The Wire China’s Substack
The Wire China Podcast
Nvidia's Partner Problem
0:00
-15:39

Nvidia's Partner Problem

Nvidia has nearly 1,000 partners around the world, including over 100 in China. Some of these firms, according to a Wire China investigation, have close ties to China’s military. What does this mean for Nvidia and the future of U.S. chip restrictions?

This week, reporter Noah Berman and editor Andrew Peaple go behind the scenes of Noah’s investigation into Nvidia’s China partners, discussing how one partner connects to a network of sanctioned universities, and what this reveals about the way the world’s most valuable company operates. His article is now available to read on our website, thewirechina.com.

We also have an excerpt from our recent interview with historian and sinologist Geremie Barmé in which he talks about his misadventures studying Maoism in Shanghai right after the end of the Cultural Revolution.

Share

Geremie Barmé. Illustration by Lauren Crow.

Transcript

Andrew: Hello, and welcome to The Wire China podcast. I’m Andrew Peaple, the Wire’s news editor, and I’m coming to you today from a very sunny UK, very sunny once again.

Half the country is now in drought, unbelievably. So crazy times we’re living in. But anyway, this week, I’m going to be talking to my colleague Noah Berman, one of our staff writers about his latest story on Nvidia business ties to China.

But before that, a quick plug for our latest issue, which is out now. We’ve got Noah’s story, of course, which we’ll talk about a bit more in a bit. We’ve also got an interview with one of the world’s leading experts on Chinese culture and history, that’s Geremie Barmé.

We’ve got a great opinion piece on whether China can maintain its commitment to openness when it comes to AI models, really hot topic in that industry right now. And a look at two Chinese companies which are in the news because of big stock market listings. Chipmaker CXMT, they’ve had their IPO, massive jump in their share price after that IPO.

So pretty interesting to catch up on the details about that company. And then this week, we’ve also published a pretty in-depth look at the numbers behind Shein, the online fashion retailer. We’ve got a timeline of its history as it now warms up towards an IPO, which may or may not be as successful as CXMT’s, we’ll have to see.

Well, anyway, you can find all of that at thewirechina.com. As ever, please do take a look. We’ll have another set of stories coming out this Sunday, too. Before all of that, let’s talk to Noah about Nvidia partners in China, his great story on that.

Hello to you, Noah. How are things over in New York?

Noah: They are nice and humid, Andrew. Thanks for having me.

Andrew: Absolutely. Well, another great story by you on Nvidia, Noah. In this one, you’re looking at what Nvidia calls its partners.

So can you just start off by explaining who these partners are and how many of them are in China?

Noah: Nvidia is a designer of computer chips, but they aren’t necessarily selling directly to the people who are using those chips to end customers, to set up a data center. For example, you may be interested in buying servers which contain loads of chips rather than buying those chips themselves. And so Nvidia relies on a network of companies that it calls partners to bring its technology to market.

It’s a dynamic list, but it currently has almost 1,000 partners, and of those, 123 are in China.

Andrew: And so what do Nvidia Chinese partners typically do? And why could that actually be a concern for the United States government more broadly?

Noah: So in general, Nvidia divides its partners into more than a dozen categories, and each is slightly different. But the two most common types of partner Nvidia has in China are what the company calls a solution provider and what the company calls an original equipment manufacturer or OEM.

Solution provider is primarily a reseller that can include Nvidia chips and other Nvidia products. And an OEM is similar in that it also uses Nvidia technology, but the difference is that OEMs typically resell under their own brand name rather than using Nvidia. So to give an example of a well-known Chinese OEM that is a partner of Nvidia is ZTE, which sells its own AI servers, well-known telecom company, potentially using Nvidia chips.

ZTE, of course, has been subject to US trade restrictions for many years since the first Trump administration, but it’s not currently illegal to export chips to ZTE. And Reuters has reported that a unit of ZTE is among the companies that have received a license from the US government to buy Nvidia H200 chip, currently the most advanced chip Nvidia can legally sell to China. So from the US government’s perspective, they’ve been trying to keep Nvidia most advanced chips out of China for almost four years now.

The reason those controls were imposed, as we have reported, is because officials in the Biden administration viewed AI as key to the future of military operations. And they therefore wanted to sustain, as former national security advisor Jake Sullivan often said publicly, as large a lead as possible. And you need the best chips to train the best AI.

So as far as Nvidia’s partners go, what could then be concerning to the US government is whether Nvidia’s own partners are seeking to supply Chinese universities or Chinese companies that the US government has determined are working to advance China’s military ambitions. And what I found is that at least 23 of Nvidia’s partners have won contracts to do exactly that, including in some cases explicitly with Nvidia products.

Andrew: I see. Can you give us an example then of the sort of relationship that you trace between Nvidia and then one of its partners in China, and then these ultimate customers inside of China that might be of concern?

Noah: One Nvidia partner is called AMAX. It is a manufacturer of servers that is publicly listed in Taiwan. And it says on its website that it is headquartered in California.

But it has a subsidiary in China, in Suzhou. And since 2022, that subsidiary has won 10 contracts to supply four of the universities known as the Seven Sons of National Defense. It’s worth noting that is not a term that comes from the US or even from English.

These schools are not military academies, but they have a strong historical relationship with the Chinese military. And that is a translation of how they are known in China.

So one of the contracts that I reviewed was a tender offered by Beihang University, one of the Seven Sons schools.

Beihang has been on a US blacklist called the entity list for 25 years since the Bush administration. And Beihang’s tender was won by AMAX earlier this month, and it explicitly called for a set of four chips called RTX 6000D, which is an Nvidia model. I should note that the procurement records I reviewed are the results of a government bidding process for contracting, but they don’t show that the product was delivered.

They only show the company that won the contract.

Andrew: Still pretty clear evidence of this relationship between Nvidia and these customers that are ultimately on US blacklist somewhere. I mean, what sort of questions does this raise for Nvidia and the kind of checks that it’s doing on the ultimate buyers of its chips?

Noah: As you say, the big question is how does Nvidia vet its partners? Does it ask them, for example, if they are seeking to supply the PLA? Are they seeking to supply universities or companies under various forms of US sanctions? And do they care if that is indeed the case? Another big question is enforcement.

One Nvidia partner in China is called Shandong Yiran Information Technology. And for the piece, it told me that it complies with a list of banned customers given to it by Nvidia, and that that list is based on a commerce department blacklist, which is probably the entity list. So if that is true, then is Nvidia aware that at least 23 of its partners are seeking to supply those same customers that it told another partner are banned? And what does that reveal then about Nvidia’s enforcement efforts if it does indeed have that ban?

Andrew: And when you went to the company, when you went to Nvidia, what did they have to say about this?

Noah: So Nvidia gave me only a terse statement.

But what they said is that companies are free to join their network of partners, but that doesn’t necessarily mean they are receiving products from Nvidia. They declined to comment on any of their specific partners or to say anything even generally about the due diligence that they conduct on their partners. And with regard to whether the Chinese military is using Nvidia chips, they are pretty adamant, both for this piece and in previous stories by The Wire China and other outlets, that China has millions of domestic chips, more than enough for every military use.

And that may well be true. But previous studies, including by researchers at Georgetown University and by Wirescreen, our sister company, show that the Chinese military is indeed seeking to buy Nvidia chips anyway.

Andrew: Yeah, I mean, one of the head scratching aspects of this whole situation, I always think, is that on the one hand, you have the U.S. government, which has various levels of export controls on Nvidia chips and has still some of those controls in place, even though they’ve weakened somewhat under the Trump administration.

But actually, on the Chinese side, they’re kind of saying, well, we don’t want your chips anyway, and they’re trying to keep out Nvidia chips. Why is that? And what does it say, then, that there are still plenty of these Chinese interests, it seems, still trying to get hold of advanced Nvidia chips?

Noah: There are a couple of factors, for sure. But one is that the Chinese government is pushing self-sufficiency.

It does not want to be reliant on U.S. technology. So it is pushing its companies to use chips made by national champions, companies like Huawei and other Chinese competitors to Nvidia. But industry experts say Nvidia chips are miles better than any Chinese competitors.

So for China to get its companies to adopt homegrown alternatives, it may well have to create artificial scarcity. And to do that, you end up with restrictions on Nvidia. But none of that changes the fact that Nvidia’s best chips, at least, are better than Huawei’s best chips.

And Nvidia also has a dominant software ecosystem, which allows those chips to communicate with one another. So once you start buying Nvidia, it’s easier to just keep buying more rather than switching to a new brand that might not mesh as well with Nvidia software. And so what these records and previous studies suggest is that even universities close to the PLA, and maybe even the PLA itself, recognize that.

And I just add one more point on this, which is there is an undercurrent of fears about cybersecurity, just as policymakers in Washington are worried about backdoors in products like Chinese electric vehicles, which are effectively banned from the United States, or internet routers. So too are Chinese regulators concerned about American technology. And just last summer, China’s cyberspace regulator summoned Nvidia to talk about exactly that.

But again, these records suggest that those types of concerns are manageable from Beijing’s perspective.

Andrew: I see, that’s really clearly explained. Thanks, Noah.

What are some of the broader implications here? I mean, it feels like we always come back to this debate. The fact that these chips from Nvidia are making their way to China, and into the hands of Chinese customers that the US government, at least on paper, doesn’t want to have them obtaining these chips. It all plays into this debate of whether export controls, ultimately, are they a waste of time? Is China and Chinese customers, are they going to get around these export controls by hook or by crook? Or should the US carry on with these export controls, because at least they’re making it harder for China to get hold of the most advanced chips? I mean, it plays into that debate.

What do you think the sort of broader implications are for that discussion that’s still ongoing about the efficacy of export controls?

Noah: I think what this shines a light on is just the difficulty of formulating export controls that intend to regulate sales of advanced technology in China, in particular, because it is a country that has long sought to integrate its military and civilian institutions. And those efforts really have gone into overdrive under Xi Jinping. So can a company, especially one that distributes its products through a network of local partners, really prevent its technology from going to a given end customer, even if that is the PLA? And if that technology is not Nvidia’s most advanced, if those companies can buy alternatives in the local market, is it worth the effort to control them? Sources I spoke to for this story said there is more that policymakers and Nvidia itself can do.

And there are a couple of bills floating around Congress that would make it even harder for Nvidia to sell products in China. But we’ll see what happens to those over the next couple of months. They may have an uphill battle.

Andrew: Thanks so much for that, Noah. It’s a really fascinating story. Once again, you’re really on top of this Nvidia beat at the moment.

Lots of really interesting implications for policymakers, for Nvidia itself, and for China and Chinese customers. So really, really, really interesting stuff. Thank you.

Before we leave you, we’ve got an extract from that interview that I mentioned at the top, the Q&A with Geremie Barmé. I hope you’ll enjoy that extract, which will play after this. In the meantime, thanks so much for listening once again to the Wired China podcast.

Do check out our stories, including Noah’s, including this interview with Geremie Barmé, all at thewirechina.com. Do take a look. Thank you for now and goodbye.

Geremie: All our classmates were former Red Guards who had been or participated in the height of the Cultural Revolution, who were selected by their local work units in the countryside or in the cities, propaganda units or wherever, to study at university because they were politically reliable and advanced.

So all of our classmates were very ideologically reliable. I arrived there in 1974, October. It was the tail end in China, the tail end of the anti-Confucius, anti-Lin Biao movement.

We were foreign students who had read the documents. And then the teachers, of course, in keeping with the way the party works, they wanted to see how the foreign friends responded to this movement. So we all got together and had meetings with them and said, well, you know, you have cars and go around in private vehicles. You should all go in a bus.

You all have special, we believe you all have special apartments and special food and privileges to get rid of all of that stuff. And of course, they took copious notes of all of our comments.

And a few days later, obviously, having gone back up to the party committee running the school or even the Shanghai party committee, and they came back and then spent a whole half a day denouncing us for incorrect and erroneous ideas.

So that was my introduction to get involved in the movement, study all the documents, and then share your views in keeping with the party’s policies to criticize and self-criticize and have open exchanges. And then immediately to denunciations and vilifications. So that was my very first encounter.

Discussion about this episode

User's avatar

Ready for more?