One percent. I read that number three times before it sank in.
Chinese cloud and AI companies lined up for one of the biggest graphics-card orders anyone can remember, and the supplier is reportedly telling them they'll get a sliver of it. Not half. Not a quarter. A rounding error. And the more I looked into it, the more it seemed like the usual framing, "Nvidia is losing a huge China sale," gets the story backwards.
Key takeaways
- Orders: about 6 million RTX Pro 5500 GPUs, roughly $70 billion, per Bloomberg's sources.
- Deliveries: "some tens of thousands," or about 1%, with some outlets putting it near 60,000 units.
- The chip matters because it's legal to sell to China, unlike Nvidia's B300 and, without a licence, its H200.
- Domestic alternatives like Huawei's Ascend aren't ready to absorb the demand.
- Nvidia hasn't publicly confirmed the 1% figure. It all rests on unnamed sources.
What actually happened
The story arrived in two steps. On 20 October, Reuters reported that ByteDance, Alibaba and Tencent had together ordered more than 3 million Nvidia RTX Pro 5500 cards, worth over $35 billion at roughly $11,400 apiece. One source in that report said Nvidia had warned partners they might get only 5% to 10% of what they wanted.
Two weeks later Bloomberg widened the picture. Citing people familiar with the matter, it said Chinese AI and cloud firms had ordered about 6 million of the cards over four months, a total near $70 billion, and that only some tens of thousands would actually be delivered. So the expected fill rate went from "5% to 10%" to "about 1%" in the space of a fortnight. That's the part I keep coming back to.
| Item | Figure | Source |
|---|---|---|
| Early order estimate (three firms) | 3 million+ chips, $35 billion+ | Reuters, 20 Oct 2026 |
| Wider order estimate | About 6 million chips, about $70 billion | Bloomberg, 4 Nov 2026 |
| Price per chip | Roughly $11,000 to $11,400 | Reuters / Bloomberg reports |
| Expected deliveries | Tens of thousands, about 1% (some outlets say ~60,000) | Bloomberg and follow-ups |
| Rough value of 1% | Under $1 billion | Simple arithmetic on the figures above |
A quick note on the math: 6 million chips at about $11,000 comes to roughly $66 billion, so the "$70 billion" headline is a rounded number. Order books also tend to be padded when buyers know supply is tight, a point I'll return to below.
Why this one chip? The RTX Pro 5500 explained
The RTX Pro 5500 is a Blackwell-based workstation GPU with 84GB of GDDR7 memory. Nvidia launched it in September 2026. Tom's Hardware notes it shares the CUDA core count of the flagship RTX 5090 but carries far more memory, which is exactly what you want for running large AI models.
Here's the catch that makes it so popular: it's legal. US export controls bar the B300 data-centre chip from China outright, and the older H200 is reviewed case by case with licensing conditions. The RTX Pro 5500 sits outside those limits, so it's the most powerful Nvidia product Chinese companies can buy without special permission. That's why demand piled onto one model.
| Nvidia chip | Type | Status for Chinese buyers |
|---|---|---|
| B300 (Blackwell) | Data-centre accelerator | Not allowed under US export controls |
| H200 (Hopper) | Data-centre accelerator | Case-by-case licensing with conditions |
| RTX Pro 5500 (Blackwell) | Workstation GPU, 84GB GDDR7 | Export-compliant, no licence needed |
There was also a political wrinkle. In late September, The Information reported that Beijing's industry ministry had signalled it intended to approve purchases by Alibaba and ByteDance. Reuters couldn't verify that at the time. So for a few weeks the road looked open from both ends, and buyers rushed to get in line.
So why can't Nvidia ship more?
Honestly, the reporting is a bit murky here, and I'd rather say so than pretend otherwise. Different outlets point to different pieces of the puzzle:
- Capacity. Bloomberg's sources say demand far outruns what Nvidia planned to build, with capacity for 2027 already stretched.
- Components. eTeknix, summarising the report, points to manufacturing limits and shortages of key parts hitting the whole chip industry.
- Scheduling. Per Seoul Economic Daily, Nvidia told customers supply would stay limited for now because of conflicts in its production schedule.
- Rules on both sides. Nvidia's own line is that US companies are squeezed by outdated American export controls, which still cover gaming-era products nearly five years old, and by China's limits on importing US goods.
One outlet, BigGo Finance, also reported that Nvidia verbally guided customers to keep Chinese buyers to 49% of a cloud provider's business, aimed at offshore workarounds. I could only find that in one place, so treat it as unconfirmed.
Why this may hurt China more than it hurts Nvidia
Now to the angle in the headline. This is analysis, not a settled fact, but here's my reasoning.
1. Nvidia already lost most of China
Jensen Huang has said Nvidia's share of China's AI data-centre market is now close to zero. Delivering about 1% of this order means roughly $0.7 billion of chips. It's real money, but it's a missed upside, not a collapse of an existing business.
2. China's homegrown options can't absorb the demand yet
At Huawei Connect in September, rotating chairman Eric Xu said Huawei's AI computing output couldn't yet meet domestic demand. Reports from German outlet Borncity add that Huawei raised prices on its top accelerators by around 60% in September and that SMIC's manufacturing yields remain a problem. Beijing has been steering firms toward Ascend chips for a while. Wanting something and being able to supply it are two different things.
3. The workarounds are costly and fragile
With chips scarce, Chinese companies have reportedly been training models on rented capacity in offshore data centres, in places like Singapore and Malaysia. That works until rules change, or until costs climb. It's the opposite of the stable, owned infrastructure that big AI labs want.
The other side of the argument
To be fair, there's a counter-view. Beijing has long wanted its companies to lean on domestic silicon. A forced shortage of Nvidia chips could speed up adoption of Huawei, Cambricon and others, and Huawei says it expects wider use of Ascend processors for AI training from next year. If that happens, "hurt" might turn out to be short-term pain for long-term independence. I don't think anyone can say yet which way it breaks.
A reporter's take
I've spent seven years in news and media watching how big numbers get used, and the first instinct is always to repeat the headline figure. Here I'd slow down. A $70 billion order that was never going to be filled tells you less about what China will buy and more about how scared buyers are of missing out. When supply is rationed, everyone orders triple what they need. The real signal isn't the $70 billion. It's that Chinese firms are willing to chase a workstation card for AI work because nothing better is available to them.
That's a story about scarcity and policy, not a clean win or loss for any one company. And since every number here comes from anonymous sources, I'd hold the specifics loosely until Nvidia or Beijing says more on the record.
What to watch next
- Beijing's decision. Whether the approvals The Information reported actually come through, and for how many units.
- US export rules. Any change to how gaming-era and workstation products are treated.
- Huawei's ramp. Whether the next Ascend chips arrive in volume next year, as the company has indicated.
- Nvidia's next earnings call. Management comments on China, and whether the 1% number gets confirmed, corrected or ignored.
Frequently asked questions
How much did Chinese companies order from Nvidia?
Bloomberg reported on 4 November 2026 that Chinese AI and cloud firms, including Alibaba, Tencent and ByteDance, ordered about 6 million RTX Pro 5500 chips worth roughly $70 billion. An earlier Reuters report had put the three giants' orders at over 3 million units and $35 billion.
How many chips will Nvidia actually deliver to China?
Bloomberg's sources say only some tens of thousands, about 1%, over the next few months. Some outlets estimate around 60,000 units. Nvidia hasn't publicly confirmed the number.
What is the Nvidia RTX Pro 5500?
It's a Blackwell-based workstation GPU with 84GB of GDDR7 memory, launched in September 2026. It complies with current US export rules, which is why Chinese buyers can order it and why it's popular for AI inference.
Why can't Nvidia ship more chips to China?
Reports cite tight production capacity, component shortages and scheduling conflicts. Nvidia also says US firms face outdated US export controls and China's own limits on US imports.
Why might this hurt China more than Nvidia?
Nvidia's share of China's AI data-centre market is already near zero, so it loses relatively little. Chinese firms lose access to compute they need, and domestic options like Huawei's Ascend are still supply-constrained.
Can Chinese firms use Huawei chips instead?
Partly. Huawei says it can't yet meet domestic demand, and yield problems at SMIC are reported to be a bottleneck. Beijing is pushing firms that way, but closing the gap will take time.
Where this leaves us
A $70 billion order, a 1% fill, and two governments pulling in opposite directions. For Nvidia, it's a headline about money it won't make. For China's AI companies, it's a practical problem: not enough good chips, and no easy substitute. Which side feels it more will depend on how fast Beijing's preferred suppliers can scale, and on whether Washington or Beijing changes the rules again.
I'll keep following this one. If new numbers surface from Nvidia, Bloomberg or Beijing, this article will be updated.
