This article is authorized to be reproduced by Automatic Insight, author: Rhythm Editorial Department, copyright belongs to the original author.
When PayPal announced its collaboration with OpenAI, allowing ChatGPT users to complete payments directly within conversations, the market's reaction was almost instantaneous, with PYPL soaring 15% in pre-market trading and closing with its best performance in months.
The significance behind this collaboration goes far beyond stock price fluctuations; AI is redefining the pricing for all payment companies.
In a future where AI can also spend money, the moat for payment companies is no longer just about licenses, channels, and rates, but rather whether they can abstract payment rules, risk control, identity, and settlement into capabilities that can be called upon by agents.
Whoever can teach AI to spend money in a compliant, secure, and accountable manner the fastest will regain a premium in the new pricing system. Companies that miss this window may be compressed into low-profit pipelines.
Why does "AI + Payments" reprice the existing payment system, and what costs are involved behind this? What costs is the entire industry bearing to teach AI how to spend money?
The change in valuation frameworks comes from changes in cost structures. Making payment companies' capabilities "API-enabled, accountable, and auditable" so that AI can safely spend money on behalf of people incurs costs far exceeding a single equipment purchase.
By 2025, just four tech giants—Alphabet, Meta, Microsoft, and Amazon—are projected to spend over $360 billion on AI infrastructure. Alphabet's budget was initially set at $75 billion at the beginning of the year, raised to $85 billion by mid-year, and again to $93 billion by year-end.
Overall, Silicon Valley's investment in AI is expected to reach $400 billion in 2025, primarily for building data centers, purchasing GPUs, and laying networks, aimed at enabling AI to understand language, recognize images, and generate content.
But this is just the foundation; teaching AI how to spend money on behalf of humans is another significant expense.
Currently, we cannot provide an exact figure because it is not a one-time capital expenditure but a continuous, multi-dimensional gamble. It consists of visible direct investments and invisible indirect costs, which we can estimate through different dimensions.
First, the visible investments, with the most direct expenses coming from the "arms race" in protocol development.
Top engineers, product managers, and legal advisors from giants like Google, Visa, and Mastercard are working day and night, with a singular goal: to make their designed protocols the universal language of the future AI economy.
Based on Silicon Valley's salary levels, an average team of several hundred people incurs annual operating costs of hundreds of millions of dollars. Considering that the development and promotion cycle for these protocols typically takes 2-3 years, the total industry investment for this alone could reach tens of billions of dollars, not including the substantial marketing and business expenses incurred to promote the protocols.
Following this is the complete overhaul of existing infrastructure. To support AI agent payments, thousands of banks and payment institutions worldwide must undergo "open-heart surgery" on their outdated systems. This includes establishing new API interfaces, deploying more robust anti-fraud models, upgrading authentication systems, and increasing real-time monitoring capabilities, among others.
According to a report by Autonomous Research, while AI technology can save financial services companies up to 22% in operating costs in the long run, it requires massive upfront investments. For the entire industry, this transformation cost is conservatively estimated to be in the hundreds of billions of dollars.
Meanwhile, venture capital is flooding into this sector, trying to capture the next generation of payment giants. According to a report by S&P Global, in the third quarter of 2025, AI agents and stablecoins became two major hotspots for fintech financing.
Let’s look at some real cases.
Ramp is a company providing autonomous financial services that completed two rounds of financing in just 45 days, with its valuation soaring from $16 billion to $22.5 billion, showcasing the market's enthusiasm.
Circuit and Chisel is an agency commerce platform that secured $19.2 million in seed funding. Lava Payments is a company providing payment systems for agency-native companies, raising $5.8 million in seed funding.
In addition, companies like Kite AI, Basis Theory, Kira Financial AI, and Neurofin AI have also secured funding, and this list continues to grow. Conservatively, the total venture capital flowing into AI payment-related startups is expected to exceed $5 billion by 2025.
However, these visible investments may only be the tip of the iceberg compared to the invisible costs.
The largest cost is the "opportunity cost," which is the disruption of traditional business models.
McKinsey pointed out in its report "The Disruption of Retail and SME Banking by Agentic AI" that nearly half of the over $27 trillion in annual revenue from the global payment industry comes from "customer inertia," where people lazily or unknowingly leave large sums of money in low-interest checking accounts or forget to redeem credit card points.
The emergence of AI agents will completely end this era of easy money.
A rational AI agent would act like an tireless personal financial advisor, monitoring your accounts in real-time and automatically transferring idle funds into the highest-yield savings products.
It would choose the credit card with the highest cashback for you every time you make a purchase; it could even bypass traditional credit card networks entirely through open banking A2A (account-to-account) payments, thereby eroding the core revenues of issuing banks and card organizations.
McKinsey predicts that in Europe alone, if 10-20% of consumers adopt AI agents, banks' net interest margins could narrow by 30-50 basis points, resulting in revenue losses of billions of dollars. Globally, this figure could reach hundreds of billions of dollars.
In summary, the industry reshaping to teach AI how to spend money, including protocol development, infrastructure transformation, and startup financing, may involve direct investments in the hundreds of billions of dollars, while indirect costs such as opportunity costs and revenue losses could reach hundreds of billions or even trillions of dollars.
This is a "tuition fee" that no single company can bear; the entire industry is collectively footing the bill.
Having AI buy you a cup of coffee sounds simple, but the actual operation is a complex project involving technology, law, business interests, and security.
First is the standard dispute. When AI is to represent you in payments, it needs a universal protocol to communicate with merchants, banks, and payment networks, but the question is who will define this protocol.
In May 2025, Coinbase launched the x402 protocol, attempting to activate the long-dormant HTTP 402 status code (Payment Required) to establish a Web3-based, decentralized payment standard. Its core idea is that each transaction is verified through cryptographic signatures, ensuring security through code rather than centralized institutions.
The advantage of this approach lies in its openness and permissionless nature, allowing any developer to innovate based on this protocol. However, its disadvantages are also apparent; under the existing legal framework, once disputes arise, determining responsibility and tracing accountability can be very difficult.
In September, Google, along with over 60 companies, released the AP2 protocol, proposing a completely different solution—Mandates. This scheme is more like creating encrypted digital contracts for each transaction by AI under the legal framework of the real world, ensuring that every transaction is traceable and accountable.
Google's advantage lies in its vast ecosystem and good relationships with regulatory bodies, making the AP2 protocol more easily accepted by mainstream merchants and financial institutions. However, its problem is centralization; Google may exploit its position as a standard setter to gain unfair commercial benefits.
Following this, in October, Visa and Mastercard released their respective protocols on the same day. The two giants of the traditional financial world declared their presence in this crucial standard dispute.
Their proposals are more like incremental improvements to the existing payment system, attempting to incorporate AI agents without disrupting the current interest structure. Their advantages lie in their deep understanding of payment business and large existing user bases, but their disadvantages include slower innovation speeds, which may be replaced by more aggressive solutions.
Behind this protocol dispute is a struggle for control over future commercial infrastructure. Whose protocol becomes the standard will determine who can play the role of "tap" in the AI economy and extract "tolls" from every transaction.
Next is the challenge of permission control. When you authorize AI to spend money on your behalf, boundaries need to be set: how much can it spend? In what scenarios can it spend? Does it require user confirmation? If the AI is hacked or encounters a bug, who is responsible?
These questions seem simple, but they are extremely complex in terms of technical implementation, requiring the establishment of a finely-tuned permission management system that ensures AI has enough autonomy to complete tasks while preventing it from going out of control.
This requires a complete overhaul of the underlying payment system, adding new API interfaces, authentication mechanisms, real-time monitoring systems, and emergency shutdown mechanisms.
For example, Visa's proposal includes a "Programmable Spending" module that allows users to set complex payment rules through natural language, such as "Spend a maximum of $50 on coffee each week, but if it's at my regular store, it can go up to $70."
The third challenge is determining responsibility. When AI buys the wrong item on your behalf or pays the wrong amount at the wrong time, who bears the responsibility? Is it the AI developer, the payment service provider, or yourself?
Traditional payment systems are built on the assumption of "human decision-making," and the legal framework is designed accordingly. But when the decision-maker becomes AI, the entire chain of responsibility needs to be redefined. This involves modifying legal terms, designing insurance products, and establishing dispute resolution mechanisms, each requiring significant time and financial investment.
Currently, most protocols attempt to address this issue through detailed logs and traceable digital signatures, but when disputes arise, how to convert these technical evidences into legal evidence remains an unresolved issue.
Finally, there is the redistribution of commercial interests. In traditional payment systems, banks, card organizations, payment networks, and merchants each occupy their own positions, sharing transaction fees. But when AI agents enter this ecosystem, they will become a new, powerful intermediary.
They can choose the most favorable payment methods for users, bypass traditional credit card networks, and conduct real-time price comparisons and negotiations. This means that the originally stable distribution of interests will be disrupted.
Whoever can occupy a favorable position in this new ecosystem will be able to share in the future trillion-dollar market. Those players who cannot adapt may be completely marginalized.
To give a simple example, you ask an AI agent to help you plan a trip for next month. For an AI agent, this involves a series of complex transactions. It must find and book flights, reserve hotels that match your preferences, and rent a car.
Each step involves making separate payments to different suppliers, possibly using different currencies, and adhering to various cancellation policies and refund rules. The agent needs a secure way to manage multiple payment credentials, handle multi-step transaction logic—such as only booking the car after both the flight and hotel are confirmed—and deal with potential refunds or disputes.
Building a system that can reliably and securely handle this complexity and scale to millions of users and billions of transactions is a massive engineering challenge.
The entire payment industry is deeply anxious.
In the era of AI agents, traditional payment companies face a harsh reality: they are no longer the protagonists of transactions but may become mere "plumbers" in the background.
Users no longer care whether they are using PayPal or Stripe, as AI will automatically choose the optimal payment method. Merchants also no longer need to place various payment buttons on their websites, as AI will complete transactions directly.
In this scenario, what is the value proposition of payment companies? How can they maintain their position in the value chain?
PayPal's answer is to deeply integrate with OpenAI, becoming the default payment method for ChatGPT.
This is a high-risk strategy. If ChatGPT becomes the mainstream entry point for AI agents, PayPal could gain significant traffic and transaction volume. However, if other AI platforms rise or OpenAI decides to handle payments itself, PayPal's investment could go to waste.
Other payment companies are also making their own choices, with Visa and Mastercard opting to launch their own protocols in an attempt to seize the initiative at the standards level.
But regardless of the path chosen, these companies face the same anxiety: in an AI-driven payment ecosystem, their profit margins will be significantly compressed.
What is even more unsettling is that the power to lead this transformation lies with AI companies, not payment companies.
Tech giants like OpenAI, Google, and Microsoft possess both AI technology and vast user bases, along with ample funding. They can easily bypass traditional payment companies and establish a new payment system themselves. In fact, Google's AP2 protocol has already demonstrated this ambition.
In this context, traditional payment companies can only participate in this new ecosystem as much as possible to avoid being marginalized. This is precisely the logic behind PayPal's collaboration with OpenAI.
In this reshuffling of the payment industry led by AI, the future landscape of players is gradually becoming clear. At the top are AI platforms like OpenAI, Google, and Microsoft. They control AI technology and user access, making them the absolute leaders of the new ecosystem. They may treat payment capabilities as infrastructure for their platforms, charging developers and merchants, similar to how Apple takes a cut from every transaction in the App Store.
The middle layer consists of infrastructure service providers like Visa and Mastercard. They provide the underlying payment networks and clearing services, acting as the "pipes" of the AI economy. They may earn slim profits from processing massive transactions, but they no longer have the power to define the rules. They are more like highway maintenance companies rather than owners.
At the bottom are vertical solution providers, such as startups like Ramp and Kite AI. They offer customized AI payment solutions for specific scenarios or industries. They may find their survival space in niche markets, but it is difficult for them to challenge the status of platform companies. They are the "gold miners" in the new ecosystem, hoping to carve out a share in this transformation.
Traditional payment giants like PayPal find themselves at an awkward crossroads. They lack the technological advantages of AI platforms and do not control the underlying networks like Visa and Mastercard.
Their only advantage lies in their large existing user base and merchant relationships, but in this new era, these advantages are rapidly eroding. They resemble the former Nokia, holding a massive market share in feature phones while watching the arrival of the smartphone era.
The surge in PayPal's stock price on October 28, 2025, may just be the prologue to this transformation. Hundreds of billions of dollars in direct investments and trillions of dollars in revenue redistribution all point to the same conclusion—a world of payments dominated by AI is on the horizon.
In this new world, those companies that once defined payment rules are becoming the enforcers of those rules.
Related: Opinion: The changing demographics of cryptocurrency users pose new demands for crypto security.
Original: “How Much Did It Cost to Teach AI How to Spend Money?”
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
