Stablecoins, business of intelligent agents, and AI in the eyes of Mastercard

Original by Will A-Wang

Stablecoins and agentic business have been the most talked-about topics in the past two years, but the ones truly making trade-offs on their balance sheets are not the ones issuing stablecoins or the ones creating agents, but the card organizations. Others may see both as business opportunities, but before commercialization, card organizations must first answer a more difficult question: If money no longer flows through my channel, what do I have left?

Mastercard's CEO Michael Miebach answered three questions on the same day: How does money flow, who is buying, and how do we make judgments?

On the stablecoin question, he said there are "no problems to solve" in daily consumption; on the agentic business question, he said cards will prevail; on the AI question, he said the winners are companies holding proprietary data. The three answers sound unrelated, but they are performing the same action—settlement can be multi-track, but trust and security must be secured.

On July 30, the day of the second-quarter earnings report. At 9 AM, Michael Miebach held a conference call with analysts; afterward, he sat down in the Motley Fool podcast studio and chatted with Tom Gardner for forty-six minutes. One occasion was aimed at institutions, the other at retail investors, with varying tones, but the judgments were the same. Three days later, on August 3, Mastercard completed the acquisition of BVNK for $1.8 billion, five months ahead of schedule.

In a previous article titled "Stablecoins and Agentic, Not Included in Visa's Earnings Report," we analyzed the revenue model of card organizations and Visa's strategic positioning. This article aims to clarify three things from Mastercard's perspective: why the settlement layer made concessions, why the trust and security layer is uncompromising, and the significance of the $1.8 billion acquisition of BVNK three days later.

1. What is this company actually selling?

Tom Gardner opened by asking him to set the stage and clarify the relationship between banks, merchants, and cardholders. Miebach first corrected a number: not 4 billion cardholders, but 3.7 billion, and then gave a statement that many have never thought deeply about: cardholders are not our customers. The customers are the banks, and they can also be the merchants, which could be Walmart or JPMorgan Chase.

This statement determined everything that followed. It means that what this company has never sold is payment, but rather the layer that allows payments to be trusted.

What does trust look like specifically?

When you go shopping, you can either leave the website waiting for delivery or walk out of the store with your items. The reason you can do this is that there is a payment guarantee issued by Mastercard behind it, telling the merchant: you can let this person go, we guarantee you will receive the money. The four-party model's transfer then occurs, where your bank withdraws money from your card account and transfers it to the merchant's bank. And this needs to be established in 220 countries and regions, with 3.7 billion cards, where the regulatory rules and infrastructure differ in each market.

The other side of the trust guarantee is risk control. If you paid on a fake website, you are still protected because it’s not your fault; to block fraud at the source, the backend scans trillions of data points at the nanosecond level: Is it really possible for you to be in this location right now? Is this expenditure greater than any previous one?

Miebach's own description is the operating system of the digital economy. This sounds like PR language, but what is valuable is how he subsequently breaks it down: the operating system has a security layer, a funds transfer layer, and above that, data and insights. The description of the funds transfer layer should be read word for word; it spans stablecoins, account-to-account, and cards.

The trust and security layer is singular, while the funds transfer layer is plural.

The revenue structure is also moving upward. In the second quarter, net revenue grew by 12% on a currency-neutral basis, and value-added services and solutions grew by 18%, outperforming by six percentage points, with about 60% tied to the network. Security, identity, fraud management, personalization—these are not payments, but the judgment sold around payments.

On the security side, he provided a magnitude: by 2030, losses from fraud and cyber risks will reach $15.6 trillion; if cyber risk were a country, it would be the third-largest economy in the world. The strategic shift he described can be summarized in one sentence: from defense to offense. Based on the threat intelligence capabilities built with Recorded Future, over the first three quarters, more than 7 million test card transactions were identified in 192 countries, preventing an estimated loss of $172 million.

2. Stablecoins: How does money flow?

Miebach's most complete statement on stablecoins was not in the podcast but in the opening statement of the conference call, where he said only one thing: stablecoins have great potential, but to really get them running, several necessary principles are required—reliability, security, and interoperability, which is exactly what Mastercard provides.

He does not deny stablecoins; he denies that stablecoins currently possess the three conditions.

He then delineated the boundaries of the scenario, with the version in the conference call being more straightforward than in the podcast: stablecoins have clear utility in some B2B and P2P flows, but there are no problems to solve in P2M.

In the podcast, he expanded this into a visual. Why would you use stablecoins to buy coffee at the café downstairs? It could be replaced by a small business paying a supplier in another country, going through an agent bank system, which is costly and lacks transparency; you send out $100, not knowing that both parties take $5 each, leaving only $90.

His principle is very straightforward: this has never been about technology, but about whose problems we can solve. He emphasized that Mastercard has been investing in account-to-account systems since 2016 and is now one of the largest A2A solution providers; cards are a significant part of the answer but not the answer for all payment types. Today, this network can handle dollars, any fiat currency, and stablecoins, and not only does it allow stablecoins to run in the system, but it also needs to transfer the protections you expect from card payments over as well.

He did not say stablecoins are not viable; he said stablecoins have not yet developed an acceptance network.

Because Mastercard has spent 60 years building the world's largest acceptance network. No one wants a payment solution that can only reach a small number of terminals; what you want is scale, predictability, and protection. And these things are not currently provided by stablecoins.

In the conference call, he articulated the same idea more structurally: we expect a world of multiplicity, many coins, many chains, and all of these need a trusted interoperability layer because people will transact across different currencies.

In March, they partnered with SoFi, in May they obtained a BitLicense from New York DFS, in June they expanded settlement to six regulated stablecoins and eight chains, and earlier there were multi-currency networks MTN, Crypto Credential, and One Credential, which consolidates fiat and stablecoin balances into one credential.

Despite the dense actions, what about the results?

In the conference call, he provided only one growth figure related to digital assets: the transaction volume of crypto co-branded cards has more than tripled in the past two years, with Bitget and Kraken added this quarter. This number is real, but it is important to see what it measures—it measures how many stablecoins have been spent, and the path of spending is still through cards.

As for the number that could prove "stablecoins running on our network," he did not provide it. Some analyses pointed out that Mastercard has never disclosed stablecoin settlement volumes comparable to Visa, and most actions may still be in pilot stages. In contrast, the numbers on the other side are public: by April of this year, Visa's annualized stablecoin settlement throughput was $7 billion, with a quarter-over-quarter growth of 50%, covering nine chains and over 160 stablecoin card projects. More striking is the on-chain card settlement share distribution: Visa 97%, Mastercard 3%, despite both supporting nearly equal numbers of projects.

The framework is more complete than its competitors, but the traffic is less. He is discussing a methodology regarding stablecoins, while the evidence he can present is the growth of stablecoin holders using cards for consumption.

3. Agentic Business: Who is buying?

Analyst Ramsey El-Assal asked the best question of the day during the conference call: Are there scenarios in agentic business that absolutely require stablecoins? Or can traditional Mastercard credentials cover everything?

Miebach did not directly answer yes or no but categorized the transactions into two types.

3.1 AI Shopping for People: Cards are Sufficient

Suppose you want to go camping and ask AI what to bring; it gives you 15 items, knowing you already have a tent, so it doesn't recommend one. But after receiving the list, you still have to search one website at a time, wasting a lot of time. If you could check out directly at that moment, delegating the checkout to an agent to shop, with a Mastercard running behind it, things would go smoothly.

But problems arise. There is now an entity that did not exist before; how do you know this agent is legitimate and not a forgery? How do you know it is buying what you want? If it orders two grills instead of one, what proof do you have when the bill is deducted?

Mastercard Agent Pay's answer is to transfer the existing systems from the card framework: tokenization, zero liability protection, dispute resolution. Among the most critical is the verifiable intent he named during the conference call, which allows you to challenge a transaction, saying you never intended to buy this, and the chargeback process restarts. He also added a point not mentioned in the podcast: this was developed in collaboration with Google.

Chargebacks are valuable because processing them is costly. The reasoning is simple: merchants always need to reach and predict user experiences, and consumers do too. So we truly believe that cards will prevail in that world.

He applied the same logic to the enterprise side. He clearly stated in the conference call that there are a number of agent transactions on the B2B side, where an agent makes purchases for a company, which can fully run on the card ecosystem, with the same logic. Amounts, speeds, and purposes all match, and they also need protection and global reach. In other words, AI shopping for you and AI shopping for companies, in his view, do not pose a challenge to the card channel.

Where is the business? Tokenization itself is a service sold by Mastercard, and in the second quarter, the token penetration rate just surpassed 40% of all routed transactions, leaving the remaining 60% as runway.

But essentially, people will not buy five more tents just because there is an agent; this is more about replacing existing traffic.

3.2 Machines Buying: On-Chain Authorization, Off-Chain Settlement

The dividing line is drawn here.

A company purchasing APIs, computing power, data, and content—why still send an invoice? You would want to pay based on usage, needing to add 10% more computing power and then scale it back down, paying for what you use, which immediately improves working capital efficiency. This is what a Chief Procurement Officer wants, and also what a CFO wants. But it requires a payment ecosystem that can support always-on, high-frequency payments priced in tiny fractions of a dollar, which does not exist today.

In the opening statement of the conference call, he summarized this architecture in one sentence:

on-chain permissioning, off-chain settlement.

Mastercard is the only network supporting machine-to-machine payments.

The eight characters complete the division of labor. Authorization is on-chain because machines need to instantly recognize a transaction; settlement is off-chain because the path money takes is another matter. He expressed the latter more loosely in the Q&A: settlements will occur on different types of tracks, possibly involving stablecoins or other methods, and we are quite open to that.

The version in the podcast was more straightforward:

The underlying tracks and infrastructure may differ from card tracks. It could be stablecoins or something else. It depends on the enterprises' choices; we are quite neutral on this, but the upper layer that maintains trust and interoperability is key.

The initial ecosystem includes over 30 companies, with names like Adyen, Ant International, BVNK, Checkout.com, Coinbase, OKX, and Cloudflare. The appearance of a CDN company on the initial list of payment protocols indicates that this line is not targeting traditional procurement processes but rather computing power and API billing.

His position of concession is therefore very precise. As long as the buyer is still a person, whether an individual or a company, cards will win; only when the buyer becomes a machine does he acknowledge that there is more than one track, but he wants to maintain the upper layer of trust and interoperability.

Settlement can be multi-track, but trust and interoperability must be secured.

4. AI: How to Make Judgments

Tom Gardner posed a heavy question: Elon Musk said AI will surpass human intelligence in five years, and almost all jobs can be done more cheaply, while the tech companies with the strongest cash flows and balance sheets are streamlining their workforce. What does this mean for consumer spending?

Miebach's answer first focused on the technology itself. This is something that needs to be explored; if deployed properly, it can drive paths to prosperity and growth, but it also has downsides, as generative AI is empowering scammers, fraudsters, and hackers. The same set of technologies can be used for attack or defense, so this is an arms race.

He elaborated on this duality more specifically in the conference call. What occupies the minds of CEOs and boards now is cybersecurity, and discussions around cutting-edge models are precisely whether they pose a threat or serve as tools to identify vulnerabilities. His answer is both. Mastercard itself is using cutting-edge models to accelerate the identification of internal vulnerabilities and sharing this best practice with clients.

On a human level, he said this industry, and actually most industries, ultimately depend on having the best talent, so it is essential to upgrade skills. He defined the direction as "human-centered AI applications," using tools to do work better, not doing repetitive tasks. He gave his example of creating an AI assistant to handle emails, still taking a glance, but the trivial parts are taken away. He also mentioned that many clients currently want to discuss agentic business and stablecoins with Mastercard, and the team uses AI to organize public information to prepare for these conversations, saving time to focus on the truly important parts.

As for why this company is not afraid, his exact words were that Mastercard has always been about technology, not about numbers, because it is fundamentally a network company, with a relatively light human structure compared to its market value, yet operating in 220 countries and regions. So "our industry is not one that requires fundamental rethinking."

He also conveniently distanced himself from AI trading: we are not AI trade, nor AI infrastructure; we are doing applied AI. The context for the $4.9 billion share buyback in the second quarter lies here; when the AI sector dominated the market, Mastercard became a source of funds to be sold off, with the stock price once dropping from $570 to $470. He said the buyback is opportunistic; we do not engage in the buyback business, only doing it when it makes sense.

His definition of the winners in the AI era is the simplest statement of all: the companies that stand out are those that can use various different models and have proprietary data to feed into those models. Mastercard possesses one of the most unique datasets—transaction data. He said this is what gives this company long-term viability and eligibility to participate.

5. The Significance of Acquiring BVNK

The acquisition that closed three days later actually hides the answer within the four-party model.

In the traditional fiat currency system, the final transfer of funds is completed by banks, while Mastercard is responsible for the instructions and clearing logic. In the scene described in the first chapter, where "you can walk out of the store with your items," the money transfer is done by your bank and the merchant's bank, providing guarantees and routing. Member banks constitute a fiat currency track, and on this track, Mastercard does not need to hold a license or touch the funds—it never controls the settlement.

The problem is that when he states that settlements may occur on blockchain stablecoins, there is no corresponding member bank system on another track to do that for him.

BVNK was bought to fill this position. He positioned it during the conference call as follows: with BVNK, Mastercard will act as that trusted interoperability layer, allowing customers to send, receive, store, and convert assets.

5.1 The Stablecoin Track Running in the Market Today

Closing on August 3, the base consideration of $1.5 billion plus $300 million in performance guarantees was completed five months ahead of the original year-end schedule. The asset list brought over includes: approximately $30 billion in annualized stablecoin transaction volume, over 130 markets with more than 25 licenses, the MiCA authorization obtained in February, and direct access to the SEPA euro track through Lithuania. The client list includes Worldpay, Deel, Rapyd, Flywire, and Visa Direct.

Once integrated into Mastercard, it will undertake three specific tasks:

• Provide 24/7 stablecoin settlement for processors and acquirers,

• Integrate stablecoin checkout capabilities into Mastercard's payment gateway,

• And facilitate exchanges between fiat and stablecoins.

What member banks do on the fiat side—receiving payments, settling, and exchanging currencies—BVNK will do the same on the stablecoin side.

Miebach himself provided a noteworthy reason for the acquisition: BVNK's payment orchestration capabilities, license reserves, and connectivity are highly differentiated, and they are already running in the market today. This is what the $1.8 billion buys—not technology, but time.

Chief Product Officer Jorn Lambert provided an official description in the acquisition press release: In a multi-currency world where fiat, stablecoins, and tokenized deposits coexist, the next payment paradigm will be defined by the efficiency of connections between various tracks. Translated, it means we do not create tracks; we create connections. But to create connections, we must first hold a track ourselves.

5.2 Details of Transactions from All Parties

The bidding process for this transaction was described more clearly by BVNK's earliest institutional shareholders than by either party. Concentric founding partner Kjartan Rist entered at a $4 million valuation in 2019 and has not sold a single share in eight years. He provided three facts, each more informative than the press release.

The source of pressure is Stripe. By the end of 2024, Stripe will buy Bridge for $1.1 billion, putting a cat among the pigeons. Rist said Mastercard holds the deepest respect for Stripe, then translated that into another way of saying it: this means they are looking back. He broke down the threat into three points: execution capability, product simplicity, and lack of historical baggage.

Coinbase offered a higher bid but lost due to cultural fit. Reports indicate Coinbase once offered $2.5 billion, but the founders valued chemistry when selecting a buyer, and Coinbase is an exchange while Mastercard is a financial services company. BVNK tried, but it didn’t work out; Mastercard was always waiting on the porch.

Visa, on the board, chose not to pursue. Visa is both a shareholder and holds a board observer seat, yet ultimately did not take action. Rist's interpretation is that Visa adopted a different strategy, not owning operators but collaborating with multiple operators.

The same target, the same table. One rents the pipeline, the other buys the pipeline.

$1.8 billion is not considered expensive; in the same quarter, Mastercard repurchased $4.9 billion of its own stock, with the acquisition cost being just a little over a third of the buyback. But there is a fundamental difference between the two: stocks bought back can be issued again, while a company bought cannot be returned.

The BitLicense obtained in May is another signal. Card organizations traditionally do not need to hold licenses; licenses are held by member banks. Mastercard itself applied to New York DFS for qualifications to clear tokenized deposits and payment stablecoins, along with the 25 licenses and SEPA direct connection brought by BVNK; this company is building a new track, progressing even faster than those Fintechs.

On the stablecoin track, it has no member banks, so it acts as a member bank itself.

Conclusion

Miebach once said something that sounds plain but is actually the whole answer: what consumers want in payments has not changed for decades—simplicity, security, and knowing what will happen if something goes wrong.

This is what card organizations have truly been selling for sixty years. Not clearing, not routing, but "who is responsible when something goes wrong."

Stablecoins cannot provide this answer; there is no chargeback on-chain. Agents cannot provide this answer, so they need to create a verifiable intent with Google. AI also cannot provide this answer; models can recommend but cannot hold accountability.

Thus, his three answers are essentially the same: you can choose any track, but if something goes wrong, someone must be responsible, and I will charge that responsible person. He places this system on the trust and security protocol layer, and then to ensure the protocol has something to connect, he bought a new track he once said did not need to be owned.

But this bet has one premise that has yet to be tested: will responsibility itself still be as valuable as it was in the past?

The statements from Miebach in this article, unless otherwise noted, are from the Motley Fool podcast, while those marked "conference call" are from Mastercard's Q2 2026 earnings conference call.

This article does not constitute any investment advice. The stablecoin industry is currently undergoing changes in regulation, competitive landscape, and interest rate environment; any variable brought forward or delayed could render the judgments in this article untenable—including the parts the author is most confident about. The calculations in the text are a way to understand the issues, not a commitment to stock price direction. The market is always more complex than the framework; please treat this article as a starting point for thought, not a basis for action.