Why Kevin Carter thinks China is winning the AI race

0:00I'm Ashton Addison from the Crypto Coin Show and today we're joined with Kevin Carter, founder and CIO of EMQQ Global and EMXETF. Kevin has been building emerging markets index funds since 2014 and just in the last month his team has launched new funds, the China AI ETF, and as of this week, the China AI ETF is under AICH, a broader bet on the whole AI stack.

0:27There's a lot to discuss between Western AI, Chinese AI, and where we're all at. It's moving so fast and you're the man that has some of the insights that I would love to dive into today. So thank you so much for taking the time, Kevin.

Sure, glad to be here.

Yeah, so most people in AI or investing in the West, they're looking at AI and trying to find where's the perfect place to be investing?

0:50Are we buying the picks and shovels? Are we investing in the models? Are we looking at Nvidia for chips? Data centers, power companies? It's all exploding and some are higher upside, higher risk than others. And I think as we are in this exponentially fast growth phase, there's a bit of noise and I think that's going to fade out over the next couple years and we'll see some clear winners in where the most asymmetrical upside was.

1:16What's your take on where the money is right now in AI?

Well, right now the money is going into the infrastructure. And if you know the best way to think about AI is the five-layer stack. The bottom layer is power, the second layer is chips, the third layer is the infrastructure data centers. And those first three layers, that's really the platform on which AI runs.

1:42We had power before AI, we had chips before AI, we had data centers before AI, but it's that fourth layer, the magic layer as Jensen calls it, the model layer that requires all of that capital expenditure. And so it's that picks and shovels layer. And right now, and not just now, but over the last couple years, there's been anxiety about the scale of the spending.

2:06The hyperscalers have basically deployed all of their free cash flow towards building data centers. And there's rightfully a lot of concern about whether that's going to pay off. But on the flip side of that, you have companies in Asia and in Taiwan, in Korea in particular, the mag three of AI: TSMC, SK Hynix, and Samsung. And they're making a lot of money.

2:30I mean, if you look at the earnings of those three companies between this year and next year, it's expected to be about $965 billion of profit.

Wow, the picks and shovels are, in a profitability lens, the ones making all the money off of that spend.

So that's a clear place the money is being made. And the models, it's not clear any of the models anywhere are making money yet.

2:54But that's one of the trillion-dollar questions.

Definitely, and that's an interesting point because it seems like from an end-user perspective right now, people are paying $100, $200 and they're getting PhD-level inference and coding for a month. And I feel like this should be worth tens of thousands. How is the company not going bankrupt? And how do you actually see that playing out in the cost of tokens and the AI companies doing the modeling?

3:15How can they stay profitable in the long run and how will that work out?

Sure, well, as you know, one of the funds that we launched, in fact the first fund that we launched as part of EMXETF, was the China AI Tigers, TGRZ, which tracks the Chinese models. The large language models, the models, the labs, whatever you want to call them, the US and China are the only real players.

3:40And in the United States we of course have OpenAI with ChatGPT, we have Anthropic with Claude, Google with Gemini, we have other players. Musk has a strong offering as well. But most, indeed nearly all of our models are closed models. Now on the other side, you have the China models and you have DeepSeek, the one that most people know because it's the one that rattled the cage on the US AI story when it launched.

4:05You know, 20 months ago in January of 2025, there was already a lot of anxiety about all the capex and DeepSeek said, "Well, we made the same thing for $6.5 million." And that really freaked people out. And I think Nvidia dropped 35%. And it started to question whether we needed to deploy that much money. But beyond DeepSeek, which has continued to push the envelope, you have the so-called AI Tigers of China.

4:29Now these are six China AI companies, all of which have come out of the Tsinghua University computer science lab. This is a leading science lab in the world, likely, and those include Z.AI, which has gone public and is part of the Tigers. That's MiniMax, which went public in January as well. So those are the only two publicly traded large language models today.

4:52And then you have Kimi and Moonshot, or Moonshot and the Kimi K3 model as well, which a lot of people know about. That was sort of the second DeepSeek moment. And the China models have a lot of advantages. They are open-weight and can be downloaded and run on premises, so you don't have to log in and pay those expensive inference fees. You can customize them for your own use.

5:18And they're increasingly just as good and getting better, catching up with our models and they cost a tenth or even less the price. And because of that, you're seeing them get used, and not just used in China, they're getting used in the United States. Alibaba is built on Alibaba's model. Cursor, which was acquired by SpaceX, was built using China models. So the China models have actually passed the US models in terms of token use in the last several months.

5:45And I think that's going to continue.

Yeah, well, I think at least end users in the West would be happy to be paying less, and the fact that you can download it and not have to pay the inference fees. I feel like the blockchain industry and the crypto sector, these guys, they want more open-source models that you can own or download, but maybe it's always been a matter of convenience.

6:09And I feel like they get caught up and be like, let's just use ChatGPT and give it all our information and pay more because that's what other people around me are doing. I don't know if that's the right way people should be going about it.

Well, I mean certainly there is a convenience factor and there's going to continue to be use cases for both models.

6:32But again, as we saw earlier this year when token maxing became a thing and then within two months when the bills started to show up, then they started to do token mining. And when the bills came, you're going to see a lot more open models and you're already seeing that the open models are getting used more. You can use the China open models at AWS and at Microsoft.

6:56So the cloud hosting companies are making them available. DeepSeek even, I think, is available at Microsoft, and the Moonshot people are working on revenue-share deals with all three of our major cloud companies. So it's really early in this model race. And of course, there's been a lot of noise. I mean, there's been an incredible amount of noise in the last five days with the push by Dario and Sam to get regulated.

7:17And then Trump, of course, responding it's a hoax. And so it's a pretty heated moment right now. And with Xi Jinping coming to visit next week, I think we'll see more fireworks in the next couple weeks.

Definitely, I think we're going to be seeing a lot of headlines about China AI and that's why this conversation is very timely.

7:39And there was also an announcement from the leading government bodies in the US putting out an advisory for the Chinese models saying there was distillation, whether they were basing it off the American models or just putting out a warning. But the government loves to put out warnings. What are your thoughts on whether people in the West should be cautious about these China AI models and from what aspect of that?

8:03

I don't think there's any reason to or need to be cautious on these China models any more than any other providers' models. In terms of the question of distillation, I mean, the entire AI industry is based on distillation. I mean, they distilled everything off the internet and by the way, they've been sued by lots of people whose IP was stolen or distilled.

8:26

I don't think there's any reason to or need to be cautious on these China models any more than any other providers' models. In terms of the question of distillation, the entire AI industry is based on distillation. They distilled everything off the internet and they've been sued by lots of people whose IP was stolen or distilled.

8:50Distillation is also a common practice amongst our labs. I can't speak to the veracity of some of the claims of distillation, but I don't think it's a long-term issue.

Yeah. No, I agree. You can put it out, you can see the warnings and then you can take a look at the Western models and they're based off of scraping information off the internet and also information that we're freely giving them.

9:13When we check the terms and conditions box, they're learning from whatever we say to make themselves smarter and building their own models. The frontier is in recursive building, so it keeps building from itself. There's been discussion around that recently—that's part of the issue, that it's moving too fast because of the recursive training. What are your thoughts on that?

9:39

Well, I think it's good that there's so much happening in AI so quickly. It's hard for anyone to really keep their arms around it. But I think the move that basically started last week—you had the guy that quit and said, "Everyone's going to die in five years and no one's taking it seriously." People had done that before, but for whatever reason last week it did jar the main players.

10:04issue.

Yeah. No, I agree. And you can see the warnings and then take a look at the Western models and yeah, they're based off of scraping information off the internet and also the information that we're freely giving them. When we check the terms and conditions box, they're learning from whatever we say to make

10:26themselves smarter and building their own models. The frontier is in recursive building, so it keeps building from itself. And there's been discussion around that recently that it's part of the issue—that it's moving too fast because of the recursive training. What are your thoughts on that?

Well, I think it's good

10:51that there's just so much happening in AI so quickly. It's hard for anyone to really keep their arms around. But I think the move that basically started last week—I mean, you had the guy that quit and said, "We're going to die in five years and no one's taking it seriously." And people had done that before, but for whatever

11:12reason last week, it did jar the main players in this—Dario, Sam, and Elon—to agree that we really should have some procedure to review at least the models before they're released. And this makes a lot of sense. You also had on top of that Dario calling for government regulation to slow down and they got pushed back pretty hard by the president and by Sachs and

11:39they basically said, "Look, do it yourselves. Figure it out." And Elon had been calling for this previously. And the idea is, look, all of these guys are in their bubble, right? They're doing their thing. They have their models. You just get locked into something and sometimes it's good to have a fresh set of expert eyes come in and review and look. It's like proofreading, right?

12:01If you're writing a book or something, it's good to have other people come and take a look at it and give you a reality check. So I think those efforts seem to be moving very quickly and I think that's a good place to at least try to minimize risk associated with things like recursive alerting. And importantly, you also saw Elon Musk saying we need China to be involved with this as well.

12:20

Right, the Chinese models also need to participate in this sort of industry group.

And that's why this summit that's coming up a week from tomorrow with the president and Xi Jinping could be one of the most important summits in modern history. I mean, there's a lot going on.

A lot of people are framing this as a race, as a battle between the US and China. And I think that's the wrong way to

12:46think about it. I mean, this is a wave, right? Nobody won electricity, right? We all won. So I think if the US and China—because we ultimately were the two main superpowers to start with and we are clearly the two AI superpowers—and I think it's very important that we have some alignment and safety considerations

13:10and some sort of apparatus put in place so as things continue we can better keep our arms around it.

Yeah, it's a great point, Kevin. And to that about limiting the recursive models and people considering it a battle between the US and China, part of the thought was that if we do limit it, then China's models are going to get ahead.

13:38And to your point there about everyone wins when there's electricity—do you think that's an issue with China models getting ahead or does it matter whose models are better?

Well, here's what I would say. First of all, I think the reality is and one of my heroes is this man Nandan Nilekani, who's one of the founders of Infosys, India's

14:03largest tech company. And as he has said for a long time now, we don't need the models to get any better for ten years. For the vast majority of things that AI can do, it's already good enough. And it's more about companies figuring out how to use it, right? They don't need the latest version of Claude or the latest version of OpenAI's models

14:28to do almost all of the practical things that need to be done. So there will continue to be this race or drive to improve models and improve efficiency. That's one of the other things the China models have become—a lot more efficient—largely because they were forced to because they didn't have access to Nvidia's leading chips. And so

14:54they've been very crafty in terms of finding ways to tweak the models and the hardware so that they can compete. And it very well may be that China models will catch up with ours. I mean, they're pretty close now.

I know a couple of months ago someone tweeted to Elon about when the China models would catch up and his

15:16response was first quarter of 2027 and the founder of Z.AI responded probably sooner. [laughter] Look, China's got more AI patents by a lot. They have more AI scientists. A lot of the leading scientists in the US labs are from China.

China leads in almost every leading technology. So even Jensen Wong has stated that he thought

15:45China was going to win AI. Now he got booed basically by the media after that and Nvidia had to put out a press release. But China's very capable and I don't think we should think of it as a race.

Yeah. But I think China's very likely to end up leading in AI as they lead in almost everything technology at this point

16:14other than EUV—extreme ultraviolet lithography—which is what makes TSMC and Nvidia's chips. That's the real machine that powers the advanced chips, which comes from ASML in the Netherlands.

Yeah, I'd love to jump back into the ETF and sort of the China AI models that are

16:40in there. You know, we've mentioned and people understand from the end user it's much cheaper—much, much cheaper—and the production of the models was at a fraction of the cost as we saw with DeepSeek: six million versus hundreds of millions. I don't know what the exact amount it cost OpenAI at the time, but when you compare those numbers, how

17:08much people are paying versus how much it costs them to make, what does the profitability look like for investing in these companies versus investing in the US companies, which we've seen people in San Francisco dying to get equity in some of these private companies, selling their San Francisco Bay house to get Anthropic stock. But

17:31in terms of the profitability and long-term profitability, how does that look for the models in the ETF?

Sure. Well, the Tigers ETF, TGRZ, which is the China models—right now there's three holdings in this. It's a very narrow group of companies. It's a very volatile group and they're not profitable. They have just like our labs, a lot of revenue growth. They think they can find a way to profitability but

17:52right now it remains a big question mark. So these are amongst the riskiest investments you can make right now, especially these China models. And the volatility reflects that. Now we also have AICH, which is the China full-stack ETF. So that's not just the models. That's where you get power, so contemporary Amperex, which is a leading battery company in the world. On the chip side, everyone knows Nvidia and Intel but they probably don't know Huawei. They probably don't know Montage Technologies. They probably don't

18:15know China's leading chip companies. China has its own group of companies that are competing with Nvidia and AMD. Two of those four companies were founded by Nvidia's China head and AMD's China head. So they come with good pedigree. You have optics makers, which China leads the world in. You've got Eoptis, for example, and Inolite, the other leader. And then the models which we talked about. And then finally you get to the application layer and that's really the most important layer—how do we take all this other stuff and create value with it. And that's where everyone's early, right? You have only a few AI-native apps. Cursor, for example, has been acquired, but there's no great example yet of an application company. And one could argue that it's the applications, the super apps of China like Tencent and its WeChat app that will ultimately win at the application layer, as may Apple here in the US at some point.

18:37But we're still pretty early in that application layer.

And of the AI models, DeepSeek and Moonshot—these are also still earlier-stage private companies from what I understand. Does the future of the ETF profitability or where you see it depend at all on how quickly these companies grow and if they go public?

19:02know you know the China's leading chip companies that China has its own group of companies that are competing with Nvidia now and AMD two of those four companies were founded by Nvidia's China head and AMD's China head. So they're, you know, they come with good pedigree. [snorts] you have optics makers which China leads the world in optics. You've got Eoptolink

19:26for example, Inolite the other leader. And then the models which we talked about and then finally you get to the application layer and that's really the most important layer is how do we take all this other stuff and create value with it and that's where everyone's early, right? you have only a few sort of AI native apps, cursor for example, that

19:51has been acquired, but there's no great example yet of an application company. And one could argue that it's the applications, the super apps of China like Tencent and its WeChat app that will ultimately win at the application layer, as may Apple here in the US at some point. Mhm. Yeah. We'll see how that plays out with Apple and Google. I feel like they've they've been

20:18a little bit behind compared to some of the two main ones in the US. I don't know how the China-US race will fit into that, but I'm eager to see how they really own the devices that everyone is using. How will they take back control on the AI front?

Yeah. Well, ultimately, you need distribution and if you've got a billion people holding a device that you made and runs on your software, that's a pretty good head start.

20:46That's another reason why Tencent I think remains very interesting. It's got one and a half billion people. Tencent's main business is WeChat, which is like the Facebook of China and other parts of the world. They have a great presence and they're already deploying it. We talked about this before, but the QClaw, which is their OpenClaw agent, is available to all one and a half billion people on the app now.

21:05So we're still pretty early in that application layer.

And of the AI models, you know, DeepSeek and Moonshot, these are also still earlier stage private companies from what I understand. Does the future of the ETF profitability or where you see it depend at all on how quickly these companies grow and if they go public?

Well, right now you could say the three leading labs in China are DeepSeek, Z.ai and Moonshot.

21:29Z.ai is public. It's the largest holding in TGRZ. Moonshot is an interesting story. Moonshot was actually founded by a quantitative hedge fund that happened to have a lot of Nvidia chips for their money management business. The founder was a math prodigy from China who wanted to become a quantitative hedge fund and he did that very successfully. They never took any outside funding until just a couple months ago.

21:57They raised $7 billion and the founder put in three billion of his own money. So his hedge fund's obviously doing pretty good. They are moving towards an IPO which should happen in early 2027.

And then Moonshot and the Kimi app, they may go public this year. They've already started to file for a Hong Kong IPO. So we'll add those as they IPO. The fund has the ability to invest up to 15% in private pre-IPO companies.

22:22As the fund scales up, we'll look and see if that's an opportunity we can find as well.

That's exciting. And what do you think are the thoughts of US investors, maybe not ones that are specifically focused on AI, but branching out into more alternative investments like the China models? Do you think there's a sense of nationalism or misunderstanding that maybe there's upside but it's scary territory to invest in these models that are competing with the US?

22:44

I think that people in the United States in particular have a lot of biases against China. The one thing I've learned investing in emerging markets—because we don't just invest in China, we invest in India and Brazil and any emerging or frontier market is part of our universe. It just so happens in the case of AI there's really only two players: the US and China. Taiwan and Korea, which are emerging markets, are making hardware but they're not really in the real part of the story.

23:11But the thing is most people have never been to China, right? They've seen headlines, they hear stories. Because it's communist, people assume it has all these oppressive elements. And I think people would be stunned if they went to China. I tell people, especially if they're young and don't have a family and a dog or a cat to hold them down, it's not hard to get to China, and it's quite cheap when you get there.

23:40And it's amazing. It's really the most advanced country in many ways. I mean, this is a terrible analogy, but if the aliens decided they finally wanted to land on the planet and talk to the leaders and they scoped out which country just looking at the cities, it's pretty clear they'd probably go to Chongqing or Shanghai because these are incredible places. So I recognize that there's a fear level that people have with China, but I think a lot of it's unfounded.

24:01

Definitely. Yeah, I have been. I spoke at a blockchain conference there nearly 10 years ago now in Beijing. So I did get the chance to experience it, but I know it's so big that was just one dot on the map. So have you personally, as you're talking about China, visited the companies or the warehouses and talked to them? And do you think that's an important factor for building this fund and presenting it to investors?

24:24

We haven't met with all of the companies, but I have spent lots of time on the ground in China over the last 20 years since I got involved. Our primary focus has been on the emerging market internet sector. I launched the first China technology ETF in 2010 with the ticker CQQ, so that was my first launch of an emerging markets tech-focused product 16 years ago.

24:45But starting around 11 years ago, I realized all the things that were happening in China with Alibaba and Tencent were starting to spread. So I launched EMQQ, which is our flagship fund that does emerging markets internet. I launched that to capture things like Mercado Libre, the Amazon.com of South America, and companies like Sea Limited in Southeast Asia. I've been focused on technology in emerging markets from the time it basically started, before there were smartphones.

25:08

So I haven't made a trip specifically to go and visit a lot of these hardware companies which are part of the EMX portfolios, the AI-focused portfolios.

What are some of your thoughts or predictions on how the AI race will move into 2027? You sort of referenced that tweet from Z.ai and Elon Musk. Are there any other thoughts and predictions on how the models will evolve in each of the countries by next summer?

25:34

Well, one thing that is just incredible about this AI thing as compared to the dot-com thing is the speed and breadth of developments. It's staggering. Consider this: a year ago, if you asked almost anybody, OpenAI was winning, right? That was the US model that was going to be the leader. And then now, less than 12 months later, people are saying Anthropic's worth twice as much as OpenAI.

26:03But consider this: one of the main reasons Anthropic has leapt ahead is the coding product, which wasn't even planned. It was a product that one of the engineers made for his own use in his spare time, and a couple of his buddies borrowed it or wanted to use it. All of a sudden it's one of the biggest parts of the company. That started this calendar year.

26:33So trying to predict where it's going to be a year from now is quite difficult. I would probably say we'll have more US open models, right? We're already seeing that. That's one prediction I would make: there will be more open models. I think there's going to be lots of other smaller models and more use-case-specific models. Everything's just happening so fast.

26:59In fact, Demis Hassabis, who's now the chairman of the DeepMind part of Google, was interviewed a couple months ago after he gave a talk. They had a Q&A afterwards and the first question was: what advice would you give to a college graduate coming into this AI world? He said the same thing everybody says: I would learn how to use all the different models. I would become the best at prompting.

27:30I would just make myself really good at using AI. That was a pretty standard question, but then he made the following comment. He said, "You know, I don't even know what's going on because I have a job. I can't keep up with the models that have already come out, let alone the five or 10 new things that are going to come out today."

28:00So everybody's sort of in this whirlwind and it's hard to make predictions because things are going so quickly.

Definitely, I agree. And there's going to be a lot. We just don't know what they're going to stumble upon, and then that's going to take off and that will be the thing, like you said, about the coding product.

28:27anthropic's worth twice as much as opening some some people have said that and

but consider this one of the main reasons Anthropic has leapt ahead is the coding product which wasn't even a plan it was a product that one of the engineers made for his own use.

Mhm.

In his spare time and a couple of his buddies borrowed it or wanted to use it

28:52and all of a sudden it's the biggest one of the biggest parts of the company. And that started that's been this calendar year

that happened. So trying to predict where it's going to be, you know, a year from now is quite difficult. Like I guess I would I would [snorts] probably pretty safe to say we'll have more US open models,

right? We're already seeing

29:17

that. I guess that's one prediction I would make that there will be more open models and I think there's going to be lots of other smaller models and more use case specific models but as as you know as we've touched on it's everything's just happening so fast in fact Disab and he's now the chairman the deep mind

29:49part of Google. He was interviewed a couple of months ago and after he gave a talk actually and then they had a Q&A afterwards and the first question was what advice would you give to a college graduate right that's coming into this AI world and he said the same thing everybody says he said I would learn how to use all the different models I would

30:12become the best at prompting I would just make myself really good at using AI now that was a pretty standard question but then he made the following comment. He said, "You know, I don't even know what's going on because I have a job. I can't keep up with the models that have already come out, let alone the five or 10 new things that are going to come out

30:33today." So, everybody's sort of in this whirlwind and it's it's hard to make predictions because things are going so quickly.

Definitely, I agree. And there's going to be a lot. We just don't know what they're going to stumble upon. And then that's going to take off and that will be the thing like you said about about the coding product and

30:58with the fund, what's the best way for people to learn more about the fund and access it for US investors?

Sure. Well, EMXETF.com is the website. That's EM for emerging markets, X, and then ETF. And then you'll see AICH, which is the full-stack China AI portfolio—a 24-company top-to-bottom portfolio of Chinese

31:29AI companies. And then the Tigers, TGRZ, you'll find. And that's just the currently three leading companies in the model space.

That sounds great. Well, I'm definitely going to be looking into it and following it. There's an opportunity that I feel like a lot of US investors are dying to get into US AI companies that are

31:53already at a trillion-dollar valuation. And there are other opportunities outside of this country that aren't at a trillion dollars yet and they have potential upside. And I think whether it's nationalism or misunderstanding of other cultures and what's going on in other parts of the world, it might limit people from having potential bigger returns elsewhere or at least better diversification of their portfolio. So I really appreciate your insights into the global AI growth and what we're seeing outside the US.

32:18And I feel like not enough people are paying attention to this yet.

Well, it's not going away. [laughter]

That's for sure. Well, thank you so much, Kevin, for taking the time. I will leave a link to the platform and all of the other ETF information as well as EMQQ Global in the show notes below. And I'd love to follow up in the near future.

32:43There will be much more to talk about with updates on AI. We could be back next week and there could be hours to talk about. But when there's more on the Tigers ETF, I think that's a great part to follow as these other China AI companies grow and see how those ETFs evolve. So congrats on the launch of that and looking forward to following up in the near future.

33:08the Tigers ETF, I think that's a great part to follow as these other China AI companies grow and see how those ETFs evolve. So, congrats on the launch of that and looking forward to following up in the near future.

All right. Thanks, Ashton.