Bernstein: ChatGPT Wallet Launches, Visa and Mastercard Gain First-Mover Advantage in AI Payment Tokens
nashnova research
OpenAI launched the ChatGPT wallet, choosing payment tokens from Visa, Mastercard, and Amex as its payment backbone; Bernstein sees this as a material step for the AI-agent payments ecosystem and rates both Visa and Mastercard outperform.
How does the ChatGPT wallet actually work?
A user enters card details once. Each time the AI agent — a program that shops online on your behalf — is authorized to buy something, the payment network generates a unique payment token (an encrypted stand-in for the real card number) to settle the transaction.
This means → the AI agent never sees your actual card number; security sits with Visa, Mastercard, and their networks.
Existing dispute resolution, rewards, and merchant checkout flows all carry over. In plain terms = every perk you get when you swipe your own card still applies when AI shops for you.
Why did OpenAI pick card networks only — no third-party wallets?
The ChatGPT wallet currently supports only Visa, Mastercard, and Amex tokens. PayPal, Shop Pay, and other digital wallets are excluded.
By contrast, Meta's Muse launched with Stripe Link and Shop Pay, and plans to add PayPal — a clearly different approach.
Bernstein reads this as OpenAI aiming to build its own strategic position in consumer payments, much like Apple kept Apple Pay walled off from rival wallets for years.
What can payment tokens earn for Visa and Mastercard?
The value goes beyond "keeping AI purchases on card rails." Once tokens are widely adopted, Visa and Mastercard can sell value-added services — transaction security, identity verification, risk management — to AI platforms and merchants.
Bernstein points to Stripe's fee model for shared payment tokens as proof that tokens themselves can be monetized.
In a bull case where AI-agent transactions reach a mid-to-high single-digit share of total consumer spending, the business could add several percentage points to Visa's and Mastercard's growth.
How early-stage is AI-agent shopping right now?
Uber disclosed that out of roughly 320 million transactions in Halloween week, only 504 were completed by AI agents — essentially a rounding error.
Visa's shopping-capability test found that mainstream AI models passed just 57.2% of tasks. Worse, among the tasks they failed, about 54% still resulted in an actual purchase.
In plain terms = AI agents not only get it wrong often, they also spend money when they haven't figured out what to buy — refunds and disputes will be a major issue.
Why are Visa and Mastercard well-placed to handle those risks?
Transaction disputes, refunds, and returns are precisely the domain where Visa and Mastercard have decades of mature infrastructure.
This means → their consumer-protection machinery from traditional card spending extends directly into AI payment scenarios — a barrier new entrants cannot easily replicate.
Mastercard stated on its earnings call that one-time virtual cards are not a long-term solution, because they sever the link between a transaction and the consumer's existing rewards and spending controls.
What comes next?
Bernstein rates Visa and Mastercard outperform, with price targets of $486 and $764 respectively — implying roughly 15.6% and 17.1% upside.
The firm flags three checkpoints: ① how quickly issuing banks and merchants actually integrate AI payment tech; ② how widely AI-agent shopping is adopted during the year-end holiday season; ③ whether Google and Meta's Muse also adopt purpose-built AI-agent payment tokens.
This reflects a core thesis: whoever sets the token standard dominates the AI payments ecosystem — and right now, Visa and Mastercard are running in front.
市场有风险,内容仅供研究参考,不构成投资建议。
