Transcript

China’s AI catch-up begins to look inevitable

By Reuters

This transcript was created using speech recognition software, it may contain errors. Please review the episode audio before quoting from this transcript.
Disclaimer
The views expressed on this podcast are those of the participants, not of Reuters News.
Robyn Mak
And when OpenAI released ChatGPT back in 2022, there was a sense that China was falling behind in the AI race. And I think that really was a bit of a wake up call for a lot of the local players to step up their game.
Jonathan Guilford
The global race to dominate the artificial intelligence market heated up again this year, when Chinese upstart DeepSeek revealed its low cost, large language model R1 in January. Unlike in the US market, where OpenAI and Anthropic and a few select peers are ballooning in size, China is seeing growth in a large swathe of tech companies, big and small, all keen to roll out bits and pieces of artificial intelligence technology. But the bigger question is perhaps how this impacts the chip war. Trade tensions between the US and China have never been higher, and the administration of President Donald Trump is continuing to pull back on availability of the most cutting edge silicon used in training and running AI models. Furthermore, AI is the big story in US equity markets. How worried should US tech companies be about this seemingly cheaper and more plentiful competition? This and more is the focus of this week’s Viewsroom.
Jonathan Guilford
Welcome back to the Viewsroom, the weekly podcast that invites you, the listener, into a lively debate with Breakingviews columnists about the biggest stories of the week. I’m your host, Jonathan Guilford. China is muscling in on the AI party. News of DeepSeek’s technology sparked a surge in the country’s tech companies just as much as it cut back valuations, if only briefly, for the likes of Nvidia. If China is able to provide a much cheaper alternative to ChatGPT, how worried should US companies be? Is there a sense that China may just end up doing things cheaper and better, as it has in some other markets? Fortunately, I have two experts in AI, one who focuses on China and the other the US with me today, Robert Cyran, who’s based in our New York bureau, and Robyn Mak, who’s based in Hong Kong. Robyn, Rob, welcome to the Viewsroom.
Rob Cyran
Good to be here.
Jonathan Guilford
Rob, if we could start with you. So obviously, the US has gone to great lengths to limit China’s access to the most advanced AI chips or the equipment required to make them. What does this development tell us about where things are in the chip wars?
Rob Cyran
Well, a bunch of stuff. I mean, the first thing is that the walls are porous. You can actually buy, you know, servers with Nvidia’s latest chips in China, and the price isn’t that high. The premium isn’t that much higher. The other thing is that, you know, with the rise of DeepSeek, I mean, one thing DeepSeek showed is that there was a question over -- so over the past few years, there’s been a simple formula. If you make systems bigger, you have more chips, add more data, more and more parameters to the to the AI, what happens is the AI becomes more powerful. And then we started to hit those limits last year. And what DeepSeek showed is that, you know, so there was, there was that in the background. And then DeepSeek showed that they were able to become much more efficient using their chips and also and, and also because that’s happening, that means the importance of Nvidia chips becomes a little less. Also, AI is moving more towards inferencing rather than training. So that means instead of actually training the systems, it’s more about running the systems. And in that there’s more competition because you can use Nvidia chips, you can use other types of chips for the inferencing. And so, you know, add that all up. And what does that mean? It means that there’s simply more competition at least for Nvidia. And also it means that, you know, if there is more competition, if it’s easier to make the cutting edge systems, then there’s, then the prices you’re able to charge for them will be simply less, right.
Jonathan Guilford
And I kind of wonder, Robyn, turning to the picture in China here, it feels like what Rob is saying about, oh, you know, the barriers will come down, we’ll see a lot more competition, at least from afar -- it really looks like that’s happening. It seems like China is awash with AI companies at this point. I would imagine that’s a certain element of just being born out of necessity, because OpenAI and its ilk don’t operate in China. But like, how much of this do you think is a result of this, this kind of technological breakthrough that we’re seeing at DeepSeek?
Robyn Mak
I mean, keep in mind China has been investing a lot in AI well before DeepSeek emerged and even well before OpenAI emerged. So we’ve we’ve seen a lot of really big companies like Alibaba, Tencent invest billions to build out their cloud infrastructure to train their models. So it’s it’s -- AI has been a very hot buzzword for a very long time. And we’ve even had, you know, the Chinese AI dragons emerge at one period --
Jonathan Guilford
The Chinese AI dragons?
Robyn Mak
It didn’t -- yes, yes. There was four companies, including Megvii and and SenseTime that basically were at the forefront of China’s AI push, but it never really lived up to the hype. And when OpenAI released ChatGPT back in 2022, there was a sense that China was falling behind in the AI race. And I think that really was a bit of a wake up call for a lot of the local players to step up their game. And when OpenAI cut access, to, for Chinese developers, that was also another big wake up call. So starting 2024, you’ve really seen a lot of the Chinese companies make some real progress, including DeepSeek, because that was around the time that they started releasing some of their open source models. But you also have Alibaba and Tencent. They have open source models that consistently ranked very high globally and on some metrics, perform just as well or even better than, than their Western peers. So the the whole competition in China, I think it’s very, it’s quite a very familiar story for those that have been following the China tech. So that’s and there are good sides and downsides to that. I mean it’s like you said, I mean it’s it’s really just sparked a huge frenzy. You know, on the investment side, a lot of people are very excited about it. But at the same time, there’s a huge price war playing out in China as well, and that’s starting to spread abroad. But DeepSeek cutting prices, and I think now a lot of their global competitors, not just Chinese competitors, are having to cut prices.
Jonathan Guilford
And yet that’s so interesting because it kind of reminds me of what you’ve seen in a couple other markets, especially -- I cover automotives -- so Tesla had this, you know, initial big lead in the electric car market. And then you kind of saw the rise of China story and really, really brutal price competition there. I kind of wonder like what does, I guess, DeepSeek’s success, and whatever we’re seeing happening with pricing, tell us about the AI business model? Is it actually possible if you can build a passable large language model without all the chips and computing power and so forth? Like, what are the implications for OpenAI? And I guess just the rest of the market here, because there’s a ton of stuff that feeds into questions about how much people are going to be spending on AI, right?
Robyn Mak
Absolutely. I think, you know, DeepSeek has shown that you can train these AI models for a fraction of the cost. And also in terms of inferencing, I think there is a very critical question now that everyone needs to ask, which is how much chip demand does the world really need? And if DeepSeek can train an inference, you know, all its models using just, you know, 2000 Nvidia GPUs, then is what Alibaba, or Tencent, you know trying to amass in their in their chip pile, is that necessary? So I think that’s that’s a question that everyone is going to be wondering going forward.
Rob Cyran
Yeah. And for for the American side for the hyperscalers, you know they’ve they’ve made just absolutely gigantic investments. And at least part of them are predicated on the idea that they’re going to be some attractive margins in AI. That’s up to question. If there’s a lot of competition, you know, you’ll have price wars that may spur demand. But at the same time, you know, if there are a lot of these things out there, the margins probably aren’t going to be awesome. The other thing which we haven’t really touched on is, you know, there have been a a bunch of trade wars now launched. And you know, if there are, you know, a lot of the, a lot of the capex investments being made by a few American tech companies, you have to wonder, okay, will Europe be happy with a data center with all the European data being over here, or are there other countries being, you know, with their data abroad? And so it may spark, that may encourage countries around the world to say, okay, it we have to have the data center within our country. And what does that do for capex intensity, probably makes it go up. But on the other hand, maybe returns go down. So it, it introduces a ton of uncertainty.
Jonathan Guilford
Right. And it’s interesting you mentioned the trade war aspect to this because obviously we we have some pretty big announcements on that front this week. US president Donald Trump has upped tariffs on China to 20% across imported goods, adding on to various other measures. I mean, Robyn, can you, do you have a sense of what Xi Jinping’s views on the sector are and kind of how that plays into negotiation and jockeying with the US and elsewhere?
Robyn Mak
I mean, I think for the for the China tech sector, what matters a bit more is the chip restrictions. More so than than potential trade tariffs. So the chip restrictions have been a huge impediment and has really hampered progress for China’s AI. And I think, you know, I think there is a sense here that that’s not going to change. And if anything, the chips restrictions will only get tighter. Just given the way that things are going between the US and China, in terms of, you know what, how the government sees, you know, the tech sector and I, I mean, it’s a very much a very domestic issue. And there are lots of indication and signs and policies that, you know, the government intends to fully be, you know, in control and in the driving seat when it comes to regulating AI and not just for censorship reasons, but also because, you know, for example, it wants to develop industry standards. It was, you know, it’s the first government in the world to regulate algorithms and come up with AI regulations. So I think, you know, the view from Beijing is that this is a technology that we want to control and we will regulate. And it’s also critical that China is self-sufficient in this, in this area.
Jonathan Guilford
Do you think we are seeing signs that there’s a real sort of indigenous chip industry that is growing up here? That could be a challenge to Nvidia? I just ask because I, I don’t even have a good sense of who the players are or how sophisticated they are, or whether we’ve had recent news about them making kind of advances similar to what DeepSeek did on the software side.
Robyn Mak
I think so, I mean, you know, Huawei is leading the forefront of, you know, China’s chip progress. And, you know, they have shown that, you know, from a design perspective, they are able to design chips that are probably just like as good as Nvidia. But the biggest hurdle is the hardware and chip making equipment that they can’t get access to. And, you know, there’s been a lot of progress made on that front too. So Huawei has redesigned its chips, you know, so it’s a bit more efficient. And maybe it requires less computing power, even a lot on the software side. You know, companies like DeepSeek are finding ways around the chip restriction as well. So it is I mean, there’s still a long way to go, but China is making progress and it is catching up. And I think, you know, the the belief here is that it’s not a matter of if it’s a matter of when.
Jonathan Guilford
I’m so curious about that because, Rob, I wonder if you have a take on this kind of, I guess, the US view on what the moats are, maybe the oh you know, this will ensure, let’s say, Western technological supremacy on the chip side. Like, do you have a sense of kind of what those are seen as being and, you know, maybe how durable that is?
Rob Cyran
Yeah, I mean, I agree with Robyn that, you know, China’s just pouring money into chipmaking. And we’ve seen this happen, this game, before. We saw it with, you know, the rise of the chip industry in South Korea. Countries can catch up if they if they pour a lot of money into, you know, scientists and sending people to grad school and, you know, developing chips and forcing their own companies to use their chips. So it’ll happen. It’s a question of when. In terms of how durable, you have to look at various moats, you know, part of the reason why Nvidia chips are so common is because the software as well, that, you know, it’s become the de facto standard. So for a lot of people around the world, when they start up, they’re going to start using Nvidia chips just because that’s what everyone else uses. But on the other hand, you know, they could provide an impediment in China because, you know, it may be more efficient to be like, hey, let’s not use CUDA, let’s use some other, you know, locally grown version. I mean, maybe squeeze some efficiency gains. Like that was the whole story behind DeepSeek is that they had limited access to chips, and or at least part of the story. And so they were they were more concentrated on wringing out as much efficiency out there as they could out of the system. And then, you know, just in terms of capital, if anything, I would expect Chinese chip companies have more, more access to capital and Western companies because it isn’t necessarily just an economic decision. It’s also a national security decision.
Jonathan Guilford
Right. And, you know, we’ve been going back and forth and mentioning the DeepSeek efficiency breakthrough. One thing I do wonder is we had a big market freak out. Obviously, when DeepSeek first released its all reasoning model, information in that initial paper implied the huge efficiency gains that we’re all talking about, that run counter to the narrative that everybody needs to spend many, many billions to make AI work. And I think, Robyn, if I’m right, just last week, the company followed up with another paper that included some more color that made the claims, if anything, like even more astounding. I know you took a quick look through that, obviously with the caveat that none of us here are semiconductor engineers -- I mean, do the numbers make sense to you, or at least does the do they imply things that seem radical enough that the the scale of the reaction makes sense?
Robyn Mak
It’s it’s a hard one to answer because I think you can argue both sides. I mean, you know, on one hand they did release some pretty impressive profit margin numbers, but that was all based on the theoretical revenue. So how much sales they can generate, if they were able to charge the prices they wanted to charge. And based on that, I mean, think they had something like an 80%-plus gross profit margin, which is very impressive. And I think it does confirm what a lot of people thought in January, which is that, you know, this doesn’t need, you, there is a business model out there, you don’t need to be spending so much money. There can be profit. But even DeepSeek acknowledge that this is theoretical revenue and that they in reality do not charge us much. And, you know, in fact, they actually offer discounts. And there is a wider question of if they do charge what they wanted to charge, will they even have this much revenue to begin with or this many users? So there’s that. But yeah, I think it really did, I think it was significant in the sense that it just confirmed a lot of what people believed about DeepSeek in January, which is that, you know, the business models really need to be examined going forward.
Rob Cyran
And another interesting thing about that, with all your caveats totally in mind, you know, like, I mean, the fact that they think the gross margins could be that high, it also implies there’s a lot of room for cost cutting, right? I mean, in other words, when, you know, if other companies get in here, the prices are going to totally drop. And that may, you know, maybe that’ll spur adoption.
Jonathan Guilford
Right. And I guess that’s kind of the final thing that I wonder about, which is just whether the market, even after all of the the drops and the swings and the seesawing that we’ve seen, whether there’s a sense that investors are actually beginning to kind of take this into account and, and adjusting their views in a coherent way with that. I mean, Rob, on the US side, we’ve seen, you know, OpenAI working towards a fundraising round. We’ve seen Anthropic reportedly hit the kind of new valuation milestones, like when you think about the conversations that you’re having with folks or kind of what you see in the market, do you think there has been a meaningful kind of investor adjustment based on on, I guess just like processing what it is that DeepSeek has done here?
Rob Cyran
No, not really. I think, I think they are still, you know, there was that momentary freak out. But then there’s also kind of the, the, the desire for it to keep on going on. And, you know, you have to also step back and be like, okay, so not everyone is doing the same thing. Like, you know, a lot of companies like, you know, big companies like Microsoft, they’re also selling cloud services. So, you know, even if it becomes a, you know, commoditized AI, they’ll still make money off of it. Or, or Meta will do as well. So it’s kind of hard to say, okay, well all of them are banking on one, you know, kind of dominant AI thing is, as was the case, kind of more, was more the case, like last year. So there’s been some adjustment in business models. But in terms of valuation, it’s still I think everyone is still kind of hoping for a market which isn’t here yet. And that could present problems, you know, because, okay, right now they can afford the capex figures, all the big US companies, but they’re growing so fast that it becomes a bigger and bigger swing every quarter.
Jonathan Guilford
Right. I mean, does the story seem sort of similar in China, Robyn, or do you think investors are taking a slightly different view?
Robyn Mak
I think it’s a it’s a slightly different view. So China tech stocks have been completely beaten down for the past few years. We’ve had regulatory crackdown. We’ve had the pandemic. So the valuation multiples were extremely, extremely low. And then when DeepSeek released this model, I think investors were really excited again about China tech and what it could do. And I think that led to a reevaluation of a lot of the valuations. So we’ve seen, you know, Alibaba’s stock go up, I think more than 50% this year as well as, you know, the Hang Seng Index as well has rallied quite a bit since January. And this was all down to DeepSeek. But having said that I mean, you know Alibaba is still trading on, you know, an earnings multiple well below its historical average. It’s still the market cap, the market cap is still half of what it was, in its, you know, 2021 peak. So it’s all relative I think.
Jonathan Guilford
Right. So collar tugging optimism in the US and trying to get out of the dumps in China. It’s an interesting kind of split there. But anyway, thank you so much, Robyn and Rob. This has been fascinating, and I’m sure we’re going to have you back on to dive through the inevitable next big thing that comes down the pipe in two weeks. So thank you so much for joining us on the Viewsroom.
Robyn Mak
Thanks Jonathan.
Jonathan Guilford
Thanks for tuning in. This podcast was produced by Oliver Taslic in London. You can listen to a new episode of the Viewsroom every Thursday on the Reuters app or your favorite platform. And don’t forget to tune in to our sister podcast, The Big View, every Tuesday, as well as other great podcasts from the Reuters team. If you like what you heard, please follow the Viewsroom and let us know what you thought. And check out our views on the biggest stories in business and finance every day at Breakingviews.com and Reuters.com.