and I were traveling there and gave him a heads-up that we'd like to meet in San Francisco to discuss strategy.

He said there was a family emergency and he couldn't make it from San Jose. We didn't think much of it and let it go.

But then we ended up staying a few extra days to meet with local investors, and he still never showed up. Eventually he admitted that he was actually in Miami at the time, on vacation with his family while working remotely. We fired him immediately — it was completely unreliable.

That whole process of building trust, having it break down, and then recruiting again is incredibly draining. But we still want to do the right thing.

Even facing setbacks like this, we don't want to send someone from headquarters to run the US market — that still won't work. We insist on hiring locals to run local operations. This is extremely important.

👦🏻 Koji

Beyond localizing talent, when building out such a massive global footprint, how do you define the priority of different regional markets and allocate resources accordingly?

👨🏻‍💻 Wu Kai

At first we used a two-axis matrix: market size on one axis, ease of entry on the other. But later we realized it was unnecessary — and it didn't help us decide anything.

Everyone wants markets that are both large and easy to enter, but in reality those markets don't exist.

So we later realized we don't need to consider entry difficulty at all. We only need to consider market size. Market size is the sole criterion for our market entry priority.

And measuring market size doesn't need to be complicated — just rank by GDP. We'll enter all top ten countries by GDP first. After that, tackle countries 11 through 20. Beyond 20, decide case by case.

This logic is extremely simple, yet it gave us tremendous strategic conviction throughout our growth. We don't waver, because the biggest fear in going global is fluctuating resource commitment — pouring in more today, less tomorrow — which rarely yields good results.

👦🏻 Koji

But we've also seen many traditional East Asian companies — Japanese and Korean giants, and even Huawei in its early days — largely adopt a deployment model where headquarters parachutes in core executives and rigidly transplants processes and systems.

How does this compare with your philosophy of thorough localization — where are the similarities and differences?

👨🏻‍💻 Wu Kai

I have thought about this question.

My personal view is: if you're a manufacturing company, the parachute model works very well. For example, when building a local factory — on that specific task, your familiarity with headquarters processes and your superior factory-building experience and management methods matter far more than your understanding of local culture.

But our main work on the ground is twofold: compliance and marketing. For these, understanding local culture and rules matters far more than understanding headquarters products.

Because our products aren't manufactured locally at all. They're entirely fintech products developed by headquarters. In this case, parachuting people from headquarters doesn't make much sense.

👦🏻 Koji

Since you've positioned "marketing" as one of the core missions of your overseas local teams. In practice, do different country markets show huge cultural and habitual differences in customer acquisition methods and touchpoints?

👨🏻‍💻 Wu Kai

There actually are differences, mainly in the composition of marketing costs across countries.

In Asia, because the digital economy is highly developed, online traffic costs are very high. But conversely, labor costs aren't as high as in Western countries. So sometimes in Asia, online marketing is less effective than simply deploying marketing staff for ground-level outreach.

Western countries are completely different. While their online traffic costs are similarly expensive, when you compare online marketing costs against their extremely high labor costs, online channels actually don't look so costly.

So when we entered the UK, Europe, and North America, we mainly built online marketing teams and drove growth through digital channels.

From $11 billion to $100 billion, a very simple arithmetic

One million customers, each contributing $10,000 per year — that's $10 billion in annual revenue.

👦🏻 Koji

Final question. Airwallex took roughly ten years to reach an $11 billion valuation. If you want to steepen that acceleration curve and push toward the ultimate $100 billion club, what are the most critical milestones that need to be crossed along the way?

👨🏻‍💻 Wu Kai

I think the first milestone is a leap in user scale — we need step-function growth. We currently have roughly 200,000+ active customers, and our target is to reach one million.

Second, we want our product to not just serve customers' financial operations, but to truly help them achieve intelligent finance. Behind intelligent finance, we can see our customers' finance teams significantly improving their efficiency through our empowerment.

We can do some simple math: if we can get each customer to contribute an average of $10,000 in annual revenue, at one million customers that supports a $10 billion revenue scale.

This is how we can truly achieve our future goal of becoming a $100 billion intelligent finance platform.

👦🏻 Koji

Thank you, Wu Kai, for the deep sharing today. And best wishes for Airwallex to reach its goals soon and successfully join the $100 billion club.

👨🏻‍💻 Wu Kai

Thank you, Koji. Bye.

👦🏻 Koji

Bye.