Author: David, Shenchao TechFlow
On March 18th, another blockchain mainnet went live.
It's called Tempo, backed by Stripe and Paradigm. Stripe is one of the world's largest online payment companies, processing $1.9 trillion in transactions last year; Paradigm is one of the largest venture capital firms in the crypto industry. The two jointly invested $500 million in Tempo last year, valuing the project at:
5 billion.
A $5 billion blockchain, not speculating on cryptocurrencies, not doing DeFi, not issuing memes. On the day of the mainnet launch, Tempo's most high-profile product was:
Enabling machines to pay machines.
This might sound abstract, but you can think of it as AI now having to pay for every step. Calling an API costs money, buying computing power costs money, pulling a batch of data from a database costs money...
But the existing payment systems are all designed for humans. Bank accounts require ID cards, credit cards require facial recognition, Alipay requires SMS verification codes.
AI can't pass any of these.
It can help you complete an entire workflow, but when it comes to the payment step, it has to stop and wait for a human to press "confirm."
So, launched alongside the mainnet is an open protocol called MPP (Machine Payments Protocol), co-developed by Stripe.
Simply put, it sets rules for transactions between machines, including how to request payments, how to authorize, how to settle, etc.
The envisioned scenario is that AI can autonomously spend money within preset limits without needing human approval for every transaction. On launch day, over 100 service providers had already integrated, including OpenAI, Anthropic, and Shopify.
But Tempo isn't the only one doing this this week.
Within five days, Visa established a new department and launched an AI payment tool, Coinbase's payment protocol underwent a major upgrade, Mastercard acquired a stablecoin company for $1.8 billion, and Sam Altman's World released a toolkit specifically for AI identity verification.
Five giants rushed through the same door in one week, eager to open bank accounts for AI.
Two Paths, One Door
Tempo is helping AI with settlements. But settlement is just one part of the payment system. For an AI Agent to truly spend money autonomously, it also needs payment tools, funding channels, and identity verification.
Here, traditional payment companies and crypto companies are competing for the cake using their respective strengths.
On March 18th, the same day Tempo's mainnet launched, payment giant Visa also moved. Its newly established Crypto Labs department released its first product: Visa CLI, a tool that allows AI Agents to initiate credit card payments directly from the terminal.
No API key needed, no pre-registration required. If an AI needs to buy some service while running a task, it can pay with one line of command. Visa calls this "command-line commerce."
Visa's global card network connects billions of cards and tens of millions of merchants. If AI payments can run on this existing network, it doesn't need to wait for any new infrastructure to mature.
Visa is extending the old path. Its competitor Mastercard chose another way: buying the path outright.
On March 17th, Mastercard announced the acquisition of London-based stablecoin infrastructure company BVNK for $1.8 billion. This is the largest stablecoin acquisition in the history of the crypto industry.
The purpose of this acquisition is straightforward: if AI payments are going to use stablecoins, then let those stablecoins flow through my pipes.
On the crypto-native side, the moves are equally密集 (dense).
Coinbase's x402 protocol completed a major upgrade, expanding its payment scope from a few stablecoins to all ERC-20 tokens, while also releasing the MCP toolkit, allowing developers to integrate AI tools into the payment network with one click.
The two sides seem to start from different points, but their actions point in the same direction: Traditional payment companies are embracing crypto, crypto companies are embracing AI. Ultimately, crypto infrastructure is becoming the underlying pipeline for AI payments.
One环节 (link) remains. AI can spend money, but how does the merchant know if there's a human responsible for the AI spending the money?
On March 17th, World, co-founded by Sam Altman, released AgentKit, integrated with Coinbase's x402. It does only one thing: allow the AI to prove, while paying, that there is a verified real person behind it. The merchant can confirm that someone is responsible for the transaction but cannot see who that person is.
Five days, five companies. Settlement, channels, tools, protocols, identity. Every环节 (link) has been卡位 (staked out).
AI Cake Divided, Only the Cashier Left
Over the past three years, most positions on the AI industry chain have been taken.
The model layer is the table for OpenAI, Anthropic, Google, and a number of Chinese companies. Computing power is locked down tightly by Nvidia. The application layer is a bloody red ocean from programming assistants to search engines...
Every layer is crowded, and the competitive barriers in each layer are getting higher.
But the payment layer is still relatively vacant.
It's not that no one thought of it, the timing just wasn't right. AI Agent payment has a prerequisite: AI must first have the ability to independently complete an entire task chain. If it can only chat, doesn't need to call APIs, doesn't need to buy computing power, doesn't need to hire other Agents to work, then payment is not a rigid demand.
Over the past year, this prerequisite has slowly been established.
OpenClaw allows AI to directly operate computers, the MCP protocol allows AI to access external services, and the Agent capabilities of various large models saw concentrated breakthroughs in the second half of 2025. AI changed from a "conversation tool" to a "tool that gets work done," and getting work done costs money...
The demand for spending money has arrived, but the infrastructure for spending money does not yet exist.
This is why Stripe, Visa, Mastercard, and Coinbase acted simultaneously. For traditional payment companies, this is the first time they have gained a home-field advantage in the entire AI wave. They can't make models, they can't make chips, but payment is what they've been doing for decades.
Visa's global card network connects billions of cards and tens of millions of merchants, Mastercard covers over 200 countries, Stripe processed $1.9 trillion in transactions last year. If every AI expenditure flows through these pipes, the more capable AI is, the more money they make.
For crypto companies, the logic is somewhat different.
Coinbase CEO Brian Armstrong once said something very direct: "AI can own a crypto wallet, but it can't open a bank account."
Every step of the traditional financial system confirms "who you are." Opening a bank account requires an ID card, applying for a credit card requires facial recognition, every transaction requires an SMS verification code. AI is software, not a person; it can't pass any of these checkpoints.
But crypto wallets don't need these. A private key is an account. For an AI Agent, on-chain payment is the path of least resistance.
Crypto or not, AI payment will be a new infrastructure-level market. The difference is only whose pipeline is more suitable for machines.
Road Built, Cars Not Here
The story seems all set here, five giants are in place.
But there's a number worth looking at.
Coinbase's x402 protocol is currently the earliest landed and widest-reaching AI payment protocol. According to x402scan data, the total transaction volume of the entire ecosystem in the past 24 hours is $65,400. 150,000 transactions, averaging less than 50 cents per transaction.
What infrastructure is paired with this number? Tempo is valued at $5 billion, Mastercard spent $1.8 billion to acquire BVNK, Visa specifically established a new department, Stripe personally wrote the protocol.
Billions in valuation infrastructure, serving a daily transaction volume akin to a roadside bubble tea shop.
All infrastructure businesses seem to have this常态 (normal state).
On the eve of the 2000 internet bubble, telecom companies laid millions of kilometers of fiber optic cables under the sea. After laying them, they found that global internet traffic only used 5% of it. Most of those companies went bankrupt, but the cables remained.
A decade later, video streaming and mobile internet filled those pipes. The road pavers didn't make money, but the road was real.
AI payment is at this stage now. The demand logic is sound: AI Agents are indeed becoming more capable, they indeed need to spend money autonomously, they indeed need a new financial infrastructure.
Everyone is on the starting line, but after the starting gun fires, they find that, for now, they are the only ones on the track.
As for whose road ultimately succeeds, and when the first truly autonomous AI transaction happens in your life, it might be faster than everyone expects, or it might be slower than everyone expects.
The only certainty is that this battle has already begun, and your wallet and mine might be the last to know.
