Mastercard Is Fighting to Stay the Middleman for AI Shoppers

Mastercard's stock is down more than 10% this year, trading near the bottom of its 52-week range at roughly a 30-times earnings multiple — even as the underlying business grows at double digits. So when the company used a September 9 announcement to roll out new AI tools for retailers, the market's reaction was muted. That misses the point. This is not a new product line for shops. It is Mastercard answering the single most threatening question hanging over its future: what happens to a company that charges every time a card crosses its network, when the people doing the buying are no longer people?

Mastercard makes money the way a toll road does. Every transaction that runs across its network generates a fee, and on top of that sits a fast-growing layer of "value-added services" — security, authentication, data — that also earns per use. The toll-road logic is why the company prints a roughly 46% net margin and why the market pays a premium multiple for it. The toll is the thesis. Anything that moves transactions off the network is an attack on that toll.

Now the transacting is moving off humans.

The threat is real, and it is measured

In a report released days before the merchant announcement, Mastercard projected that more than one in ten online shoppers will routinely use AI agents to buy products on their behalf by 2030 — more than 300 million shoppers globally. Early-adoption data supports the direction: teens are twice as likely as their parents to already use AI weekly to find the best price or discount, and 27% of teens say they are likely to use a fully AI-run shopping assistant that recommends, chooses, and completes purchases.

If an AI agent is the one doing the buying, who gets paid? The uncomfortable answer for a card network is: maybe nobody on its rails. The agent finds the product, checks the price, and could route payment through a dozen channels — open banking, a stablecoin, a direct bank transfer — none of which touches Mastercard. That is the disintermediation nightmare, and it is the popular narrative: AI "cuts out the middleman" in commerce the way it threatens to everywhere else.

Mastercard's counter is that the middleman was never the shopping. It was the trust.

The abundance-scarcity problem flips in Mastercard's favor

Here is the mechanism worth understanding. AI does not make payment easier to trust; it makes shopping effortless and free. Discovery, comparison, recommendation — the entire front half of commerce becomes abundant, and abundant things tend toward zero price. What becomes scarce as a result is the back half: confirming that a purchase is authorized by a real person, that the payment will settle, that a dispute can be resolved. That is exactly the asset Mastercard has spent decades building.

The September 9 announcement is that thesis made concrete. Mastercard's Agent Suite for Merchants lets a retailer feed its product catalog, pricing, and fulfillment data to AI agents so they can be discovered and bought from directly. Agent Connect is a single integration point so a merchant can connect to many AI shopping experiences at once instead of wiring up each one. And underneath both sits Agent Pay, the mechanism that carries the verifiable consumer authorization and executes the payment. As Mastercard's chief product officer put it, AI agents shift commerce "from search and discovery to checkout and disputes" — and Mastercard intends to be the company that owns checkout and disputes.

The strategic logic is clear: if AI sparks an explosion in the number of transactions — agents paying for compute, data, and digital services at machine speed, in very small amounts — the toll-road business model stops depending on individual humans and starts depending on volume. The June launch of Agent Pay for Machines made the logic explicit, with more than 30 industry leaders signing on to a protocol for agent-led transactions. The bet is that machine-to-machine payments become a volume layer Mastercard can toll.

What the news does not tell you

Here is the honest evidence boundary. The announcement discloses no pricing, no fees, no revenue attached to these merchant tools. This is positioning, not earnings. The numbers Mastercard reports today — Q2 net revenue up 12% on a currency-neutral basis, adjusted EPS of $5.04, value-added services up 18% — come overwhelmingly from the existing, human-driven network. The AI-for-retailers push is a bet on what happens after 2030, and it is priced as such: a stock already discounted on the year is the market expressing its doubt that Mastercard will win the agentic commerce race.

That doubt deserves respect. The multi-rail strategy — letting agents settle across cards, bank accounts, and stablecoins — is Mastercard hedging its own toll by owning whichever rail wins. But if agents settle natively on stablecoins or bypass the card brand entirely, even the trust layer gets thinner. The bear case is not that AI agents won't shop. It's that they'll shop without needing Mastercard's permission to do so.

The reframe is the useful part, and it applies beyond this one announcement. When a technology makes something abundant, the scarce complement is where the value concentrates. AI makes shopping abundant; it does not make settlement — trusted, authorized, final settlement — abundant. That is the prize Mastercard is competing with Visa to own, and this week's merchant tools are the cleanest statement of its bid: not a gadget for retailers, but a claim to remain the middleman the machines have to trust. Whether that claim holds is a question the data will answer over the next few years — not in this quarter's toll collections.