On November 4, Bloomberg published a report about deliveries. Several large Chinese artificial intelligence and cloud computing companies had together ordered about six million Nvidia RTX Pro 5500 chips, a quantity large enough to buy out the next few years of supply at once.
The order is large; the delivery is small. According to people familiar with the matter, in the next four months these companies will receive only tens of thousands of chips, roughly one percent of the total. The rest remains in the schedule, with no certain date.
What does one percent mean? If six million chips were lined up, the first batch would barely fill the front of the queue; everyone behind would keep waiting, with no idea until when. Tens of thousands sounds like a lot until it is placed inside six million, where it is loose change.
At more than 11,000 dollars a chip, the order is worth about 70 billion dollars. That is the figure Bloomberg gave, sourced from unnamed people, and Nvidia itself has not confirmed it.
How the Order Came About
The buyers are not one company. Alibaba, Tencent, ByteDance and other AI and cloud computing firms are all on the list. They are not buying the most advanced chip, but the tier that can still be bought at all.
In September, the US technology outlet The Information reported that China had signaled it might allow Alibaba and ByteDance to purchase this chip. The order probably grew in that gap. It was nobody's plan; it was grabbed.
Whether a chip can be sold to China depends not only on how fast it computes, but on whether it has crossed a certain line. The RTX Pro 5500 has not crossed it, so it became the one shelf still within reach, and the one everybody reaches for at once.
For that reason, its scarcity is not technical but a matter of production scheduling. The same machines are being fought over elsewhere, by buyers who bid higher, so the order of the queue changes: first come, first served gives way to who pays most.
What One Percent Means
The timetable is specific: the first few thousand chips are expected before the end of this year, and another tens of thousands around February next year. So far only those volumes are confirmed; the rest is still parked inside the word "expected."
The buyers' own internal estimates say that if Nvidia expands production smoothly, by the end of next year large Chinese cloud providers could have received about 700,000 chips in total. That is an estimate, not a commitment, and it will not be written into a contract.
Seven hundred thousand against six million is about one in nine. Even that is the optimistic calculation, and it depends on how capacity is adjusted. Bloomberg noted the figure could change with scheduling; buyers treat it as a reference, not a crutch.
So "one percent" is not the end point but the starting point. It describes the next four months, not the fate of the whole order. The two statements are far apart and should be read separately. Delivery plans roll by quarter, and nobody can say now what happens after four months.
Who Got the Capacity First
The reason for the shortage is not complicated. Nvidia has allocated more manufacturing capacity to next-generation products such as Rubin. New products carry higher margins and greater urgency; mature models wait at the back, and how long they wait depends on how much the front gives up.
The foundry is TSMC. A production line's annual output is relatively fixed: give it to one product and another gets less. Nvidia has warned customers that product schedules conflict and capacity will remain limited for some time.
This is not an arrangement aimed at China. American AI companies are fighting for capacity at the same time. Bloomberg named OpenAI and Anthropic as parties receiving priority in this round. Who gets capacity first depends on order size and margin, not on who signed earliest.
In other words, Chinese customers are not in a "China queue" but a "global queue." They stand at the back, the line ahead is long, and capacity is only so large, so their share comes out low, low enough to look like a different business altogether.
What Can Be Bought Is Not the Best
One point is easily missed: this chip is not Nvidia's most powerful tier. Its performance does not reach the threshold set by US export controls, so it can be sold, but only just.
In the Chinese market it already counts as a relatively strong tier. The sentence sounds contradictory but describes two sides of one thing: wherever the ceiling of what can be bought is drawn, the shelf ends there, with no spare room.
Buyers know this very well. They are not buying the best; they are buying what can be had now. That difference means they will not put every plan on this one chip, and will not leave themselves a single road.
While Waiting for the Shipment
Demand for what cannot be bought does not disappear. Bloomberg reported that large Chinese cloud providers will likely turn to domestic options: Huawei, Cambricon and Alibaba's T-Head all supply inference chips.
That road has bottlenecks of its own. Advanced manufacturing equipment and services are restricted, high-bandwidth memory has long been in short supply, and the two press on capacity together, pushing prices up. When goods are scarce, waiting becomes normal.
Nvidia's spokesperson declined to comment on delivery numbers. He said only that American firms remain subject to two constraints: outdated US export controls, which still cover gaming products released nearly five years ago, and China's restrictions on American imports.
That is a position, and also part of the facts. When a company is squeezed by rules from two directions at once, delivery stops being merely a factory problem, and stops being something the company alone can decide.
Over the past two years, Nvidia has adjusted product specifications for the Chinese market several times, and each adjustment forced buyers to queue again. The queue grows longer, and orders grow larger, because nobody wants to be at the back.
For the companies that ordered, waiting has a cost. Data centers must be planned ahead, power must be applied for ahead, and training schedules already run into next year. If chips arrive a month late, everything downstream shifts by a notch.
Yet they ordered anyway. Because the choices that can be bought are few, and even one percent beats nothing at all. The order here is not only a purchase but a placeholder: hold the spot first and talk later.
This is what the business looks like now: orders signed to the brim, warehouses still empty. The numbers are all large; what reaches the hand is small. The distance in between can only be filled with time, and nobody dares promise how much.
Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.