✦ Agent Pay: The Card Held by SoftwareThin moat
Mastercard (MA) — the future bets
Mastercard wants every AI agent that pays for something to pay with a Mastercard token.
Mastercard's second bet is that software will increasingly pay on people's behalf, and that the card should go with it. It launched Mastercard Agent Pay in 2025, a framework for AI-assisted and fully automated payments across its acceptance network, built on its tokenization capabilities including Agentic Tokens; it enabled all United States cardholders to participate in 2025 and scheduled a global launch for early 20261.
The logic is defensive and offensive at once. An agent that shops needs a credential it can use safely, and a tokenized card with limits and dispute rights is the credential most merchants already accept. If Mastercard supplies it, every agent purchase is a Mastercard transaction.
The company has gone further with machines paying machines. On the second-quarter 2026 call it described Agent Pay for Machines, a protocol for low-ticket, high-velocity machine-to-machine payments, which attracted over 30 industry leaders at launch2.
The risk is that agents choose the cheapest rail. An agent has no loyalty to a card brand and no interest in points unless its owner tells it to.
The strategic stakes are larger than the early revenue. The filing describes extending the network's reach to enable the tokenization of credentials, identities, assets and data and their exchange between counterparties3. An agent that buys on its owner's behalf needs exactly that: a credential, an identity and a way to dispute a mistake. Mastercard's dispute management capabilities are part of what Agent Pay builds on4, and dispute rights are one thing a bank transfer does not offer.
It is early, and nothing is reported. The number that would show the bet paying is tokenized share of transactions, about 40% in 20255; agentic commerce running on tokens should push it up.
Live for all US cardholders in 2025, global launch planned for early 2026, machine protocol launched.
Early adoption of the machine protocol; a disclosed transaction count would be the first real evidence.
Source: Mastercard Q2 2026 call highlights (Yahoo Finance) ↗- ReportedIt launched Mastercard Agent Pay in 2025, a framework for AI-assisted and fully automated payments across its acceptance network, built on its tokenization capabilities including Agentic Tokens; it enabled all United States cardholders to participate in 2025 and scheduled a global launch for early 2026.Mastercard Form 10-K for fiscal 2025 - Item 1 business: the network, competition, tokenization, account-based payments and human capital. — FY2025 · publ. 11 February 2026 · source ↗
- ReportedOn the second-quarter 2026 call it described Agent Pay for Machines, a protocol for low-ticket, high-velocity machine-to-machine payments, which attracted over 30 industry leaders at launch.Yahoo Finance / GuruFocus, Mastercard Q2 2026 earnings call highlights (third-party summary of the call). — Q2 2026 · publ. 31 July 2026 · source ↗
- ReportedThe filing describes extending the network's reach to enable the tokenization of credentials, identities, assets and data and their exchange between counterparties.Mastercard Form 10-K for fiscal 2025 - Item 1 business: the network, competition, tokenization, account-based payments and human capital. — FY2025 · publ. 11 February 2026 · source ↗
- ReportedMastercard's dispute management capabilities are part of what Agent Pay builds on, and dispute rights are one thing a bank transfer does not offer.Mastercard Form 10-K for fiscal 2025 - Item 1 business: the network, competition, tokenization, account-based payments and human capital. — FY2025 · publ. 11 February 2026 · source ↗
- ReportedThe number that would show the bet paying is tokenized share of transactions, about 40% in 2025; agentic commerce running on tokens should push it up.Mastercard Form 10-K for fiscal 2025 - Item 1 business: the network, competition, tokenization, account-based payments and human capital. — FY2025 · publ. 11 February 2026 · source ↗