Author: Fugui
0x0 Preface
In early May 2026, three things happened simultaneously: Justin Sun posted on X to promote B.AI, the Trump family released WorldClaw at the Consensus conference, and Fu Sheng launched EasyRouter.io, announcing "15% off everything." These three events, put together, seem unbelievable, but they actually happened.
A person who sold bananas, a person who sold meme coins, and a person who sold browsers, all started AI transfer stations in the same week.
This is a bit like three people from different provinces going to the same supermarket on the same day to buy the same bottled water—you can't help but feel that there might be something fishy about that bottled water.
0x1 What is an API Relay Station? Let's explain it in layman's terms.
Companies like OpenAI, Anthropic, and Google package their large models into APIs and sell them to developers on a token basis. One token is roughly half an English word or a third of a Chinese character. You ask the model a question, and it gives you an answer, consuming thousands of tokens back and forth, which might cost you less than a penny, or it might cost you a few cents.
The problem is that these official APIs have barriers: a credit card, an overseas mobile phone number, access to overseas networks, and speed limits. This is inconvenient for domestic users, and many companies with large usage volumes want discounts.
So, relay stations came into being.
The logic is simple—I get API quotas in bulk from the official website, you buy them from me, and I profit from the information gap and service fee. In industry terms, this is called "Token Import," or more colloquially, AI middleman. This business itself isn't new, nor does it involve much technology. What's new is that it's been given a makeover: Web4.0, AI Agent Economy, and Blockchain Infrastructure. With this makeover, it can conveniently sell tokens—not API tokens, but cryptocurrency tokens. It's a very old business, dressed in a very new guise. 0x2 Justin Sun's B.AI: Every Call is an On-Chain Transaction Let's start with Justin Sun, because he was the first and most direct to act. On May 1, 2026, Justin Sun posted on X: "The strongest AI hub in history, B.AI, has arrived for nothing. One API Key = Claude + GPT + Gemini + the entire series of domestic large-scale models. Blockchain login, purely anonymous payment, zero tampering, lowest price on the entire network." This statement is very accurate. It integrates mainstream models such as GPT-5.5 and Claude Opus 4.7, claims to have official zero markup, and supposedly has already surpassed 1 million users. Judging from the functional description alone, it seems like a normal API aggregation service. However, the payment method betrays everything—it does not accept USD, does not accept RMB, and must use USDT, with on-chain settlement. This detail is crucial. The TRON chain carries the world's largest USDT circulation, and Justin Sun controls this chain. Every time a user calls B.AI, a transaction occurs on the TRON chain, and transaction fees, on-chain traffic, and ecosystem data all flow into this system. In his own words, this is "Web3 infrastructure," a "financial identity that gives AI agents autonomous payment capabilities." In plain terms: Every click you make using my intermediary feeds my public chain. This narrative is packaged as Web4.0. Justin Sun defines Web4.0 as: readable, writable, ownable, plus AI autonomous execution. It sounds like a grand vision, but the tangible product is: an API aggregation platform that mandates settlement using on-chain stablecoins, turning every inference into a TRON record. The real calculation is clear: earning price differences through API intermediaries is the first layer; earning ecosystem fees through TRON on-chain transactions is the second layer; and boosting the valuation of TRX and the entire TRON ecosystem through the "All in Web4.0" narrative is the third layer. A three-pronged approach, with the intermediary platform merely the head of the fish. 0x3 The Trump Family's WorldClaw: Putting Mar-a-Lago dinners on the pricing list. The Trump family's actions are even more direct, bordering on spectacular. On May 5, 2026, WLFI (World Liberty Financial, a crypto project co-founded by the Trump family) launched WorldClaw at Consensus 2026. Its core product, WorldRouter, is an API routing platform aggregating over 300 models, claiming prices 30% lower than official rates, allowing access to all models with a single account, no KYC required, and no need for foreign credit cards. It sounds similar to other transit hubs. However, the pricing list changed drastically. Four tiers: $9.99 for the lowest tier, $99 for the standard tier, $999 for the premium tier, and $9999 for the flagship tier. The flagship tier includes 1 million AI credits and a hardware device whose brand is kept secret and specifications undisclosed—the official website thoughtfully notes that "images are for reference only, and the actual product may differ," with shipping expected in the third quarter of 2026. And, a random selection of buyers will be sent to Mar-a-Lago for a private dinner with Donald Trump Jr. This dinner raffle is the most honest part of the entire product. It directly tells you that this product isn't just selling APIs, but selling faith-based consumption surrounding the Trump family's IP. Payment is only accepted in one way: USD1. This is WLFI's stablecoin pegged to the US dollar, issued in March 2025, backed by US Treasury bonds and US dollar deposits, and currently runs on Ethereum, BNB Chain, and Solana. With WorldClaw, you must first buy USD1, then use USD1 to buy credits. If you don't want to spend money, you can lock up WLFI tokens to exchange for credits; the Pro tier locks up 250,000 WLFI, and the Flagship tier locks up 2.5 million. The logical chain of this structure is: use API services to attract traffic → force settlement with USD1, increasing stablecoin usage → lock up WLFI to reduce market circulation, driving up token prices → harvest through token economics. The API price difference is just the entry fee; the token ecosystem is the main course. Zach Witkoff said on stage: "In the future, there will be trillions of AI agents running, and they will need to pay each other and pay businesses, but traditional banking is too slow for machines, and KYC is simply impossible for machines—they need USD1." This statement itself is not wrong. However, he used it to sell a $9,999 meal package and a chance to win a prize at the Mar-a-Lago dinner.
0x4 Who Else Is in This Web3 Race?
Justin Sun and the Trump family are not isolated cases. AI intermediaries with tokens are already a crowded race.
Gate.io launched GateRouter, aggregating more than 25 models and supporting AI agents to make independent payments using USDT. Its logic is highly similar to B.AI—intermediary entry point, on-chain settlement and tax collection.
Heurist positions itself as a decentralized AI computing cloud, aggregating idle GPU resources globally and providing serverless inference APIs. It has its own ZK Layer 2 (Heurist Chain), its own HEU token, and supports x402 and ERC-8004 payment protocols. Developers calling models and agents settling with each other all use its chain.
On the surface, it's a DePIN computing network, but at its core, it's still an API relay plus token incentives. Virtuals Protocol, known in the industry as the "Stripe of AI agents," allows developers to tokenize their AI agents, issuing and trading them on-chain—your agent itself is a token, and users holding tokens are essentially holding shares in the agent. Taking this logic a step further, API relay is just one capability of the agent; what's truly being sold is the expected appreciation of the agent's token. These projects share a clear common characteristic: aggregated model APIs, linked to on-chain payments, and ultimately, a token awaits you. From B.AI to WorldClaw to GateRouter to Heurist, the packaging varies, but the token thread runs throughout. The problem with this sector isn't the wrong direction, but rather the inability to distinguish who is building the infrastructure and who is using the infrastructure narrative to sell tokens. It's like the internet boom of 2000: IDC companies building fiber optic cables and broadband, and portal websites raising capital through the "internet concept"—both were called internet companies, but their essence was different. 0x5 Three Ways Relay Stations Make Money, and Why Use the "Web 4.0" Label Relay stations make money in three layers, each independent, but together they form a complete business. The first layer is information asymmetry. Overseas APIs have geographical restrictions, price differences, and access barriers. Relay stations exploit these frictions to obtain lower wholesale prices, selling to users at the official retail price or slightly lower; the price difference is the profit. While they claim "zero markup," there are actually billing multipliers in the backend, and these are often secretly modified. According to industry insiders, nearly half of the nine suppliers are secretly diluting their offerings—you're paying for Claude's inference, but it might be running on a fake platform. The second layer involves hidden taxes in the payment process. Abandon credit cards, introduce on-chain stablecoins or your own tokens, and generate traffic within your ecosystem with every transaction. Earn gas fees, on-chain transaction data, and circulation revenue within the ecosystem. The third layer involves exploiting the token economy. Issue or bind your own tokens, reduce market circulation through lock-up mechanisms, drive up secondary market prices, and ultimately cash out at the peak of the token price. 75% of WLFI's net revenue flows directly to entities associated with the Trump family. These three layers combined constitute the true business model of "All in AI, All in Web4.0". So why label it "Web 4.0"? Because without this label, the token wouldn't fetch a high price. An API intermediary is just an intermediary; profit margins are limited, regulators are constantly monitoring it, and users have zero cost to switch platforms. But if this is "the next generation of internet infrastructure," "the financial foundation of the AI agent economy," or "the payment protocol layer of the Web 4.0 era"—the valuation logic is different. The narrative can attract more investors, the token can sell for a higher price, and the psychological cost of user migration is also higher. The value of the concept lies in making you feel that if you miss this train, you miss the entire era. 0x6 What does real infrastructure look like? Having discussed the sickle, let's talk about the real shovel. The AI agent economy truly needs underlying infrastructure, but this infrastructure comes in two forms: payment protocol layer and computing power layer. Both are being seriously developed, though not as much as the token-selling frenzy. The payment protocol layer is represented by x402. Coinbase's open protocol, launched in May 2025, is based on the HTTP status code 402—a code that has been shelved in the HTTP protocol for nearly thirty years. Originally meaning "payment request," it has never been used because there was no programmable instant payment mechanism. The combination of blockchain stablecoins and the explosion of AI agents has finally given it a place. The logic is extremely simple: the client sends a request, the server responds with a 402, along with the amount and receiving address, the client completes the on-chain payment, issues a receipt, and the server approves the transaction—the entire process requires no account and no KYC. Cloudflare, Google, Stripe, AWS, Visa, and Mastercard are all integrating with it. It has been handed over to the Linux Foundation for maintenance, and by the end of 2025, its weekly processing volume was approaching one million transactions. It is open-source, free, and requires no token lock-up. Two noteworthy application-layer projects have emerged on the x402 protocol. Pay.sh, an AI agent payment gateway released on May 5, 2026, by the Solana Foundation in partnership with Google Cloud, is built directly on the x402 and MPP protocols, with a completely open-source registry. It integrates with official Google Cloud APIs such as Gemini, BigQuery, and Vertex AI, as well as over 50 community API providers, including on-chain data services such as Dune Analytics, Helius, and The Graph. AI agents use the Solana wallet as their identity, paying in stablecoins on requests. No account or API key is needed; settlements on Solana are measured in seconds, and the provider receives fiat currency. In short, it transforms Google Cloud's enterprise-grade API into a marketplace where agents can directly make purchases. Kite AI (KITE) is a Layer 1 public blockchain designed specifically for the agent economy. Its mainnet launch is scheduled for the end of April 2026, running on the Avalanche architecture. What it does is issue each AI agent a "passport"—the Kite Agent Passport—containing a verifiable encrypted identity and programmable spending control, solving the compliance problem of "agents autonomously paying money, but humans not knowing how much they spent or where they spent it." x402 is its native payment primitive. PayPal Ventures and General Catalyst led a $33 million Series A funding round, with Coinbase Ventures also participating. PayPal and Shopify have pilot integrations. It's not an API intermediary; it's the infrastructure layer that issues IDs to the machine economy. There are three projects worth mentioning separately in the computing power layer. Bittensor (TAO) is the oldest, and it's not without reason to be called the "Bitcoin of AI"—with a total supply of 21 million tokens, a halving mechanism, and driven by "useful computation equals mining": anyone contributing AI models, computing power, and data to the network in exchange for TAO rewards. The network is divided into hundreds of subnets, each focusing on a specific AI task vertical, and subnets can call each other to form a service chain. It's not a transit point; it's an experiment attempting to use a token incentive mechanism to piece together global AI computing power and models into a decentralized brain. In May 2026, TAO returned to the market's attention, with daily trading volume exceeding $200 million. Whether it can work is another matter, but what it's doing is not API reselling. Phala Network (PHA) focuses on privacy-preserving computation. Its core technology is TEE (Trusted Execution Environment)—your data enters Phala's nodes and is processed within a hardware-isolated, encrypted environment. Even the node operators themselves cannot see the data content, and the results are verifiable on the blockchain. In November 2025, it completed its migration from a Polkadot parachain to Ethereum L2, and now processes over a billion LLM tokens daily. AI inference contains a large amount of sensitive data—medical, financial, and legal data—and no one wants to send their raw data to a centralized API intermediary. Phala's approach is: you can use AI, but you don't need to hand over your data. Fluence (FLT) is a decentralized computing power marketplace. It aggregates GPU resources from the world's top data centers into a network, allowing developers to purchase computing power on demand at prices approximately 80% lower than AWS and Azure. The demand for FLT tokens is directly linked to the platform's computing power consumption—determined by the actual number of GPU hours run, not narrative. By the end of 2025, the platform's total revenue exceeded $1 million, running on over 1400 GPUs across 32 regions and 71 data centers. This isn't an API intermediary; it's genuinely building a decentralized computing power foundation. The difference between these projects and B.AI and WorldClaw is clear: they are building the infrastructure that others use to create things, not creating a middleware layer to resell others' solutions. x402 itself doesn't issue tokens, and OpenRouter has never issued tokens yet still supports a platform handling trillions of calls. TAO, PHA, and FLT have tokens, but these tokens correspond to real computing power contributions and cybersecurity staking, not simply "locking up 2.5 million to get a Max plan."
0x7 Let me say a few words of truth
The AI agent economy is a real direction, the demand for micro-payments between machines is real, and the narrative of the "next-generation internet payment layer" is also real.
There's only one question: Who is actually building this, and who is using this narrative to sell tokens?
The method of judgment is simple: Look at this product, remove the token, can it still survive?
B.AI, without USDT settlement, is still an API transit station, it can survive, but it loses the value of being tied to Justin Sun's ecosystem. WorldClaw, without USD1 and WLFI lock-up, is still just an ordinary API aggregation, and the $9999 package has no reason to exist.
Even without FLT, Fluence's 1400 GPUs worldwide are still running, and the $4 million in cloud computing savings for customers are still real. Even without TAO, Bittensor's models within its subnets are still inferenced, and competition continues. Even without PHA, Phala's TEE's privacy computing capabilities remain, and the fact that it processes a billion tokens daily doesn't disappear. x402 never had a token to begin with, yet Cloudflare and Google still adopted it. This is how to distinguish between infrastructure and exploitation: does the business still exist after the token is removed? This isn't to say that having tokens is bad; on-chain stablecoins and decentralized computing power incentives are real needs. The problem is that payment tools and token speculation are two different things. The former is the pipeline, the latter is a water seller installing a toll gate on the pipeline while simultaneously selling pipeline stock. When someone tries to sell you API services, stablecoins, and locked tokens simultaneously, one of what they're selling is your principal. As for which one, can you guess?
Kikyo
Anais
Catherine
Joy
Alex
Miyuki
Weatherly