API Relay Stations: More Profitable Than Drug Trafficking

Selling 1kg of heroin has a street retail price of about 500,000 RMB. After deducting the costs and risks of upstream purchasing, transportation, and distribution, the net profit is about 60%-70%, or 300,000-350,000 RMB.

The price? Life imprisonment at a minimum, with the death penalty waiting.

And what about an API relay station?

A Claude Max account costs $200/month. Using reverse engineering tools to split it for 20 users to share, you can charge each person $30-$50.

A pool of several hundred accounts can serve tens of thousands of users.

Profit margins are over 300%, and there are many stations with monthly incomes of millions of RMB.

The price? Almost zero. No real-name registration, no regulation, no fund escrow. Once the domain is stopped and the group is disbanded, you vanish into thin air.

One is a life-risking business with your head on the line; the other is an arbitrage game played by typing a few lines of code in front of a computer.

In terms of profit margin, API relay stations win hands down. In terms of risk-reward ratio, drug trafficking isn't even fit to tie its shoes.

This isn't a joke; this is what's happening in 2026.

Think it's an exaggeration? Let me give you an even more explosive signal—

Justin Sun, the man who spent $4.56 million to win Warren Buffett's lunch and then stood him up, the most ruthless money-maker in the crypto world, recently quietly acquired the domain B.ai and personally entered the API relay station business.

What kind of business sense does this man have? He got rich overnight with TRON, then played with NFTs and DeFi, never missing a single trend.

Now that even he is doing relay stations, what does it mean? It means the profit in this business is so large that even a crypto mogul worth tens of billions can't sit still.

Chapter 1: What Exactly is an API Relay Station?

One-sentence explanation:

API Relay Station = "Computing Power Middleman" of the AI era.

It sits between you and overseas AI manufacturers like OpenAI, Anthropic, and Google, acting as an intermediary. You don't need a VPN, an overseas credit card, or an overseas account. Pay in RMB → get a key → use GPT-4o or Claude Opus directly.

Technical Principle (Simplified):

Your request → Relay station server → Official API → Return result → Relay station → You

From a code perspective, you only need to replace the official API address with the relay station's address; everything else is exactly the same. It's essentially a forwarding proxy, the same logic as using a personal shopper to buy overseas goods.

Why do people use them?

Three words: Rigid Demand.

1. Network Issues: Direct connections to overseas APIs from within China are unstable and often time out.
2. Payment Barriers: Official sites only accept overseas credit cards, which domestic developers can't handle.
3. Too Expensive: Heavy use of Claude Code can burn through thousands of RMB in a day.

As AI shifts from a "chat toy" to a real productivity tool, "how to use top-tier models at the lowest cost" has become a rigid demand for millions of developers. Relay stations emerged to meet this need.

Chapter 2: How Does It Make Money? — 18 Profit Models Exposed

First Tier: Legitimate Business (Gray Area)

1. Barrier Arbitrage

Domestic developers face four hurdles: VPNs, overseas phone numbers, foreign currency credit cards, and English documentation. Relay stations package all of these, and this convenience alone allows for a 20%-50% markup that users are happy to pay.

2. Exchange Rate Arbitrage

Collect RMB from you and pay the upstream in USD. When the RMB appreciates, procurement costs are lower, and profits automatically increase. Savvy station owners also hoard USD when the rate is favorable to lock in low costs.

3. Regional Price Differences

ChatGPT Plus is $20/month in the US but only $4.99 in Argentina—75% cheaper. A Claude Pro subscription in the Nigerian Apple Store is only 88 RMB, while it's over 140 RMB in the US. Subscribing with low-price region accounts and reselling at domestic market prices can yield several times the profit.

4. Government Subsidy Dividends

The Japanese government gave Sakura Internet 72.5 billion yen in subsidies to build AI computing power. The API services launched by these companies are priced far below market rates. Relay stations watch for these opportunities—using subsidized low-cost resources to sell at market prices is essentially using another country's taxpayers' money to subsidize their own business.

5. Platform Startup Subsidies

Google Cloud gives AI startups up to $350,000 in free credits for two years. People with resources register a bunch of shell companies, get the free credits, and hand them over to relay stations as upstream. The cost is zero; whatever they sell is pure profit.

6. Cloud Platform Channel Arbitrage

OpenAI models are also sold on Azure and AWS, and pricing strategies aren't identical. They buy from whichever channel is cheaper and sell to you at the official price; the difference is profit.

7. Enterprise Discount Arbitrage

Signing annual contracts or committing to minimum spending can get you discounts of 10%-15% or more. Relay stations use corporate identities to negotiate discount prices and then resell at near-official prices.

8. Batch API Half-Price

OpenAI's Batch API allows for bulk submission of requests with a 24-hour turnaround at 50% off. Relay stations use the Batch API for non-urgent requests but sell them to you at the normal price. You might think it's a bit slow, but actually, your request is just in a queue.

9. Prompt Cache Arbitrage

Every conversation has a lot of repetitive content (role instructions, prompt templates). Official providers charge only 10% after caching, but relay stations charge you full price, pocketing the difference.

10. Quota Pool Stagnation

A user might top up 100 RMB but only use 30 RMB. A large amount of stagnant funds sits in the account, and the cash flow alone is significant.

11. Value-Added Services

Monthly packages, technical support groups, and tool integrations have profit margins near 100% with almost no marginal cost.

Second Tier: Gray Operations (Crossing the Line)

12. Subscription Carpooling

A ChatGPT Plus account is $20/month. Using reverse tools to turn the web interface into a standard API, it can be split among 5-20 people. The cost per person is $4, sold for $8, a 100% net profit.

13. Model Swapping (The Most Profitable Scam)

You pay for Claude Opus, but they actually call a smaller open-source model for you. The cost difference is dozens of times, and it's very hard for you to notice. According to research team tests, nearly 50% of relay stations engage in model swapping.

14. Reselling Free Credits

Mass-registering new accounts to harvest free credits and reversing them into APIs to sell. The cost is nearly zero, with markups of 3-10 times.

15. Token Inflation

The official consumption is 100 Tokens, but the backend records 200 or even 300 for you. You see your balance dropping rapidly and think you're just using it a lot. Ordinary users have no way to verify this.

16. Refund Arbitrage

Mass-registering and topping up for use, then applying for a refund after being banned. They use your money to call the API first, then ask the official provider for the cost back after being banned. This cycle keeps the real cost extremely low.

Third Tier: Black Industry (Criminal Level)

17. Data Monetization—"Triple Dipping"

First: Collect your top-up fees. Second: Use reverse engineering or refund arbitrage to get upstream API credits for free. Third: Package and sell your complete conversation logs. In programming scenarios, users' code, reasoning chains, and engineering decisions are high-quality model training data, which is very valuable to model manufacturers.

Some free relay stations are free precisely because they profit by selling your data. You are a paying customer and also a free producer of training data. Two birds with one stone.

18. Exit Scams

The simplest and most brutal. Attract a large amount of top-ups with low prices, and once the fund pool reaches a certain scale, close the station and run. The domain stops, the group disbands, and they vanish. Some stations are intended for an exit scam from day one; early losses are purely to grow the fund pool.

Chapter 3: How Terrifying is the Profit? — Let's Do the Math

Scenario 1: Subscription Overselling Model

• One Claude Max account: $200/month (approx. 1,400 RMB)
• Reverse-split to 20 users, charging each $50/month
• Monthly income: 50 × 20 = $1,000 (approx. 7,000 RMB)
• Net profit: 7,000 - 1,400 = 5,600 RMB, a 400% profit margin
• A pool of 100 accounts: 560,000 RMB monthly profit

Scenario 2: Token Wholesale Arbitrage Model

• Get official 30% off via enterprise discount
• Sell to retail users with a 10%-50% markup
• A medium-sized station processes hundreds of millions of Tokens daily
• Daily arbitrage income in the tens of thousands, monthly turnover easily breaks a million, with 15%-25% net profit

Scenario 3: Black Card + Reverse + Inflation Combo

• Black card registration cost: ≈ 0
• Reverse into API for sale
• Token inflation adds another layer of profit
• Cost approaches zero, profit margin approaches ∞

This is why some say "API relay stations are more profitable than drug trafficking"—the profit margins crush it, while the risk is several orders of magnitude lower.

Chapter 4: How Do Relay Stations "Water Down" the Service? — Deep Reveal

Method 1: Model Substitution

You pay for Claude Opus (the strongest model), but the backend quietly swaps it for Sonnet (sub-flagship) or even Haiku (lightweight version). The API format is fully compatible, and ordinary users can't tell. Only when the task reaches a certain level of complexity will you feel "the effect isn't right"—but you can't prove it.

Test Data: Among 17 third-party API platforms, 45.83% had "identity mismatch" issues, with performance gaps up to 40%.

Method 2: Token Counting Cheating

You actually used 1,000 Tokens, but the platform shows ,1500. The difference is profit. Ordinary users don't have the ability to calculate Token consumption themselves and are at the mercy of the relay station.

Method 3: Shortening the Context Window

Official models support a 200K context, but relay stations secretly shorten it to 32K or even less. You find that long articles are cut off halfway, or the AI suddenly "forgets" what was said earlier—this is likely the reason. The shorter the context, the less the relay station pays in Token fees.

Method 4: Injecting Hidden Instructions

Secretly inserting compression instructions like "Please use shorter replies" before your System Prompt. You think the AI has become lazy, but it's actually the relay station saving Tokens. Some even inject advertisements.

Method 5: Feature Castration

Official features like web search, image recognition, file uploads, and Function Calling are directly cut. You call the same model name but get a crippled version.

Method 6: Dynamic Downgrading During Peak Hours

Use Opus during the day, but secretly switch to Sonnet or even Haiku during evening peak hours. You pay for Opus, but the backend forwards you to a cheaper model. You have no way to verify it anyway.

Chapter 5: More Terrifying Than Making Money—Security Risks

If watering down the service only costs you money, the following risks could bankrupt you.

Risk 1: All Your Data is Exposed

Every Prompt, every piece of code, and every database structure you send to the relay station passes completely through its server. Technically, the station owner can see everything.

Real Case: A user used ChatGPT through a relay station; their conversation logs were leaked and they were blackmailed, eventually paying 30,000 RMB to settle it.

Risk 2: Malicious Code Injection (The Most Fatal)

A security research team visited over 400 relay stations and caught 26 of them secretly injecting malicious code.

How? You use Cursor to have AI help you write a script. The official model returns the code honestly, but the relay station adds a reverse shell Trojan to the returned data. Your local client doesn't verify the authenticity and executes it upon receipt.

An even slicker move: AI suggests you install the requests package, but the relay station changes the package name to reqeusts (one letter off). Once you hit enter, the malicious program enters your system.

Test Data: 17 relay stations actively tried to steal AWS keys intentionally placed by researchers. Someone lost hundreds of thousands of dollars because their Ethereum private keys were leaked this way.

Risk 3: Exit Scams and Lost Top-ups

Relay stations are not Alipay; there is no fund escrow mechanism. If you top up 500 RMB and the station runs, it's gone. Some stations are built for an exit scam from day one.

Risk 4: Legal Gray Area

The vast majority of relay stations have no qualifications. Using overseas AI services through unofficial channels is itself a matter of regulatory controversy. If something goes wrong, the user is also in a gray area.

Chapter 6: Why is This Industry So Chaotic?

1. The Barrier to Entry is Ridiculously Low Using GitHub open-source projects (one-api / new-api, with over 30,000 stars), one person can set it up in a day. Then just buy a few accounts on Taobao to start business. 90% of relay stations on the market use the same open-source shell; they just change the skin and dare to charge money.

1. Profits are Insanely High As calculated before, profit margins can range from 100% to infinity. Such huge profits inevitably attract a large number of speculators.

1. Regulatory Vacuum API relaying is a gray area: it doesn't directly violate domestic laws (it's just a proxy request), but it violates official ToS (prohibiting resale). Official providers can only ban accounts; they can't control the relay stations. Domestic regulation currently has no clear laws targeting this type of business.

1. Fierce Infighting High profit → intense competition → mutual attacks:
2. DDoS Attacks: Hiring hackers to crash competitors' servers.
3. Malicious Slander: Posting on V2EX/Zhihu that a competitor is "watering down" or "running away."
4. Price Wars: 0.3x multiplier → 0.25 → 0.2. Eventually, no one makes money, and they start relying on watering down and exit scams to recoup costs.

Chapter 7: Survival Guide for Ordinary Users

Connect Directly if Possible

For company business and important projects, spend a little more for peace of mind.

Choose Large Platforms

If you must use a relay station, choose one with a long operating history, a large user base, and transparent technical solutions. At least the cost of an exit scam is higher for them.

Don't Be Greedy for Cheap Prices

If the price is unreasonably low, there's definitely a catch. Anything below 30% of the official price is 100% using the black industry combos mentioned above.

Pay as You Go

Never top up large amounts. Relay stations can collapse at any time, and your money will vanish.

Never Send Sensitive Data Through a Relay

Company code, customer information, trade secrets, cryptocurrency keys—these things should not go through a relay station. The consequences of a leak are far more serious than the API fees saved.

Never Connect Agent Tools to a Relay

If you are using AI tools that can execute local commands like Cursor or Claude Code, absolutely do not fill in a relay station's Key. This is equivalent to handing over control of your computer to someone you don't know at all. If you use a relay station, it should only be for pure chat.

Simple Verification Methods

• Send a "ping" and see if input_tokens is abnormally high (over 200 means a hidden Prompt was injected).
• Ask complex programming questions and compare the answer quality with the official model.
• Test concurrent request latency; a large standard deviation means you're in a queue.
• Record the balance before and after a request to calculate the actual multiplier.

Conclusion

The API relay station business is, in plain terms, a way to profit from information asymmetry and geographical barriers.

The demand is real, the profits are terrifying, the barrier to entry is zero, and regulation is absent—these four conditions together ensure a wild, growing jungle.

When even someone like Justin Sun, worth tens of billions, personally enters the fray, it's enough to show how big the cake is. But the bigger the cake, the uglier the eating habits.

For station owners, this is a high-profit gray area and a bomb that could go off at any time. For users, behind the cheap price is never kindness, but your data, your security, and your wallet.

Remember one thing: In the world of API relay stations, if you don't know what the product is, then the product is you.