x402, maybe we all misunderstood its original meaning.

Written by: Charlie Little Sun

If you are following the fintech, crypto, or AI circles, you must have seen the term “x402” everywhere.

A number of internet giants are entering the agentic payments arena: Coinbase and Cloudflare have teamed up to establish the x402 Open Fund, Google has integrated crypto expansion into its AP2 standard, and even the usually steady Adyen has rolled up its sleeves to join in. Visa's TAP promises to achieve interoperability, while Stripe and OpenAI have collaborated to make ACP and “instant checkout” industry standards.

An almost forgotten internet status code—402 “Payment Required”—has surprisingly become the key to determining the business model of the next generation of AI intelligences.

In fact, this undercurrent has been surging within the payment circle for more than half a year.

In February this year, when I wrote the article “AI Agents: The Next Frontier Reshaping the Future of Payments” on Substack, I noticed that star seed round projects like Skyfire and Payman—backed by payment giants from Web2 and Web3 such as Visa and Coinbase—are already laying out their plans in this field.

Moreover, companies like Stripe, Visa, and Paypal have mentioned the prospects of combining AI intelligences with payments in their annual reports, albeit with limited coverage.

But in recent weeks, x402 has truly broken out, thanks to the major moves by AI giants and fintech behemoths. x402 itself, along with the broader trend of intelligent agent commerce, has finally attracted the attention of a wider technology and investment circle.

Many comments have mentioned that it originates from the HTTP 402 status code, and also analyzed how the rising AI agents and the gradually mainstreaming cryptocurrency technology are fueling this topic.

But they generally overlook a fundamental driving force: what is the real motivation behind this wave of AI + crypto?

The starting point of everything is the survival crisis of AI companies.

On the surface, everyone is talking about “smart agents finally being able to make payments autonomously,” and from this, they are imagining what new skills AI agents could unlock in the future.

However, from a closed-door conversation I had with a prominent investor in Silicon Valley, I learned that the root of all this, the real demand pain point, actually comes from the survival pressure of large model companies like OpenAI and Anthropic.

They are telling a trillion-scale future story to the capital markets while watching copyright lawsuits pile up. Public skepticism regarding the ethics of their training data is rising. If a scalable and automated way to compensate for the data and databases they crawled cannot be established, subsequent financing and huge capital expenditures will face pressure.

One claim I've heard is that OpenAI's Sam Altman reached out to Coinbase's Brian Armstrong—both of whom come from the YC ecosystem. This may also explain why Coinbase briefly launched the AgentKit product earlier this year. It was a toolkit aimed at developers, which included support for the OpenAI SDK, enabling agents to have trading capabilities.

However, AgentKit did not last long - it was soon taken offline. Not because the intuition was wrong, but because the paths diverged, and the strategic focuses of the two companies quickly went in opposite directions.

Perhaps under the pressure of the collaboration between Perplexity and Shopify, OpenAI's growth strategy has begun to lean towards consumer e-commerce—becoming the interface for people to discover and purchase products—therefore it tends to favor a checkout solution that merchants can adopt without much hassle.

On the Coinbase side, due to the GENIUS Act and a series of crypto enforcement actions in the United States, the ambition has gradually turned into an internet currency infrastructure, which naturally pushes it towards “machine-to-machine” traffic, meaning payment is made directly by robots to pages or interfaces without the need to create accounts.

Through the lens of first principles, observe the formation of four major camps by the giants.

Although AgentKit was short-lived, it laid the foundation for the subsequent upgraded version launched by Coinbase — x402, and opened up new ideas for other major players.

It is precisely because of the original driving force mentioned above that we can better understand the competitive and cooperative relationships among the camps formed by the giants through this first principle:

Stripe × OpenAI: Capturing consumer scenarios, achieving “smooth upgrades”

The ACP protocol co-built by OpenAI and Stripe has been launched for instant checkout in ChatGPT. Essentially, it is an “elegant upgrade” to the existing payment networks.

Stripe issues a shared payment token for specific orders and merchants, which is carried by the agent to complete transactions, while the merchant side continues to use familiar fraud prevention, refund, and tax processes—without the need to restructure teams or retrain.

Etsy has already been integrated, and Shopify is also in line waiting. The core of this strategy lies in leveraging existing channels and usage habits: seamlessly embedding smart agent transactions into the payment pipeline that merchants have already established.

Coinbase × Cloudflare: Aiming at the machine economy, making “underlying reconstruction”

The upgraded version x402 of Coinbase is even more ambitious. It makes the “402 Payment Required” status code in the HTTP protocol executable: the server declares the price along with the accepted tokens, the client completes the payment (preferably the money-making machine USDC on its beloved Base chain) and continues the operation with proof. No account, login, or monthly billing is required.

This solution shows its advantages in micropayment scenarios ranging from $0.01 to $0.1—such as per-call API usage, per-article context retrieval, and per-use analysis tools—similar to the early demo of its investment Skyfire.

Cloudflare's previously launched “pay-per-crawl” has paved the way for this, and Coinbase's strong entry has further united key players such as AWS, Circle/USDC, and NEAR. Notably, Anthropic can be considered a perfect substitute for OpenAI, the original pioneer.

Compared to OpenAI's direction towards consumer shopping, Anthropic's core strategy MCP framework focuses on enabling agents to discover and call tools on demand, and its economic model naturally aligns with x402's “pay-per-request” model, far surpassing consumer settlement aimed at storefronts.

Google × Adyen: Establishing rules to ensure compliance and auditing

Google's AP2 standard aims to tackle the most challenging authorization issues in payment systems. It binds the user's intentions (such as price limits, frequency restrictions, and category ranges) to specific actions through a set of signed authorization commands. This design is inherently agnostic to payment channels and includes a cryptographic extension that allows agents to seamlessly connect to the x402 link when processing very small payments, while maintaining complete audit trails.

In addition, the participation of payment giant Adyen means that this system has enterprise-level dispute resolution and compliance capabilities from day one.

Visa TAP: No new tracks, inclusive and integrated.

Visa's TAP does not attempt to rebuild the payment rails, but instead plays a key compatibility role: helping issuers and acquirers identify agent transaction flows and apply consistent risk and dispute rules.

The key lies in the positioning of interoperability - TAP clearly states that it will align with the ACP of the Stripe+OpenAI camp and interconnect with the x402 of the Coinbase+Cloudflare camp, genuinely implementing its positioning of Network of Networks, placing itself in an invincible position.

With the “Agent Pay” from Mastercard, you'll find that bank card brands are guiding standardization rather than erecting walls around intelligent agent commerce.

Although these solutions seem to be fighting their own battles, they are actually forming a pattern of division of labor and cooperation, with layers complementing each other in practice:

Authorization Layer: The AP2 standard from the Google+Adyen camp provides verifiable credentials of user intent, becoming a “passport” for all operations.

Execution Layer:

Facing consumer shopping, the ACP + traditional card network of the OpenAI + Stripe camp offers a smooth experience.

Faced with micro-payments between machines, the x402 + cryptocurrency from the Coinbase + Cloudflare camp has minimal costs.

Coordination Layer: The Visa camp's TAP and others inform banks behind the scenes that “this is a smart agent transaction, not fraud,” ensuring smooth processing.

A typical scenario might be: your travel agent first queries the ticket prices of multiple airlines through x402 for 2 cents (subject to AP2 authorization constraints). After you confirm, it switches to the ACP channel to complete the ticket purchase, while TAP simultaneously explains the situation to the bank. The same task, two payment tracks, one set of audit logs.

In this future system, interoperability will be a core element, which is why the most important point emphasized by various industry participants at last week's Federal Reserve Payment Technology Conference was the interoperability of future payment systems.

Who is the winner in the smart contract payment race?

Existing giants won the first round because identity verification, fraud, refunds, taxes, and dispute resolution are all scale games. ACP/TAP/AP2 directly connect to a mature system that reassures CFOs.

But if we return to the original motivation that initiated this whole matter: OpenAI attempted to pay content owners. In fact, such an interconnected intelligent agent commercial ecosystem also benefits the “long tail” participants.

Long-tail participants win in scenarios where the unit amounts are extremely small and previous cumbersome processes would stifle transactions: API requests by the instance, obtaining context by the article, using tools on a per-use basis.

Many startups are emerging as additional value areas beyond the three basic layers mentioned above—such as the bonding layer—building cross-protocol identity + payment solutions, allowing small teams of two or three people to implement metered billing without the need to establish a billing department (doesn't this sound a lot like Stripe's original value proposition?).

Furthermore, if we think outside the online space - when we begin to realize that the greatest limitation of AI is actually the real world, I believe there are at least three areas where agents combined with x402/AP2/ACP/TAP will significantly change economic models:

The Prediction Market and Oracle Revolution: If DeFi oracles adopt AP2 authorization + x402 payments, it will incentivize data sources to provide more real-time and higher-quality data, fundamentally enhancing market efficiency and liquidity, and opening up greater possibilities for the recently booming prediction markets.

AI Data Centers and Grid Resilience: The demand for training computing power fluctuates significantly, and edge storage devices and EV fleets can sell surplus electricity to data centers at a per-second rate through x402, transforming rough “power rationing” into a precise “bidding market.”

Climate data and DePIN: Thousands of sensors can become data centers, selling readings of air quality, temperature, flood levels, etc. to municipal agencies or insurance companies on a per-use basis through x402, with AP2 authorization ensuring the scope of data usage, and all transactions are auditable.

Summary: The victory of division of labor and cooperation.

Looking back at the starting point, OpenAI was initially just trying to solve the “original sin of copyright” - to pay a reasonable price for the data used.

But the ultimate answer is not a single technological breakthrough, but a sophisticated division of labor in the industry:

OpenAI+Stripe's ACP makes large-scale, compliant consumer transactions possible.

Coinbase+Cloudflare's x402 provides the infrastructure for fragmented payments in the machine economy.

Google+Adyen's AP2 establishes trusted authorization and auditing standards.

Visa's TAP alleviates concerns about the traditional financial system.

Coinbase ultimately formed a deep bond with Anthropic because the microeconomic models of MCP and x402 are inherently compatible; OpenAI chose to work closely with Stripe in consumer scenarios because that is where growth can be seen most quickly and trust can be established. Both started from the same pain point, chose different battlegrounds, and ultimately wove together a more resilient ecological network.

For startup teams, the value-added levels and services outside the “authorization–execution–coordination” core framework formed by these four camps present a good opportunity for engagement.

Even outside the online realm, there are significant opportunities in areas such as the ability of prediction markets and oracle systems to reflect real-world data, the power demand of AI data centers, and the demand for weather data resulting from climate disaster risks.

If there is any copyright dispute regarding the reprinted article, please contact us for deletion.