《算力伺服器外銷：USDT、現金、第三國主體，三條路最後都撞同一堵牆》Overseas Compute Hardware Sales: USDT, Cash, a Third-Country Buyer — Three Roads, One Wall

[Zheng Ge's Take]

Lately I have been comparing notes with a few people who sell compute hardware overseas, and the same problem keeps turning up: the goods are wanted, the money is offered, and then the structure lands on the table and you can see at a glance that it will not clear.

I do not know which particular deal any of this comes from, and I am not going to describe one. But the structure is worth setting out plainly, because a buyer turning up with an arrangement of this kind is not rare in the overseas compute-hardware trade. A buyer will offer three routes: skip the dollar wire and use a stablecoin; or take cash in Hong Kong; and if neither of those works, have a third-country company sign the contract and make the payment. All three sound like payment problems, as though changing the method would open the road. My judgement is that all three hit the same wall, and that wall has nothing to do with how the money moves.

Act One: Three Routes on the Table, One Step All of Them Skip

All three routes share one feature: each of them avoids the step of saying clearly who the end user is.

The buyer will usually hand over incorporation documents, director records and corporate bank details, and will even fly in to sign. But the moment the question becomes which data centre the goods end up in, and which entity actually operates them, the answers go vague. Sometimes it is that the client will arrange it; sometimes it is a commercial arrangement; sometimes it is that you do not need to know. That vagueness is the whole story, not the currency.

After years in export trade and compute delivery, I accept that some customers genuinely cannot disclose commercial detail. But "I would rather not discuss the commercial detail" and "I will not say who uses it" are two different sentences. The first can be worked with. The second cannot.

Act Two: USDT — the Money Arrives, Its Origin Does Not

What attracts a buyer to a stablecoin is that it is instant, borderless and never touches a bank. Those are precisely the three features that should stop a seller taking it.

First, money arriving on-chain carries no provenance. Before a transfer reaches you it may have passed through many hands, one of them fraud or gambling money, and the account you use to convert it gets frozen, followed by an investigation. Second, the transaction is irreversible. Once the goods have gone, if the counterparty denies the deal, disappears, or adds a condition at the last minute, there is no lawful route to recover anything. A bank transfer at least leaves a document, a named counterparty and a path that can be traced.

Third, there is no lawful place to put it. A mainland entity cannot take USDT as payment for goods. Even through a Hong Kong company, taking virtual assets as trade consideration on a habitual basis requires a VASP licence, which is a high bar, and mainland residents cannot open that kind of account directly. So this is not a high-risk route. It is a route with no lawful landing place at all.

Act Three: Cash — You Change the Paper, Not the Counterparty

Some buyers then step back from coin: no crypto, just US dollars in cash, handed over in Hong Kong, money against goods. I will say it directly. That does not stand up. Three reasons, and the third is fatal.

Hong Kong's anti-money-laundering rules require KYC on large cash receipts in trade. If the counterparty cannot produce identity documents, you cannot verify the source of funds, which means walking an unexplained cash receipt onto your own company's books. The bank is the same story from the other side: pay a large cash sum into a corporate account and it will ask for full business documentation; without client identity papers it will refuse the deposit, and it may close the account.

The third point is the fatal one: OFAC looks at the counterparty, not at the payment instrument. Cash in your hand does not turn a sanctioned actual buyer or end user into an unsanctioned one; the secondary-sanctions exposure is identical. All cash does is make it harder, later on, to show that you did not know.

Act Four: The Third-Country Entity — Look-Through Asks About the End User

The third route looks the most professional: a third-country company signs, that company's own account sends the dollars by SWIFT, and the goods move to a third destination. The buyer will tell you the contracting party is not a sanctioned party, so the risk has already been isolated.

That reading is wrong. OFAC's review looks through: the nominal buyer is only the starting point, and the review follows the actual controller down to the end user. Changing the country of the contracting shell does not make the exposure disappear; it only makes "did you know" a harder question to answer.

A genuinely compliant structure would have to satisfy several things at once. One, the signing company must supply full incorporation documents, director records and corporate bank details, and the money must come out of that company's own account by wire. Two, the buyer must disclose the end user and the end use in full, and the seller must independently verify that the signing company, its actual controller and the end user are all off the sanctions lists. Three, the signing company must itself have a real trading business in servers. Four, the three flows must be consistent — contracting party, paying party and receiving party all matching, with contracts, delivery evidence and bank records retained.

Of those four, the third is where the structure dies. Arrangements like this appear in the first place usually because the signing company has no such business, no staff and no trading history, and exists only as a channel around sanctions. The moment you check what it has actually traded, what it has sold, whether it has the capability to absorb goods of this kind, the shell shows itself. Items one, two and four are paperwork and process, and paperwork can be produced. Item three is history, and history cannot be produced.

So my conclusion is not that this structure is difficult. It is that it does not work. And a counterparty who puts the route forward while declining to name the end user is himself the clearest red flag in the file — not a problem for you to solve, but a signal for you to walk.

Act Five: The Real Threshold Is Three-Flow Consistency and Export Control

Set sanctions aside entirely. Even with a completely clean counterparty, a seller still has two jobs to do.

The first is three-flow consistency. If the contracting party, the paying party and the receiving party do not line up — the contract with company A, the money wired by company B, the goods delivered to C's warehouse — you need a coherent commercial explanation and complete documentation behind it. Without that, the bank asks first and the tax authority asks second. Keeping the full set of contracts, delivery evidence and bank records on file is the floor, not a bonus.

The second is export control. Advanced compute hardware is a controlled item. You check whether a licence is required before you quote, not after the deposit has landed. Get that order wrong and everything afterwards is damage control, and damage control rarely saves the deal.

My summary is short. On this kind of order the binding constraint was never the payment method. It is KYC and sanctions screening. If a counterparty will not say clearly who it is, then the question is not whether the deal is worth having; it is when it goes wrong. On a large order, paying a compliance adviser for a transaction risk review and keeping the paperwork complete is the cheapest insurance available.

Payment methods can be changed. Counterparties cannot.