《深圳B300「水下盤」：一篇分析話九成實單唔見光，我點睇》Shenzhen's B300 "Underwater" Market: One Analysis Says Nine in Ten Real Deals Never Surface

[Zheng Ge's Take]

On 18 October I read a market analysis on Toutiao about Shenzhen's B300 business, headlined "Can't Make Sense of It: Shenzhen's B300 Market Is Badly Distorted, and Nine in Ten Real Deals Are Underwater." I read it twice.

Let me be clear about what this is. It is not my own field research; I have not gone around verifying title to goods on any of these deals, and I would not be able to. What I read is a market analysis written by somebody inside the trade. What I can do is use my own years in compute hardware to judge which of its sentences hold up and which need a discount.

I agree with part of what it says. Below I go through it section by section. Where numbers appear, I say plainly that they are the analysis's, and I do not pass them off as facts I have checked myself.

The core conclusion: the public price and the real trade are two separate tracks

The analysis's core conclusion is that Shenzhen's high-end B300 market has already split into two clearly separate tracks.

One track is the public market. It runs on lead-generation offers, forward and futures listings, and paper prices that look very attractive.

The other track is private settlement. That is where state-owned-enterprise projects, the hard requirement of intelligent computing centres, and the bulk lock-ups of the major integrators sit, and there the real transaction price has stayed firm.

Then comes the heaviest line in the piece. According to the analysis, more than ninety per cent of large real orders are closed-loop underwater deals.

In the article's own words: the lock-up moves of the top source channels, the long-term capital partners and the large integrators are never posted for quotation in public groups. They secure the goods in advance, match demand privately and settle inside a closed loop, and almost all the compliant, deployable stock settles inside the core circle of the trade.

There is one more line: what ordinary channels and small brokers can reach in the open, is less than ten per cent of the market's real stock.

Those two sentences are the analysis's judgments, not numbers I have verified. I have no way to check "ninety per cent" or "ten per cent" — underwater dealing is by definition not public, and anyone who claims to have counted it should be asked how. But I agree with the direction: in this market the public quote and the real transaction have become two different things.

Why cheap offers are everywhere and a delivered machine is not

The analysis then explains why cheap prices are everywhere online while nothing can actually be delivered. It sorts almost all the low listings into three kinds of delivery defect.

The first is the broker's empty order and the forward contract with no real title to the goods. What the seller holds is a price, not a shipment.

The second is defective stock: incomplete configuration, an unfinished chain, a machine that cannot be legally racked. It is cheap because it was never going to run.

The third is the pure lead-generation quote, priced to start a conversation. The number exists to make you call, not to make you a buyer.

As the article puts it: these low offers look tempting, but push them to title verification, contract signature, corporate settlement and the machine room, and almost all of them stop there.

Conversely, the article says the scarce good stock — traceable, invoiceable, available for immediate collection, ready to be racked compliantly today — has held its price firmly, with none of the sharp fall that the internet keeps reporting.

This is the part I agree with most. In all my years in electronics and compute hardware one rule has never changed: the destructive power of a quote is in inverse proportion to its ability to deliver. The less a seller intends to deliver, the lower he dares to quote, because he does not have to answer for a closed deal.

The three kinds of people in this market

The article sorts the trade into three groups. The division is crude, but it is not wrong.

The first group is the core circle. They hold source stock, deal only in real requirement, stay out of the public price war, and steadily collect the real dividend of the trade.

The second group follows the crowd in the middle. They read the market off group rumours and paper prices, so their information is always half a step late, most of their time goes into matching that leads nowhere, and good orders rarely land.

The third group is the retail tail. They are easily carried along by the rhythm of the online paper market, cannot see the logic of the layers, cannot tell good stock from bad, miss the market repeatedly and stay on the back foot.

One thing I would add: these read less like three kinds of person than three positions. The same goods, the same buyer, and where you stand decides which group you are in. So the real question is not "which kind of person am I" but "where am I standing right now, and what entitles me to stand there".

How to judge a price

The article's test is simple: the real transaction price does not take its cue from the lowest scattered listing.

The only valid reference is the price of an order that has real money received, a proper invoice record and evidence of an actual cluster deployed. The article says this is also the core test used by government and enterprise buyers and by large AI projects.

I accept that without argument. After all these years in export, a quote and a transaction price are always separated by a stack of paper: contract, invoice, corporate bank record, acceptance. A price without that stack is an opinion; a price with it is the market.

So when somebody tells you "B300 has dropped to such and such", you can ask three questions: where is the order, has the invoice been issued, and which machine room did the rack go into? If none of the three can be answered, the number is not a price.

Zheng Ge's Take

Price filters not only the quality of the goods. It filters the level of understanding of the people in the trade.

The angle of this analysis feels true to me. What people on the front line feel and what the internet quotes are completely disconnected: most of the low prices visible in public are circulating paper offers, and the stock that can actually be delivered was digested inside the core circle long ago.

So I would turn its conclusion into three lines that are useful on the B2B side.

One: when you look for a B300 supplier, do not look only at the posted price. Look at whether they can issue a proper invoice, whether you can verify title to the goods on site, and whether the machine can run today.

Two: a price far below the market almost always has a delivery defect. It is not a question of whether the seller wants to deliver. He has nothing to deliver.

Three: a genuinely good supplier does not fight on price. He earns his living from circle resources and delivery capability.

In compute hardware the contest has never been about who quotes lower. It is about circle resources, the layer your stock comes from, information asymmetry and the ability to deliver.

One last thing. I agree with part of this, not all of it. Ratios like "ninety per cent" and "ten per cent" I take as one insider's sense of the market, not as statistics. But the direction it points in — that the public price and the real transaction are now two different things — I think is right. In B2B, the most expensive thing is never the machine. It is the price you believed.

Source: 今日頭條 (Toutiao), "Can't Make Sense of It: Shenzhen's B300 Market Is Badly Distorted, and Nine in Ten Real Deals Are Underwater", 18 October 2026. Statements quoted here — including that more than ninety per cent of large real orders are closed-loop underwater deals, and that public supply reachable through ordinary channels is under ten per cent of real stock — are that analysis's own claims, restated and commented on. None of those figures has been independently verified for this article.