There are two types of businesses that are easiest to do in this world: one is selling shovels, and the other is selling water. During the gold rush, those who mined gold did not necessarily get rich, but those who sold shovels definitely did not lose. Now, in the AI craze, those who write models may not make money, while those who sell tokens have already made a profit.

Written by: Wealth

0x0 Preface

In early May 2026, three events happened simultaneously: Sun Yuchen posted on X boasting about B.AI, the Trump family released WorldClaw at the Consensus conference, and Fu Sheng launched EasyRouter.io, proclaiming "85% off for everyone." If you put these three events together, no one would believe it, but it really happened.

A person selling bananas, a person selling meme coins, and a person selling browsers, all started an AI transfer station in the same week.

This feels a bit like three people from different provinces going to the same supermarket on the same day to buy the same brand of bottled water—you can’t help but feel there might be something special about this bottled water.

0x1 What is an API Transfer Station, in Layman's Terms

Let’s clarify this first.

Companies like OpenAI, Anthropic, and Google package their large models as APIs, charging developers per token. One token is about half an English word or a third of a Chinese character. You ask the model a question, it gives an answer, consuming thousands of tokens in back-and-forth, which may cost you less than a penny or a few dimes.

The problem is that these official APIs come with barriers: you need a credit card, an overseas phone number, access to overseas networks, and they have rate limits. Domestic users find it troublesome to use, and many businesses that use a large amount want discounts.

Thus, the transfer station has arrived. The logic is simple—I buy API quotas in bulk from the official source, you come here to buy from me, and I make a profit from information arbitrage and service fees. In industry terms, this is called "Token Import," or more colloquially, it's an AI wholesaler.

But this business model itself isn't new and doesn't have much technological content. What’s fresh is that it added a new outfit, called Web4.0, called the AI Agent economy, called blockchain infrastructure. With this new outfit, tokens can be easily sold—these aren’t API tokens, but cryptocurrency tokens.

This is an old business dressed in a new outfit.

0x2 Sun Yuchen's B.AI: Every Call is an On-Chain Transaction

Let’s start with Sun Yuchen since he was the first to act and is the most direct.

On May 1, 2026, Sun Yuchen posted on X:

The strongest AI transfer station in history, B.AI, has officially launched. One API Key = Claude + GPT + Gemini + the entire range of domestic large models. Blockchain login, pure anonymous payment, zero tampering, lowest price on the internet.

This wording is very precise. It connects mainstream models like GPT-5.5 and Claude Opus 4.7, claiming zero price difference from the official sources, and it is said that one million users have already signed up. Just looking at the functional description, it sounds like a normal API aggregation service.

But the payment method reveals everything—it does not accept USD or RMB; it must use USDT and settle on-chain.

This detail is crucial.

TRON chain carries the largest circulation of USDT globally, and Sun Yuchen has control over this chain. Every time a user calls B.AI, a transaction occurs on the TRON chain, with fees, on-chain traffic, and ecological data all feeding into this system. In his own packaged language, this is called "Web3 Infrastructure," a "financial identity that gives AI Agents autonomous payment capacity."

Translated into plain language: every time you use my transfer station, you are feeding my public chain.

This narrative has been packaged into Web4.0. Sun Yuchen defines Web4.0 as: readable, writable, owned, plus AI autonomous execution. It sounds lofty, but the end product is: an API aggregation platform, mandatorily settling with on-chain stablecoins, turning every inference into a record on TRON.

The real calculation is very clear: the profit from API transfer is the first layer, the profit from transactions on the TRON chain is the second layer, and the narrative of "All in Web4.0" raises the valuation of TRX and the entire TRON ecosystem is the third layer.

It’s a tri-fold benefit; the transfer station is just the fish head.

0x3 The Trump Family's WorldClaw: Pricing the Mar-a-Lago Dinner

The Trump family's operation is more direct, almost spectacular.

On May 5, 2026, WLFI (World Liberty Financial, a crypto project co-founded by the Trump family) released WorldClaw at the Consensus 2026 conference, with its core product named WorldRouter—an API routing platform that aggregates over 300 models, claiming prices 30% lower than the official ones, allowing one account to access all models without KYC and without needing an overseas credit card.

This sounds no different from other transfer stations. But once the pricing list came out, the tone changed.

There are four package options: the lowest package at $9.9, the standard package at $99, the advanced package at $999, and the flagship package at $9999. The flagship package includes 1 million AI credits and a hardware device with "brand confidentiality, specification undisclosed"—the official website kindly notes "the image is for reference only, the actual product may differ," with expected delivery in the third quarter of 2026.

Additionally, some buyers will be randomly selected to join a private dinner at Mar-a-Lago with Donald Trump Jr.

This dinner lottery is the most honest part of the entire product. It directly tells you that what is being sold is not just an API but the faith consumption surrounding the Trump family's IP.

The payment method accepts only one type: USD1. This is a stablecoin issued by WLFI in March 2025, linked to the US dollar, backed by US Treasury bonds and dollar deposits, currently operating on Ethereum, BNB Chain, and Solana. To use WorldClaw, you must first buy USD1, and then use USD1 to buy credits. If you don’t want to spend money, you can lock up WLFI tokens in exchange for credits, with the Pro package requiring a lockup of 250,000 WLFI and the flagship package requiring 2.5 million.

The logic behind this structure is: use API services to pull traffic → enforce USD1 settlement, increasing the usage of the stablecoin → lock up WLFI to reduce market circulation, raising the token price → harvest through token economy. API profit is just the admission fee, the token economy is the main course.

Zach Witkoff said on stage:

“In the future, there will be trillions of AI agents running, they will need to make payments to each other, to businesses, but traditional bank rails are too slow for machines, and KYC is impossible for machines—they need USD1.”

This statement is not wrong. But he uses it to sell the $9999 package and the opportunity for a private dinner at Mar-a-Lago.

0x4 Who Else is in This Web3 Track

Sun Yuchen and the Trump family are by no means isolated cases. AI transfer stations that have added tokens are now an overcrowded track.

Gate.io launched GateRouter, aggregating over 25 models and supporting AI agents to make autonomous payments in USDT, with a logic highly similar to B.AI—transfer station entry, on-chain tax collection.

Heurist positions itself as a decentralized AI computing cloud, aggregating idle GPU resources worldwide and providing serverless inference APIs externally. It has its own ZK Layer 2 (Heurist Chain), has a HEU token, and supports x402 and ERC-8004 payment protocols, with API calls and agent-to-agent settlements all utilizing its chain. On the surface, it appears to be a DePIN computing network, but its core is still API transfer paired with token incentives.

Virtuals Protocol is known in the industry as the "Stripe for AI agents," enabling developers to tokenize AI agents to issue and trade on-chain—your agent itself is a token, holding a token equals holding a share of the agent. This logic goes a step further: API transfers are just one capability of the agents, and what is truly being sold is the anticipated appreciation of the agent’s token.

The common features of these projects are clear: aggregating model APIs, tying on-chain payments, and ultimately all have a token waiting for you.

From B.AI to WorldClaw to GateRouter to Heurist, the packaging is different, but the thread of tokens runs through them all.

The problem in this track is not the wrong direction, but rather distinguishing who is building infrastructure and who is using infrastructure narratives to sell tokens. Just like in the internet boom of 2000, where IDC companies pulling fiber to broaden bandwidth and portal websites listed with "internet concepts" were all called internet companies, but their essence was different.

0x5 Three Ways for Transfer Stations to Profit, and Why They Want to Wear the "Web4.0" Hat

Transfer stations make money in three layers, each layer independent, but together they form a complete business.

The first layer is information arbitrage. Overseas APIs have regional restrictions, price differences, and access barriers. Transfer stations exploit these frictions to obtain lower wholesale prices, selling to users at official retail or slightly lower prices, with the price difference being the profit. While claiming "zero price difference," in reality, there is a billing multiplier in the backend, and discreet changes are common. According to industry insiders, nearly half of the nine vendors are secretly padding the costs—you pay for the inference of Claude, but it might be running on a lesser equivalent.

The second layer is the hidden tax in the payment process. By abandoning credit cards and introducing on-chain stablecoins or proprietary tokens, every call generates traffic within your ecosystem. Earn gas fees, earn on-chain transaction data, and earn revenues from internal circulation.

The third layer is the token economy's profit extraction. Issuing or binding their own tokens, reducing market circulation through lock-up mechanisms, thereby increasing secondary market prices, and ultimately cashing out when token prices are high. WLFI's net revenue sees 75% flowing directly to entities associated with the Trump family.

When these three layers are combined, it forms the "All in AI, All in Web4.0" real business model.

So why wear the "Web4.0" hat?

Because without this hat, tokens won’t fetch a high price.

An API transfer station is just a transfer station, with limited profit margins, and regulation is always watching; users can switch platforms without cost. But if this is "next generation internet infrastructure," the "financial base for the AI agent economy," the "payment protocol layer of the Web4.0 era"—the valuation logic changes, the narrative can attract more investors, tokens can be sold at higher prices, and the cost of user migration is also psychologically higher.

The value of the concept lies in making you feel that if you miss this opportunity, you miss the whole era.

0x6 What True Infrastructure Looks Like

Having talked about the scythe, let’s talk about the real shovel.

The AI agent economy genuinely needs foundational infrastructure, but infrastructure comes in two forms: payment protocol layers and computational power layers. Both are being seriously developed but are not as lively as selling tokens.

An example of a payment protocol layer is x402. An open protocol launched by Coinbase in May 2025, based on HTTP status code 402—this code has been dormant in the HTTP protocol for nearly thirty years, originally meaning "Payment Required," and has never been activated due to the lack of programmable instant payment mechanisms. The explosion of blockchain stablecoins and AI agents has finally given it a place to be used. The logic is very simple: client sends a request, server returns 402, along with the amount and payment address, client completes on-chain payment, sends a receipt, and the server releases—entirely without accounts or KYC. Cloudflare, Google, Stripe, AWS, Visa, and Mastercard are all integrating it, and it has been handed over to the Linux Foundation for maintenance. By the end of 2025, the weekly processing volume was already nearing one million transactions. It is open-source, free, with no tokens requiring lock-up.

On the x402 protocol, two noteworthy application projects have already emerged.

Pay.sh is an AI agent payment gateway released by the Solana Foundation in partnership with Google Cloud on May 5, 2026, built directly on x402 and MPP protocols, with a completely open-source registry. It has integrated official APIs from Google Cloud such as Gemini, BigQuery, and Vertex AI, as well as over 50 community API providers, including Dune Analytics, Helius, and The Graph. AI agents use Solana wallets for identity, pay stablecoins per request, with no need for accounts or API keys, and settlement on Solana happens within seconds, with providers receiving fiat currency. To summarize: it turns enterprise-grade APIs from Google Cloud into items on a shelf that agents can directly purchase.

Kite AI (KITE) is a Layer 1 public chain designed specifically for the agent economy, launched its mainnet in late April 2026, running on Avalanche architecture. Its goal is to give every AI agent a "passport"—the Kite Agent Passport, which includes verifiable encrypted identity and programmable consumption permission control, addressing the compliance issue of "agents making autonomous payments without humans knowing how much they spent or where." x402 serves as its native payment primitive, and PayPal Ventures and General Catalyst led a $33 million Series A, with Coinbase Ventures participating, and pilot integrations with PayPal and Shopify. It is not an API transfer station; it is the infrastructure layer that issues IDs for the machine economy.

For the computational power layer, there are three projects worth discussing separately.

Bittensor (TAO) is the oldest, and being called "the Bitcoin of the AI field" is not without reason—its total supply is 21 million tokens, with a halving mechanism, driven by "useful computation is mining": anyone can contribute AI models, computing power, or data to the network and earn TAO rewards. The network is divided into hundreds of subnets, each focusing on a specific AI task vertical, and subnets can call upon each other to form service chains. It is not a transfer station; it is an experiment attempting to use token incentive mechanisms to piece together a decentralized brain from global AI computing power and models. In May 2026, TAO re-entered the market's view, with daily transaction volumes exceeding $200 million. Whether it can succeed is another matter, but its purpose is not to resell APIs.

Phala Network (PHA) specializes in privacy computing. Its core technology is TEE (Trusted Execution Environment)—your data enters Phala's nodes and is computed in a hardware-isolated encrypted space, where even the node operators cannot see the content of the data, with results being verifiable on-chain. In November 2025, it completed migration from the Polkadot parachain to Ethereum L2, and currently processes tokens of large language models exceeding a billion daily. AI inference hides a lot of sensitive data; in healthcare, finance, and law, no one is willing to send raw data to a centralized API transfer station for processing. Phala’s direction is: you can use AI without needing to hand over data.

Fluence (FLT) is a decentralized computing market. It aggregates the GPU resources of top global data centers into a network, allowing developers to purchase computing power on demand, with prices about 80% lower than AWS and Azure. The demand for FLT tokens is directly tied to the platform's computing consumption—determined by the actual GPU hours used, not by narrative. By the end of 2025, the platform's total revenue exceeded $1 million, operating on over 1,400 GPUs across 32 regions and 71 data centers. This is not an API transfer station; this is genuinely building a decentralized computing foundation.

The distinction between these projects and B.AI and WorldClaw is clear: they are building the infrastructure that others use to create things, rather than being an intermediary reselling others' products to you. The x402 protocol itself does not issue tokens, and OpenRouter has built a platform processing hundreds of trillions of calls without ever having a token. TAO, PHA, and FLT have tokens, but the tokens are tied to real contributions of computing power and network security stakes, not "lock up 2.5 million to get a Max package."

0x7 Finally, a Few Honest Words

The AI agent economy is a real direction, and the micro-payment demand between machines is genuine, as is the narrative of "next-generation internet payment layers."

The only question is: who is really building this thing, and who is using this narrative to sell tokens?

The judgment method is very simple: look at this product, and ask if, without the token, it could still survive?

B.AI without USDT settlement is still an API transfer station, it can survive, but it loses its value tied to Sun Yuchen's ecosystem. WorldClaw without USD1 and WLFI lock-up is still a regular API aggregator; the $9999 package would have no reason to exist.

Fluence, without FLT, still has 1,400 GPUs running globally, saving customers $4 million in cloud computing costs is still real.

Bittensor, without TAO, still has models in the subnet performing inferences and competition still occurs. Phala, without PHA, still has TEE privacy computing capabilities, and the fact that it processes more than a billion tokens daily won’t disappear. x402, without a token, is still adopted by Cloudflare and Google.

This is the way to distinguish infrastructure from reapers: does it still exist without the token?

This is not to say that having tokens is bad; on-chain stablecoins and decentralized computing incentives are both genuine needs. The problem is, payment tools and token speculation are two different matters. The former is a pipeline, while the latter is the water seller installing a toll gate on the pipeline while also selling pipeline stocks.

When someone simultaneously sells you API services, stablecoins, and lock-up tokens, there is one thing in the mix that is your principal.

Which one, you ask?