Bernstein: AI Agent Commercialization Boosts Strategic Value of Visa and Mastercard Payment Networks
nashnova research
Bernstein says AI agent transactions could add several percentage points of growth to Visa and Mastercard, as a new token war breaks out — whoever controls the credentials and tokens for agent commerce will hold the commanding position in this emerging ecosystem.
The agent token race — what is everyone fighting over?
Meta's AI shopping agent Muse has opened a new frontier: agents placing orders on behalf of consumers. Bernstein notes 100% of agent transactions will be tokenized — tokenization converts real card details into one-time secure codes so merchants never see the actual card number.
Visa launched Intelligent Commerce; Mastercard launched Agent Pay. Both embed richer behavioral data and security layers into the token. This means → the networks are not just routing payments — they want to become the trust layer underneath every agent transaction.
Yet Muse currently checks out via Stripe's Link or Shop Pay, using one-time virtual cards — not Visa's or Mastercard's "agent-specific" tokens. In plain terms = the networks still capture the transaction, but they are missing the higher-margin token value-add business for now.
This fight has been fought before — what does the 2014 token war tell us?
Bernstein draws a direct parallel to 2014: when Apple Pay launched, Visa and Mastercard rolled out their own token services. After years of competition, network tokens now account for 60% of e-commerce transactions.
This means → once tokens reach scale, the networks gain a distribution channel for fraud scoring, automatic credential updates, and other value-added services — that is where the real profit sits.
Stripe launched its own Secured Payment Tokens (SPT) for agent transactions roughly a year ago, charging 0.15 cents per token. This reflects that third-party processors are moving fast too — Visa and Mastercard face real competitive pressure.
Why could "liability shift" be the deciding factor?
Bernstein argues the key to network dominance in agent commerce may come down to liability shift — who bears responsibility when a transaction goes wrong. Even if third-party tokens match on technology, dispute resolution and the liability framework remain Visa's and Mastercard's moat.
In plain terms = when AI spends money on your behalf, the scariest question is not "can it work?" but "who do I call when something goes wrong?" Whichever party answers that question wins consumer trust.
The data underscores the gap: only 23% of U.S. consumers say they trust generative AI to handle payments for them. This means → the wider the trust deficit, the more valuable a platform that offers complete governance — identity verification, chargeback management, audit trails.
Beyond card networks — who else stands to gain?
Bernstein expects AI agents to push online penetration higher in complex services and select retail categories. Card networks benefit naturally, since cards carry higher penetration and value-added attach rates online than offline.
PayPal sits in what Bernstein calls "an interesting strategic position": its two-sided wallet can extend trust, protection, fraud prevention, and dispute resolution, while helping consumers dynamically find savings. Affirm and Klarna may benefit through dynamic underwriting and interest-bearing loans for near-prime and subprime customers.
Ratings: Bernstein rates Visa, Mastercard, Affirm, Adyen, Block, and Toast Outperform; PayPal, FIS, Fiserv, Global Payments, and Klarna Market Perform.
When will this thesis be tested?
Bernstein estimates that if agent transactions eventually capture low-single-digit to high-single-digit percent of total consumer transactions, the contribution to network growth will be several percentage points.
Three verification checkpoints: ① actual adoption pace of agent-specific tokens; ② whether card issuers are willing to participate; ③ whether Google or Meta ultimately integrate agent-specific tokens.
This means → the thesis holds only if agent transaction volume genuinely scales — and that depends on whether consumer trust can climb past the 23% starting point. For now, the direction is visible but the endgame is not.
市场有风险，内容仅供研究参考，不构成投资建议。