In Citrini Research's 2028 Global Intelligence Crisis scenario, Mastercard's revenue growth slows to 3.4% in Q1 2027. Visa is described as "under siege." The mechanism is not a recession. It is consumer agents routing transactions around the card networks entirely.
This is the part of the AI displacement story that financial analysts keep underweighting. Card interchange is a tax on friction, and AI is a friction-elimination machine.
The interchange business is a friction tax
Visa and Mastercard do not move money. Banks move money. The networks are coordination layers that charge merchants 1.5% to 3% per swipe in exchange for fraud rules, dispute arbitration, and a brand consumers trust at checkout.
That spread is enormous because the alternatives are worse. ACH is slow. Wire is expensive. Cash is local. Every direct settlement rail in the United States loses to Visa and Mastercard on convenience at the point of sale.
The economic logic shifts the moment a software agent does the buying. An agent does not need a Visa-branded card to feel safe. It does not abandon a cart. It does not care about reward points. It cares about the lowest-cost rail that clears with adequate finality.
Once consumer agents control a non-trivial share of purchasing — Citrini projects this happening in early 2027 — merchants gain a counterparty that will accept any rail with acceptable settlement guarantees. Stablecoins, FedNow, RTP, account-to-account pulls, even closed-loop merchant credit. Each one undercuts interchange.
The signature is already in the pipe
You do not need to accept the 2028 scenario to see the early signal. AI-mediated commerce is already routing around several friction layers documented in the Federal Trade Commission's 2025 surveillance pricing study: cart-abandonment dynamics, individualized pricing, and the entire user-experience layer the networks depend on.
When the agent is the customer, the abandoned cart is a bug, not a revenue opportunity. The discount that used to be triggered by an exit-intent popup never has to fire because the agent is already comparing rails.
Citrini's projection: by March 2027, the median American is consuming roughly 400,000 tokens per day, ten times the volume of late 2026. That is the quiet metric that should make a Mastercard equity analyst nervous. Tokens are the unit of agentic action, and agentic action is exactly what does not need a Visa logo on a checkout page.
Compare this to the trajectory of Chinese payments over the last decade. UnionPay was the dominant rail. Then Alipay and WeChat Pay rebuilt the consumer-merchant interface as a QR code, bypassed the interchange model entirely, and reduced merchant fees by an order of magnitude. The networks did not get disrupted by a better network. They got disrupted by a different layer of the stack absorbing the function. Agentic commerce is the same shape, applied globally.
What 3.4% growth actually means for the franchise
Mastercard has compounded revenue at roughly 12% to 15% per year for a decade. The market values the stock as if that compounding is structural. A step-down to 3.4% does not look like a bad quarter. It looks like a regime change.
The downstream balance sheet effects are mechanical. Card-network revenue funds a large share of bank fee income. Bank fee income supports consumer credit underwriting. Consumer credit underwriting supports the white-collar mortgage book that Citrini flags as $13 trillion structurally impaired by 2028. Each layer assumes the layer below keeps compounding at trend.
The networks will fight back. They will buy stablecoin issuers, partner with agent platforms, lobby for regulation that requires a "trusted card-network rail" inside agent commerce. Some of that will work for a quarter or two. None of it changes the underlying physics: a commerce layer where the buyer is a piece of software does not need a brand at checkout, because there is no checkout.
If interchange is a tax on human shopping friction, what is the equilibrium fee when the shopping is done by an agent that experiences no friction at all?