AI Agents Are Spending Money Now: Virtual Cards Power Agentic Payments

Visa's bet: let AI spend for you

In June 2026, Visa and OpenAI officially unveiled their partnership: Visa Intelligent Commerce. This is not another memo. BBVA completed the world's first AI-agent-initiated card payment. Santander ran an end-to-end agentic payment flow in Latin America. CaixaBank followed. Animoca Brands finished a pilot in Hong Kong.

The core idea is simple. You give an AI agent a virtual card, technically a set of tokenized card credentials. It compares prices, places orders, and pays on your behalf. You set the budget and the rules.

Sounds convenient. But if you have used PokePay, FotonCard, or any virtual card platform to subscribe to ChatGPT Plus or run Google Ads, you probably already sense the problem. When the card is no longer typed in by a human but called automatically by software, how does fraud detection react?

Robinhood's AI credit card: 3% cashback, but who is spending

In May 2026, Robinhood opened its platform to AI agents. You can let AI trade stocks for you and buy things with your credit card. They even launched a credit card specifically for AI agents, with 3% cashback.

At the same time, Stripe went public with the payment infrastructure it built for agents. AI agents can spend real money, fully autonomously. This is not a demo. It is production.

For the virtual card industry, this creates a new demand stack:

• Layer 1: Humans subscribing to AI services (ChatGPT, Claude, Midjourney). This is the mature scenario today. VCard or DogPay handles it fine.
• Layer 2: AI agents executing payments on behalf of humans. Requires tokenized credentials, granular spend limits, and transaction audit logs. Most existing virtual card platforms are not ready.
• Layer 3: Fully autonomous agent spending. Requires trust frameworks, dispute resolution, and identity verification. Only Visa and Mastercard network layers are working on this.

Most users today sit at Layer 1, transitioning toward Layer 2.

Tokenization: from BINs to credentials

The core technology both Visa and Mastercard are pushing is tokenization. Your card number is not handed to the AI agent directly. Instead, it gets swapped for a set of one-time or purpose-limited tokens. Each transaction maps back to your real card account, but the agent never sees the original PAN.

This is the same logic behind Apple Pay and Google Pay. The difference: Apple Pay tokens are bound to your phone's biometric authentication. AI agent tokens need to be bound to authorization policies. How much per day, which merchants are allowed, which categories are blocked.

For existing virtual card BINs, this is a fork in the road. Common BINs like 556150 and 531993 may need to support dynamic token generation down the line. Some platforms are already moving. Most are still stuck at giving you a card number and telling you to fill it in yourself.

Practical impact: will your virtual card get declined

Short answer: not yet. The Layer 1 scenario, where a human manually types in a card number to subscribe to an AI tool, is completely unaffected.

But if you start using automation features in AI tools, like having AI buy API credits in bulk or auto-renew multiple SaaS subscriptions, you may run into upgraded fraud checks. In 2026, users have already reported that the same virtual card gets flagged by 3D Secure when it fires off multiple automated payments in quick succession.

A few practical responses:

• Use different cards for different purposes. Keep subscriptions and spending separate. Avoid a single card showing an abnormal transaction pattern.
• Check whether your card platform supports sub-cards or spend isolation. Crospay and BitMart, with their multi-BIN architecture, have an edge here.
• If you run Google Ads or Meta Ads, AI-driven automatic bidding and billing is increasingly common. Make sure your virtual card limit is high enough, or your campaigns will stall.

Who is fighting for this market

Visa is not the only player. Mastercard is pushing Agent Pay in parallel, running pilots with multiple banks. The two giants are fighting over the same thing: who defines the payment standard for AI agents.

Fortune put it bluntly in June: Visa, Mastercard, and Coinbase are fighting over how AI agents pay. The competition is not about the cards themselves. It is about tokenization protocols, authorization frameworks, and the settlement layer.

For users, this is good news. Competition means fees go down and features go up. But in the short term, standards are fragmented. Different platforms and different AI tools may have compatibility gaps.

The numbers behind this

According to Grand View Research, the global prepaid card market exceeded 3 trillion dollars in 2025, with open-loop prepaid cards taking the largest share. Virtual cards are the fastest-growing segment within this market.

Fortune Business Insights projects through 2034 with 2025 as the base year. Their report notes that mobile payment and digital wallet penetration is driving sustained virtual card demand. Agent-based payments will become a new growth driver in 2026 to 2028, though the exact magnitude cannot be reliably quantified yet.

What vccdir users should do

If you mainly use virtual cards for AI subscriptions and cross-border payments, do not panic. Your current cards still work. But three things are worth getting ahead of now.

First, find out whether your card platform supports tokenization. Ask support directly: do you support Apple Pay and Google Pay? If yes, the underlying tokenization capability exists, and the barrier to future AI agent integration will be lower.

Second, start using a multi-card strategy. Do not tie every AI subscription to the same card. If one card gets blocked or triggers a fraud check, at least your other services keep running.

Third, track the Visa Intelligent Commerce rollout timeline. BBVA and Santander enterprise customers are already on it. A consumer-facing version will likely launch in late 2026 or early 2027. By then, whether your virtual card can spawn sub-tokens for AI agents will become a key criterion for choosing a platform.