AI transcript
0:00:02 Support for the show comes from Panerai.
0:00:12 For more than 150 years, Panerai watches have showcased the brand’s rich heritage of Italian design and Swiss horological expertise, making it the perfect timepiece for modern heroes.
0:00:15 So this is my weakness.
0:00:17 I have several Panerai watches.
0:00:21 It’s the only watch I wear, and trust me, Panerai is the way to go.
0:00:30 Panerai’s new Luminor Marina timepieces honor the brand’s storied legacy and history of pioneering innovation while enhancing iconic aesthetics, reliability, and functionality.
0:00:35 Visit Panerai.com to discover the world of Panerai and find the boutique nearest you.
0:00:37 That’s P-A-N-E-R-A-I.com.
0:00:51 Support for this show comes from the Audible original, the downloaded two, ghosts in the machine, quantum computers.
0:00:57 The next great frontier of technology, offering endless possibilities that stretch the human mind.
0:01:05 But for Roscoe Cadoulian and the Phoenix Colony, quantum computing uploads the human mind with life-altering consequences.
0:01:17 Audible’s hit sci-fi thriller, The Downloaded, returns with Oscar winner Brendan Fraser, reprising his role as Roscoe Cadoulian in The Downloaded 2, Ghosts in the Machine.
0:01:20 This thought-provoking sequel from Robert J.
0:01:24 Takes listeners on a captivating sci-fi journey.
0:01:30 A mind-bending must-listen that asks, what are you willing to lose to save the ones you love?
0:01:33 The Downloaded 2, Ghosts in the Machine.
0:01:36 Available now, only from Audible.
0:01:49 Support for this show comes from the Audible original, The Downloaded 2, Ghosts in the Machine.
0:01:52 The Earth only has a few days left.
0:01:58 Roscoe Cadoulian and the rest of the Phoenix Colony have to re-upload their minds into the quantum computer.
0:02:03 But a new threat has arisen that could destroy their stored consciousness forever.
0:02:12 Listen to Oscar winner Brendan Fraser reprise his role as Roscoe Cadoulian in this follow-up to the Audible original blockbuster, The Downloaded.
0:02:18 It’s a thought-provoking sci-fi journey where identity, memory, and morality collide.
0:02:23 Robert J. Sawyer does it again with this much-anticipated sequel that leaves you asking,
0:02:26 what are you willing to lose to save the ones you love?
0:02:30 The Downloaded 2, Ghosts in the Machine.
0:02:32 Available now, only from Audible.
0:02:44 The latest Chinese AI sensation, which is called Kimi K2,
0:02:50 cost just a fraction of what it would cost to train an AI model in America.
0:02:54 It cost about 4.6 million US dollars.
0:03:01 That is just a tiny fraction of what OpenAI has been investing in its large language models.
0:03:05 Welcome to China Decode. I’m Alice Han.
0:03:06 And I’m James King.
0:03:12 Well, in today’s episode of China Decode, we are discussing whether China is building the Android of AI,
0:03:18 while the US builds the ultra-premium iPhone, powerful but too expensive to dominate globally.
0:03:25 Why the latest spat between Japan and China could reshape the balance of power in the Asia-Pacific region.
0:03:30 And how Starbucks and the coffee wars reveal the fierce fight for China’s consumer market.
0:03:32 All right, let’s get right into it.
0:03:37 Right now, the biggest story in tech may be the one the US didn’t see coming.
0:03:43 China’s low-cost, open-source AI is suddenly everywhere in Silicon Valley.
0:03:48 While the US is spending billions of dollars building ever-bigger models and data centers,
0:03:55 and inflating what some analysts say looks like an AI bubble, China is doing the exact opposite, it seems.
0:04:02 Cutting prices, opting for open-sourcing weights, and shipping models that are cheaper, lighter, and surprisingly competitive.
0:04:05 And American companies are flocking to them.
0:04:13 Airbnb’s CEO says his team ditched ChatGPT for Alibaba’s Quen model because it’s, quote-unquote, fast and cheap.
0:04:17 Venture firms are switching to Moonshot’s Kimi, LLM,
0:04:23 and developer data show nearly half of the most used models in the US last week were Chinese.
0:04:30 Alibaba just slashed prices on its flagship model again as part of a full-blown AI price war,
0:04:33 and a sign that involution has come to Chinese AI.
0:04:41 Former Google CEO Eric Schmidt is warning that most countries may end up using Chinese models simply because they’re free.
0:04:46 James, we’ve talked a lot about, I think, Chinese AI over the last few weeks,
0:04:51 and I definitely think, I’m very excited to see the models that are coming out of the mainland,
0:04:56 but also the fact that the West is catching up to this theme that we’ve been talking about for a while,
0:04:59 which is don’t write off Chinese AI.
0:05:01 They are going to go for it in a very different model.
0:05:05 And I think the recent announcements from these tech companies,
0:05:12 some even, I would say, pretty big startups that are using and relying Chinese LLMs is pretty dramatic,
0:05:18 and I think it’s a different narrative than what we’ve been seeing in the last few years.
0:05:22 I just want to go right into it and see what your take is on this Chinese model of AI,
0:05:29 and whether or not this kind of Android AI approach to making cheaper, faster, more efficient models
0:05:33 is ultimately going to win out in a global race between China and the US of AI.
0:05:35 Yeah, that’s a really great question, Alice.
0:05:41 I’ve started to notice just in recent days, actually, that people are using the word bubble
0:05:46 with regard to America’s AI efforts quite a bit more now.
0:05:49 I think in the past they’ve been reluctant to do that.
0:05:54 My own take is that this is still very far from conventional wisdom,
0:05:59 and the US-China AI race is going to be a very long-run thing.
0:06:01 We can’t call it yet.
0:06:04 There’s going to be many ups and downs and fits and starts.
0:06:12 But as things stand right now, I would say that some big names in the US are looking quite bubbly.
0:06:14 And I think the key point is this.
0:06:21 The bubbly aspects of American AI, as I see it, are made in China,
0:06:31 by which I mean that China’s approach to AI is to build AI models at a much lower cost than we can find in the US.
0:06:36 Also, the cost of using those Chinese AI models is much lower.
0:06:46 And crucially, the performance of the Chinese AI models is kind of just as good or catching up to US standards at the moment.
0:06:53 So I think that that is the reason why we saw Eric Schmidt, the former CEO of Google,
0:07:00 very recently say, as you’ve already mentioned, you know, the vast majority of governments and countries
0:07:08 who don’t have the kind of money that the West has will end up standardizing by using Chinese models,
0:07:10 not because they’re better, but because they’re free.
0:07:18 So I think really we’re entering a very interesting phase in the US-China AI race.
0:07:22 I should just mention before I pass it back to you, Alice, that, you know,
0:07:24 there needs to be a little bit of nuance on this.
0:07:34 Not all US LLMs are closed source, and not all Chinese large language models are open source.
0:07:36 So it’s a mixed picture.
0:07:43 You know, we’ve got Lama from Meta, we’ve got Fi from Microsoft and Gemma from Google.
0:07:46 All of those are open source.
0:07:50 And of course, it’s famous that many of the big and successful Chinese models,
0:07:55 such as DeepSeq and Q1, as you already mentioned, are open source too.
0:08:02 But I just want to, I just want to make that clear where, you know, it’s, it’s a variegated picture.
0:08:07 It’s not the case that everything done in China is open source and everything done in the US is closed source.
0:08:12 I just want to leave you with one extraordinary statistic.
0:08:24 And that is that the latest Chinese AI sensation, which is called Kimi K2, cost just a fraction of what it would cost to train an AI model in America.
0:08:29 It cost about 4.6 million US dollars.
0:08:37 That is just a tiny fraction of what open AI has been investing in, in its large language models over in the US.
0:08:39 But how do you see this, Alice?
0:08:42 Do you, do you think that, you know, it’s too early to call?
0:08:47 Or do you think that the American model is looking bubbly and the Chinese model is looking lean and mean?
0:08:50 I think that’s a great way to phrase it.
0:08:52 Bubbly versus lean and mean.
0:09:10 I think I’m more in the camp of that kind of narrative, although with the caveat that, as we’ve previously talked about, James, the US has thrown billions of CapEx into the data center rollout, into leading edge LLMs, into the frontier labs and hyperscalers.
0:09:19 And as a result, I think is probably going to be further along in the AGI, artificial general intelligence race or super intelligence, as we previously mentioned.
0:09:29 But the Chinese model, to your point, James, is leaner and meaner in the sense that, you know, it is doing these models more efficiently, much cheaper to train.
0:09:39 What was interesting for me to notice, and you already mentioned it, James, is that not only are they cheaper on the inference costs, but also on the training costs.
0:09:49 I remember meeting, I believe, the father of AI in China, who wrote that famous book, AI Superpowers, which seems like a lifetime ago, but I highly recommend that read, I believe, in 2018.
0:09:52 So before this really ramped up as a serious issue.
0:10:03 And he had said at the time a year and a half ago that China may not be able to compete on training costs because the US has dedicated a huge amount of CapEx towards training.
0:10:10 And it has some of the best data sets and training models in the world, but China can rapidly decrease the inference costs.
0:10:13 Now it seems like they’ve decreased both the inference and training costs.
0:10:24 To your point, the Kimi K2 open source model is just $4.6 million, as you say, that’s 1,500 times less than what open AI spent on R&D this year.
0:10:29 And again, just to follow up on your point, James, so that listeners can appreciate this.
0:10:33 I think it’s a bit of a misnomer to say that China is leading the way on open source.
0:10:48 I’ve been talking to some AI scientists more recently who’ve said that actually the majority of the quote-unquote open source models are really open weight, meaning that the underlying code and data set are not freely available, but the weights or the training parameters are.
0:10:51 That’s the difference between an open weight versus an open source.
0:10:59 But the fact of the matter remains that China is going through what I see as a Cambrian explosion in terms of different LLMs.
0:11:08 And we’ve talked about this theme of involution, of price wars, of excess capacity, of hyper-competition.
0:11:16 I think this is already starting to be played out in the Chinese context, where you have a lot of different LLMs in the race or in the ring.
0:11:23 Kimi just overtook of QN and DeepSeek very recently in terms of its parameters and performance.
0:11:35 I think the fact that China is being used, its LLMs are being used by American companies, is a pretty telling and a compelling sign that consumers are driven by price.
0:11:44 And the fact that these LLMs can be cheaper than their American counterparts, I think, is making them more desirable, even amongst some of these Western companies.
0:11:57 Which is why, you know, I find that the recent talk about Chinese LLMs being competitive is coming at a time in which a lot of people watching Wall Street are very worried about the AI bubble in the US.
0:12:00 I think the two of them coexisting as narratives is not a coincidence.
0:12:05 One question I do have for you, James, is have you actually used the different models?
0:12:13 I’ve been playing around a little bit across the different models from the US and China, but I want to get a sense of what you’ve experienced.
0:12:15 Yeah, I have actually.
0:12:16 I’ve used DeepSeek quite a lot.
0:12:18 I continue to use DeepSeek quite a lot.
0:12:24 Obviously, I’ve used ChatGPT quite a bit and I’ve just started to play around with Kimi as well.
0:12:31 You know, to be honest with you, I’m using it at such a low level that I don’t really notice that much of a difference.
0:12:35 I’m sort of using it as a dictionary, really, and as a database.
0:12:38 So, yeah, I mean, they’re all pretty good.
0:12:49 I have noticed that the DeepSeek results are a little bit skewed away from politically sensitive topics that, you know, you might expect from a Chinese LLM.
0:12:53 But I haven’t noticed anything too egregious, I must say.
0:12:58 But there is this question about security, which is hanging in the background.
0:13:05 And I think that a lot of people are concerned by the question of security with regard to Chinese large language models.
0:13:18 I should cite an article by the Financial Times just recently that says that Alibaba provides tech support for the Chinese military against targets in the United States.
0:13:20 Now, I need to be very careful about this.
0:13:29 The claims, which couldn’t be independently verified by the FT, were rejected very forcefully by Alibaba.
0:13:35 They said that this information leaked from your source is complete nonsense.
0:13:37 So that’s a very strong denial.
0:13:45 But there is always going to be this issue of security hanging over the Chinese large language models, I think, to users in the West, at least.
0:13:47 Yeah, I completely agree.
0:13:49 And I think that we should watch the space.
0:13:55 I think it’s a matter of time before these LLMs get politicized in the same way that semiconductors have been.
0:13:58 I actually think that that is a likely outcome.
0:14:03 The more that we talk about Kimmy and DeepSeek and the more that we worry about an AI bubble in the U.S.,
0:14:08 I think the more political backlash there will be against some of these Chinese LLMs.
0:14:14 But just to round up this discussion, I thought it was interesting as I looked into Kimmy’s architecture.
0:14:21 One of the reasons that it is more efficient than the American LLMs is that it uses this thing known as the mixture of experts design,
0:14:25 whereas ChatGPT and some of these other LLMs are using a dense model.
0:14:33 This mixture of experts design only utilizes certain parameters when created, and a dense model uses them all.
0:14:39 And apparently, the Kimmy K2 is over a trillion parameters, but it will only use about 32 billion at a time.
0:14:50 And this points to, I think, one last note that I’ll end on, is it points to the example in which Chinese AI is, at a research level, evolving in a very, very different way.
0:14:55 And so the more that I think about this space, and I’m trying to read up very quickly, because it’s changing so quickly,
0:15:04 I see that sort of whole evolution of two very, very different species of AI that I think is happening right now between China and the U.S.
0:15:07 Okay, we’ll be back with more after a quick break, so stay with us.
0:15:15 Support for this show comes from Odoo.
0:15:21 Running a business is hard enough, so why make it harder with a dozen different apps that don’t talk to each other?
0:15:23 Introducing Odoo.
0:15:26 It’s the only business software you’ll ever need.
0:15:30 It’s an all-in-one, fully integrated platform that makes your work easier.
0:15:34 PRM, accounting, inventory, e-commerce, and more.
0:15:35 And the best part?
0:15:40 Odoo replaces multiple expensive platforms for a fraction of the cost.
0:15:43 That’s why over thousands of businesses have made the switch.
0:15:44 So why not you?
0:15:47 Try Odoo for free at odoo.com.
0:15:50 That’s O-D-O-O dot com.
0:15:56 Saks Fifth Avenue makes it easy to holiday your way.
0:16:02 Whether it’s finding the right gift or the right outfit, Saks is where you can find everything.
0:16:08 From the perfect Chloe bag for your hard-to-shop-for sister to a Prada jacket for a fancy holiday dinner.
0:16:15 And if you don’t know where to start, Saks.com is customized to your personal style, so you can save time shopping.
0:16:20 Make shopping fun and easy this season, and find gifts and inspiration to suit your holiday style.
0:16:24 Well, welcome back.
0:16:27 Tensions are well and truly boiling in East Asia.
0:16:36 China and Japan are locked in a fiery diplomatic clash over Taiwan, with threats, insults, and even references to beheading now making the headlines.
0:16:45 And it all started when Japan’s new Prime Minister, Senai Takeichi, said a Chinese attack on Taiwan could trigger a Japanese military response.
0:16:49 This is a rare explicit stance from a sitting leader.
0:17:03 And Beijing responded swiftly, summoning ambassadors, issuing warnings of a crushing defeat if Japan intervenes, and lashing out in state media, calling Takeichi a quote-unquote evil witch, and warning Chinese citizens to avoid travel to Japan.
0:17:06 The spat highlights more than just the issue of Taiwan.
0:17:13 It also involves historical grievances, territorial disputes, and Japan’s shifting security posture.
0:17:20 Japan, America’s key ally in the region, may increasingly act as a deterrent while the U.S. pulls back.
0:17:27 And Takeichi’s hard line could boost her actual standing at home and abroad, even with conservative American allies.
0:17:33 James, I think that this is giving me 2010 deja vu all over again.
0:17:39 When we had the Senkaku Island incident, we had the Rare Earths embargo for a couple months in Japan.
0:17:47 We had a lot of Chinese, everyday Chinese citizens boycotting Japanese goods, burning Japanese goods, looting Japanese stores.
0:17:53 And I’m not surprised, frankly, that China has reacted in a pretty aggressive manner.
0:18:00 And I wouldn’t be surprised if everyday Chinese boycott Japan entirely, as they’ve done in the past.
0:18:05 Anytime there has been an anti-Japan or hyper-nationalist movement.
0:18:19 It’s no surprise that Chinese airlines announced over the weekend that they would allow all passengers to cancel or change Japan-bound flights free of charge, after Beijing basically urged the citizens to avoid visiting Japan.
0:18:28 This comes at a time, I think, where Trump is clearly trying to pull back from U.S. commitments in the region.
0:18:37 Japan, I think, is sitting there pretty worried about the state of affairs, given that it relies so heavily on the deterrence that the U.S. offers.
0:18:41 And that Japan is not a nuclear power, which we’ll get into in just a bit.
0:18:48 I certainly would be nervous sitting in this region at a time where politically both sides want to play hardball.
0:18:52 But what’s your take on this, James, having lived in the region for quite a bit of time?
0:18:59 Yeah, I mean, I’ve lived in Japan and China and Taiwan and, you know, the countries that are in play.
0:19:02 It’s amazing to me how quickly this has escalated.
0:19:08 It’s gone from zero to 60 or zero to 120 in what feels like a very short period of time.
0:19:21 You know, I woke up on Sunday morning to see, on my social media feeds, three Chinese warships steaming towards the Senkaku Islands, which are territory disputed by Japan and China.
0:19:23 It really was quite dramatic.
0:19:34 And I think it’s important for the West as well, particularly for the United States, because, of course, Japan is America’s leading ally in Asia.
0:19:41 Japan also plays host to several U.S. military bases, and these are really major.
0:19:49 The Yokosuka Naval Base, just near Tokyo Bay, is the home port for the U.S. 7th Fleet.
0:19:52 And there’s a large number of U.S. ships there.
0:19:58 And it also has the nuclear-powered aircraft carrier called the USS George Washington.
0:20:02 This is one of the, you know, the premier ships in the U.S. Navy.
0:20:13 So, any kind of a flare-up between Japan and China, especially if it has a military dimension, which this one very clearly does, is a big issue for Washington.
0:20:20 And I imagine it’s quickening pulses all over the U.S., just as it is for you and I.
0:20:25 Of course, the Japanese side is now trying to de-escalate.
0:20:35 They have sent over Masaaki Kanai, who is head of the Foreign Ministry’s Asian and Oceanian Affairs Bureau.
0:20:40 They sent him to Beijing to see if he can sort of try and patch things up.
0:20:47 But, in the meantime, as you’ve already alluded to, this is hammering some Japanese stocks.
0:20:53 You know, we’ve seen the Isetan and Mitsukoshi department store owner.
0:20:58 That stock price has fallen over 10%.
0:21:04 And also, the Tokyo Disneyland operator, that’s lost nearly 6%.
0:21:05 Japan Airlines is down.
0:21:11 This is all because these companies benefit hugely from Chinese tourists going to Japan.
0:21:19 And as you’ve said, Alice, once nationalist fervor grips the Chinese people, as it is clearly starting to do,
0:21:26 then you immediately see a drop-off in the willingness of Chinese to buy Japanese stuff and to travel to Japan.
0:21:28 How do you see this?
0:21:30 I mean, there is obviously historical context here.
0:21:38 I mean, why is it that Chinese feel so triggered, as it were, by Japan at these moments?
0:21:43 Well, I mean, China had a harrowing experience in World War II at the hands of Japan.
0:21:49 Japan occupied parts of China, even getting up to Shanghai, where my family is originally from.
0:21:57 And the rape of Nanjing or even the systemic wiping out of populations in China are deeply, deeply remembered
0:21:59 within the Chinese collective.
0:22:08 And I think that that probably intensifies a lot of already, I think, strategic rivalries and issues at play.
0:22:11 So there’s a deep set historical mistrust.
0:22:17 But at the same time, I think when we look at it from a real politique standpoint,
0:22:24 Japan is deeply worried about an ascendant China and what that means for the Asia-Pacific region.
0:22:36 If the U.S. decides that it ultimately doesn’t want to be a regional and global hegemon, Japan is in a very, very, I think, isolated state because it has no nuclear deterrence of its own.
0:22:47 It cannot, according to the treaty that it signed after World War II with the Americans, it cannot have an offensive military capable of warfare.
0:22:55 And it has the three non-nuclear principles, meaning that it cannot possess, produce or allow the introduction of nuclear weapons onto its territory.
0:23:00 So I think Japan, from a real politique, realist standpoint, is in a very desperate situation.
0:23:07 And I’m not surprised that Takeichi is trying to draw into question the three non-nuclear principles.
0:23:12 That is another reason why China is worried, because, and this is according to people close to her,
0:23:19 she apparently wants to call to question the strict adherence of the no introduction of nuclear weapons principle,
0:23:25 because ultimately she wants to allow U.S. nuclear-armed vessels to sit in Japanese ports
0:23:32 to basically have a credible nuclear deterrence against China in a potential showdown over Taiwan or in the region.
0:23:41 So this is, I think, all driven by the tectonic shifts of power, mainly in terms of the U.S.-China relationship in the region.
0:23:47 And I think, ultimately, I wouldn’t be surprised, even if we do see a de-escalation in the short term,
0:23:54 that there is an increasing rivalry between Japan and a mistrust between Japan and China,
0:24:00 and that Japan gets closer with, say, Korea, India, Australia,
0:24:06 some of these other countries that could be on the opposing side of China if there is a showdown in the region.
0:24:10 But just very quickly, James, having lived in Japan,
0:24:15 do you think that there has been a shift in Japan’s attitude towards China in the last few years?
0:24:22 Well, I mean, I think that certainly Prime Minister Takeichi comes from the wing of the Liberal Democratic Party,
0:24:27 the LDP, which is known to be more hawkish towards China.
0:24:32 She’s seen as a little bit of a successor to Shinzo Abe, the previous prime minister,
0:24:36 who was, you know, considered to be fairly hawkish towards China.
0:24:40 And I think that is already playing out in the minds of the Chinese
0:24:48 and making China extra anxious about what the next few years of Takeichi’s tenure is going to produce.
0:24:57 So I think that this flare-up that we’ve seen at the moment could set the tone for however long Takeichi is prime minister,
0:25:01 even if there are attempts by the Japanese, as I’m sure there will be, to de-escalate.
0:25:07 Like you, I see this as a, well, maybe not a continuous source of tension,
0:25:12 but, you know, a general undertow of tension between the two countries.
0:25:14 Well, James, we’re going to take a quick coffee break.
0:25:17 So stay with us as we wait for the next segment.
0:25:29 Support for this show comes from the Audible Original, The Downloaded 2, Ghosts and the Machine.
0:25:32 The Earth only has a few days left.
0:25:38 Roscoe Cadulian and the rest of the Phoenix Colony have to re-upload their minds into the quantum computer,
0:25:43 but a new threat has arisen that could destroy their stored consciousness forever.
0:25:48 Listen to Oscar winner Brendan Fraser reprise his role as Roscoe Cadulian
0:25:52 in this follow-up to the Audible Original blockbuster, The Downloaded.
0:25:58 It’s a thought-provoking sci-fi journey where identity, memory, and morality collide.
0:26:03 Robert J. Sawyer does it again with this much-anticipated sequel that leaves you asking,
0:26:06 what are you willing to lose to save the ones you love?
0:26:10 The Downloaded 2, Ghosts and the Machine.
0:26:12 Available now, only from Audible.
0:26:21 Well, welcome back.
0:26:26 This is my favorite section of this podcast, not just because I am a caffeine addict,
0:26:29 but also I find this story quite astonishing.
0:26:34 China’s coffee wars are growing up and Starbucks is taking cover, it seems.
0:26:37 After nearly 30 years at 8,000 stores,
0:26:42 America’s Starbucks is selling a majority stake in its Chinese business to bore you your capital,
0:26:44 keeping just 40%.
0:26:49 Local rivals like Luckin and Koti are surging ahead, opening thousands of outlets,
0:26:53 cutting prices, and serving drinks, very bizarre drinks, I would say,
0:26:58 tailored to local tastes, from pork drizzled lattes to alcohol-infused specials.
0:27:04 Starbucks’ premium strategy and slow product rollout just couldn’t keep up in the end.
0:27:09 This joint venture gives Starbucks local insight, faster expansion, and digital integration,
0:27:14 critical in a market that remains dominated by WeChat Pay and Alipay.
0:27:18 Now, other U.S. brands like Burger King and Decathlon are following suit,
0:27:24 selling their steaks to local partners to survive amid China’s hyper-competitive consumer landscape.
0:27:30 Now, the big question is, is China still a land of opportunity or have homegrown brands
0:27:32 completely rewritten the rules?
0:27:37 James, I love this story for many, many reasons, as already alluded to.
0:27:44 But also, I had noticed from a friend who sent me a message earlier in the year that they had picked up a coffee,
0:27:46 a pineapple cold brew from Luckin’s New York store.
0:27:52 And I realized then that Luckin, a coffee brand that I’d long associated just with China,
0:27:55 has become very global in the last few years.
0:28:00 And it was funny for me to see this picture because I still remember during COVID went bankrupt.
0:28:05 And prior to that, I had cases of financial fraud.
0:28:10 So I was surprised to see that they not only have become profitable again,
0:28:16 but they are trying to relist in the NASDAQ and now have at least two stores that I know of in New York,
0:28:21 where apparently the regular priced ice latte costs as little as $5.75.
0:28:26 Now, I’ve compared that to some of my colleagues living in New York who say that at Starbucks
0:28:32 and other places you could get an ice latte for $6 to $8 US dollars, but listeners should definitely correct us.
0:28:39 I do know for a fact that people who download the app as first-time users can have a coffee for as little as $1.99.
0:28:46 So this is a strategy that they’ve definitely kept alive that has been successful in the Chinese context.
0:28:52 But I am a little bit skeptical that they can do a reverse Starbucks and be as successful in America
0:28:55 because I do think the Chinese market is quite different
0:29:00 and that they’d have to burn a lot of cash to really survive in America.
0:29:05 But it’s interesting to see these hungry executives in Luckin,
0:29:08 but also another tea company that I really love in China called Chaji.
0:29:11 They do specialty iced teas.
0:29:16 These executives are really trying to jump over the Pacific and be as successful in America as they were in China.
0:29:20 I’m a little skeptical, but James, please refute me if you disagree.
0:29:22 I love that.
0:29:23 I mean, I hadn’t heard that story.
0:29:27 Luckin coffee going to America and selling coffee for $1.99 a cup.
0:29:31 Yeah, I mean, that’s really quite something.
0:29:33 It’s going to help Nandani make New York more affordable.
0:29:35 That’s right.
0:29:36 Well, it certainly will.
0:29:41 I mean, I certainly wouldn’t have any qualms about, you know, buying a coffee for $1.99 if it tasted OK.
0:29:49 But, you know, this story has quite a lot of resonances for me, too, because I was in Beijing when Starbucks arrived in 1999.
0:29:55 I reported on the first Starbucks shop, which was in the World Trade Center.
0:30:08 And, you know, at that time, we all, I certainly wasn’t the only one, or the journalists there were all writing about how Starbucks had the chutzpah to try to sell coffee to a country with thousands of years of tea culture.
0:30:12 There was a lot of quotes from people saying, this will never catch on.
0:30:15 And then look how wrong we were.
0:30:18 You know, there’s literally thousands and thousands of stores.
0:30:20 So that was one thing.
0:30:31 But the other really fascinating aspect of this story to me, which I hope you’ll be able to help me with, because I think you’re much more of an expert than I am, is what it says about Chinese tastes.
0:30:41 You already mentioned that the halo effect of foreign brands, not just Starbucks, but plenty of others in China, is beginning to slip.
0:30:45 We’ve seen Apple giving ground to local competitors.
0:30:48 We’ve seen Estee Lauder giving ground to local competitors.
0:30:52 LVMH in the luxury sector doing the same thing.
0:31:04 So the other aspect of it is the localization, which maybe includes a better understanding by Chinese companies of what Chinese really want to drink or to eat.
0:31:09 And I don’t know, maybe this is a bit outlandish, Alice, please tell me.
0:31:19 But I found something online about a coffee shop that opened in southwestern China that had cooked pork intestines, like on a skewer on top of the coffee.
0:31:28 And then there was also – actually, this was Starbucks earlier on, I think it was the end of last year – putting like pork fat drizzle onto their coffee.
0:31:41 I don’t know if this is the sort of the outlandish edge of this, but maybe there’s something more mainstream to say about Chinese taste and the ability of Chinese companies to understand Chinese taste better than foreign companies.
0:31:43 Do you think there’s anything in that?
0:31:46 Well, firstly, I think that those are probably a bit more outlandish.
0:31:55 Distinctly remember, now that you’re talking, last year, during the summer, what was viral at the time was people putting Malta in coffee.
0:32:04 So Malta, for listeners who don’t know, is China’s sake, although many people who don’t like Malta would say it probably tastes like a poor man’s sake.
0:32:06 It’s a type of white spirit.
0:32:10 And so they were trying to revive the Malta brand and mix it with coffee.
0:32:19 And apparently it was really viral online, and especially Gen Z, young Chinese people were going out and drinking these Malta iced coffees.
0:32:26 But the reason I tell this story is that, I mean, the Chinese marketplace is so quick moving and evolving, and they love fads.
0:32:27 They love new trends.
0:32:33 And so, in a way, this sort of hyper-experimentation fits within the Chinese model.
0:32:44 But the other thing that I think that the Chinese model shows that I think will be more well-received, rather than these crazy pork fat lattes in the U.S. context, is the technology platform.
0:32:52 So in China, when you go to a Lakin or a Chaji or any of these chains, almost nobody orders at a till.
0:32:55 In fact, if you do, people kind of stare at you strangely, and you don’t pay with cash.
0:32:59 Almost everyone is paying in-app or with a QR code, so they pick up.
0:33:05 It’s a hyper-efficient business, and when you see the people move, it’s actually quite a sight to behold.
0:33:16 And I think bringing some of that technology and operation system to the U.S. context could be, I think, quite well-received, and just make the whole operation more efficient.
0:33:24 I’m not so sure about these crazy tastes, and already when I see the Luckin Coffee menu in New York, it’s a bit more tame in terms of what’s available.
0:33:35 Right now, they’re going on the Thanksgiving theme with these caramel popcorn ice lattes, these cinnamon pumpkin ice lattes, so more within the U.S. Thanksgiving theme.
0:33:45 But certainly, the way you were telling that story, James, about how coffee really did take China by storm, because it changed the culture, I think is an important lesson.
0:33:54 Because I don’t think anybody looking at Starbucks in the early 90s could have foreseen the fact that coffee really took over China.
0:33:59 And the flip side of this is that Starbucks has already reached peak China.
0:34:05 Its market share in China fell from 34% in 2019 to just 14% as of last year.
0:34:15 And this is, again, another sign, to your point, James, that as tastes have been westernized within China, they’ve gone back to finding Chinese alternatives.
0:34:22 And on future episodes, I’d love to explore that with different brands in luxury, in retail, in food.
0:34:27 There’s a story about gold here to tell as well, with Lowepul being Chinese Cartier.
0:34:35 I think that this is the ongoing trend, is that there’s going to be these Chinese alternatives that cater to change to palate over the last few decades.
0:34:43 But yeah, certainly, I’m excited to head over to New York in a week or two and try some of these New York-made Luckin coffees.
0:34:45 Let us know. Let us know what you think.
0:34:52 Yeah. And before we end, James, I have to ask you, what is your favorite or preferred coffee and choice?
0:34:56 Oh, goodness me, I’m incredibly boring. I’m just Americano with a dash of milk. That’s it.
0:35:00 Okay. Well, I’m going to try that when I go to New York and tell you.
0:35:09 I’m very much, I would say, I love a matcha latte right now, but I’m very partial to a flat white or a cortado, heavier on the coffee.
0:35:15 All right, James, this is my favorite time of the episode, prediction time.
0:35:19 What’s your prediction as you look at your crystal ball for the near future?
0:35:26 Okay. I’ve got something about humanoid robots. This is one of my favorite topics.
0:35:31 I mean, we have mentioned humanoid robots before, but I’ve got a specific prediction.
0:35:39 So Wang Xingxing, who is the head of Unitary, that’s the company that makes one of China’s most advanced humanoid robots,
0:35:49 said that humanoid robots in China will reach their, what he called, chat GPT moment within three years.
0:35:58 Now, he defined the chat GPT moment for robots as when a robot will be able to walk into a room that it’s never been in before,
0:36:04 find something like a glass of water that it’s been asked to find, pour the water into a glass,
0:36:08 and then bring it back to its host and give it to them.
0:36:11 So at the moment, this is not how humanoid robots work.
0:36:15 Humanoid robots are pre-programmed to do specific tasks.
0:36:21 What he’s talking about is where a humanoid robot can assess its surroundings, think what it needs to do,
0:36:24 find a solution, and then deliver on the solution.
0:36:27 He says that should happen within three years.
0:36:30 I reckon it’s going to happen sooner than that.
0:36:38 I reckon that by the end of 2027, we will see the chat GPT moment for humanoid robots,
0:36:40 and it will happen first in China.
0:36:41 That’s my prediction, Alice.
0:36:42 What about you?
0:36:47 To my Luddite ears, a horrifying eventuality.
0:36:49 I’m very scared of robots, I have to confess.
0:36:56 But as you were talking, I was thinking, gosh, the Chinese have definitely prioritized a different part of the AI stack,
0:37:01 whereas the Americans have gone in and poured a ton of money into LLMs.
0:37:06 I don’t think it’s a surprise then that you have Meta’s AI head,
0:37:10 Yan Lecun, basically leaving and going into world models,
0:37:15 so trying to be more in the space of embodied AI, humanoid robots, real world AI data.
0:37:21 So again, definitely watch this space, but I would agree with you that China is leading
0:37:25 when it comes to this humanoid AI robotic space.
0:37:30 So mine is also AI-focused, and I’m going to piggyback off of what we first talked about.
0:37:37 It honestly surprises me that the Trump administration hasn’t gone out yet and banned Chinese AI models.
0:37:39 I feel like that’s a pretty low-hanging fruit.
0:37:42 There was speculation earlier in the year that they could ban DeepSeek.
0:37:49 I think this is one of the policy areas in which you have an alignment between the Silicon Valley tech companies
0:37:56 that are competing potentially with Chinese LLMs, so Meta, Google, etc., OpenAI,
0:38:04 and then an alignment between the Silicon Valley tech entrepreneurs and the national security hawks in Washington,
0:38:13 who, to your point, James, are very worried about the security implications of companies and even officials using Chinese LLMs.
0:38:21 So I think in 2026, we may see more politicization around this issue that might ultimately push the Trump administration
0:38:27 to issue some kind of entities listing for these Chinese AI models, effectively banning them from being used.
0:38:33 Maybe starting first with ministry officials and then spreading them out.
0:38:35 All right, that’s all for this episode.
0:38:38 Thank you so much for listening to China Decode.
0:38:40 This is a production of Prof G Media.
0:38:43 Our producer is David Toledo.
0:38:46 Our associate producer is Eric Janikas.
0:38:49 Our video editor is Ness Smith-Savidoff.
0:38:52 Our research associate is Dan Shallon.
0:38:55 Our technical director is Drew Burrows.
0:38:57 Our engineer is William Flynn.
0:39:01 And our executive producer is Catherine Dillon.
0:39:08 Make sure to follow us wherever you get your podcasts so you don’t miss an episode and talk to you again next week.
0:39:22 Support for this show comes from Odoo.
0:39:28 Running a business is hard enough, so why make it harder with a dozen different apps that don’t talk to each other?
0:39:30 Introducing Odoo.
0:39:33 It’s the only business software you’ll ever need.
0:39:37 It’s an all-in-one, fully integrated platform that makes your work easier.
0:39:41 CRM, accounting, inventory, e-commerce, and more.
0:39:42 And the best part?
0:39:46 Odoo replaces multiple expensive platforms for a fraction of the cost.
0:39:50 That’s why over thousands of businesses have made the switch.
0:39:51 So why not you?
0:39:54 Try Odoo for free at Odoo.com.
0:39:57 That’s O-D-O-O dot com.
0:40:02 Support for this show comes from Volkswagen.
0:40:13 As the U.S. gets ready to host soccer’s biggest moment on a worldwide stage, Volkswagen is helping people discover new turfs and new ways to play the beautiful game right here in the U.S.
0:40:22 From deaf and power wheelchair soccer, to beach and futsal, Volkswagen is actively supporting all the communities and teams within the U.S. soccer ecosystem.
0:40:32 They’re supporting talent from across the U.S. soccer extended national teams and are focused on helping to give these less widely known forms of soccer a platform moving forward.
0:40:38 From the pitch to the sand and everything in between, welcome to our turf.
0:40:38 you
0:00:12 For more than 150 years, Panerai watches have showcased the brand’s rich heritage of Italian design and Swiss horological expertise, making it the perfect timepiece for modern heroes.
0:00:15 So this is my weakness.
0:00:17 I have several Panerai watches.
0:00:21 It’s the only watch I wear, and trust me, Panerai is the way to go.
0:00:30 Panerai’s new Luminor Marina timepieces honor the brand’s storied legacy and history of pioneering innovation while enhancing iconic aesthetics, reliability, and functionality.
0:00:35 Visit Panerai.com to discover the world of Panerai and find the boutique nearest you.
0:00:37 That’s P-A-N-E-R-A-I.com.
0:00:51 Support for this show comes from the Audible original, the downloaded two, ghosts in the machine, quantum computers.
0:00:57 The next great frontier of technology, offering endless possibilities that stretch the human mind.
0:01:05 But for Roscoe Cadoulian and the Phoenix Colony, quantum computing uploads the human mind with life-altering consequences.
0:01:17 Audible’s hit sci-fi thriller, The Downloaded, returns with Oscar winner Brendan Fraser, reprising his role as Roscoe Cadoulian in The Downloaded 2, Ghosts in the Machine.
0:01:20 This thought-provoking sequel from Robert J.
0:01:24 Takes listeners on a captivating sci-fi journey.
0:01:30 A mind-bending must-listen that asks, what are you willing to lose to save the ones you love?
0:01:33 The Downloaded 2, Ghosts in the Machine.
0:01:36 Available now, only from Audible.
0:01:49 Support for this show comes from the Audible original, The Downloaded 2, Ghosts in the Machine.
0:01:52 The Earth only has a few days left.
0:01:58 Roscoe Cadoulian and the rest of the Phoenix Colony have to re-upload their minds into the quantum computer.
0:02:03 But a new threat has arisen that could destroy their stored consciousness forever.
0:02:12 Listen to Oscar winner Brendan Fraser reprise his role as Roscoe Cadoulian in this follow-up to the Audible original blockbuster, The Downloaded.
0:02:18 It’s a thought-provoking sci-fi journey where identity, memory, and morality collide.
0:02:23 Robert J. Sawyer does it again with this much-anticipated sequel that leaves you asking,
0:02:26 what are you willing to lose to save the ones you love?
0:02:30 The Downloaded 2, Ghosts in the Machine.
0:02:32 Available now, only from Audible.
0:02:44 The latest Chinese AI sensation, which is called Kimi K2,
0:02:50 cost just a fraction of what it would cost to train an AI model in America.
0:02:54 It cost about 4.6 million US dollars.
0:03:01 That is just a tiny fraction of what OpenAI has been investing in its large language models.
0:03:05 Welcome to China Decode. I’m Alice Han.
0:03:06 And I’m James King.
0:03:12 Well, in today’s episode of China Decode, we are discussing whether China is building the Android of AI,
0:03:18 while the US builds the ultra-premium iPhone, powerful but too expensive to dominate globally.
0:03:25 Why the latest spat between Japan and China could reshape the balance of power in the Asia-Pacific region.
0:03:30 And how Starbucks and the coffee wars reveal the fierce fight for China’s consumer market.
0:03:32 All right, let’s get right into it.
0:03:37 Right now, the biggest story in tech may be the one the US didn’t see coming.
0:03:43 China’s low-cost, open-source AI is suddenly everywhere in Silicon Valley.
0:03:48 While the US is spending billions of dollars building ever-bigger models and data centers,
0:03:55 and inflating what some analysts say looks like an AI bubble, China is doing the exact opposite, it seems.
0:04:02 Cutting prices, opting for open-sourcing weights, and shipping models that are cheaper, lighter, and surprisingly competitive.
0:04:05 And American companies are flocking to them.
0:04:13 Airbnb’s CEO says his team ditched ChatGPT for Alibaba’s Quen model because it’s, quote-unquote, fast and cheap.
0:04:17 Venture firms are switching to Moonshot’s Kimi, LLM,
0:04:23 and developer data show nearly half of the most used models in the US last week were Chinese.
0:04:30 Alibaba just slashed prices on its flagship model again as part of a full-blown AI price war,
0:04:33 and a sign that involution has come to Chinese AI.
0:04:41 Former Google CEO Eric Schmidt is warning that most countries may end up using Chinese models simply because they’re free.
0:04:46 James, we’ve talked a lot about, I think, Chinese AI over the last few weeks,
0:04:51 and I definitely think, I’m very excited to see the models that are coming out of the mainland,
0:04:56 but also the fact that the West is catching up to this theme that we’ve been talking about for a while,
0:04:59 which is don’t write off Chinese AI.
0:05:01 They are going to go for it in a very different model.
0:05:05 And I think the recent announcements from these tech companies,
0:05:12 some even, I would say, pretty big startups that are using and relying Chinese LLMs is pretty dramatic,
0:05:18 and I think it’s a different narrative than what we’ve been seeing in the last few years.
0:05:22 I just want to go right into it and see what your take is on this Chinese model of AI,
0:05:29 and whether or not this kind of Android AI approach to making cheaper, faster, more efficient models
0:05:33 is ultimately going to win out in a global race between China and the US of AI.
0:05:35 Yeah, that’s a really great question, Alice.
0:05:41 I’ve started to notice just in recent days, actually, that people are using the word bubble
0:05:46 with regard to America’s AI efforts quite a bit more now.
0:05:49 I think in the past they’ve been reluctant to do that.
0:05:54 My own take is that this is still very far from conventional wisdom,
0:05:59 and the US-China AI race is going to be a very long-run thing.
0:06:01 We can’t call it yet.
0:06:04 There’s going to be many ups and downs and fits and starts.
0:06:12 But as things stand right now, I would say that some big names in the US are looking quite bubbly.
0:06:14 And I think the key point is this.
0:06:21 The bubbly aspects of American AI, as I see it, are made in China,
0:06:31 by which I mean that China’s approach to AI is to build AI models at a much lower cost than we can find in the US.
0:06:36 Also, the cost of using those Chinese AI models is much lower.
0:06:46 And crucially, the performance of the Chinese AI models is kind of just as good or catching up to US standards at the moment.
0:06:53 So I think that that is the reason why we saw Eric Schmidt, the former CEO of Google,
0:07:00 very recently say, as you’ve already mentioned, you know, the vast majority of governments and countries
0:07:08 who don’t have the kind of money that the West has will end up standardizing by using Chinese models,
0:07:10 not because they’re better, but because they’re free.
0:07:18 So I think really we’re entering a very interesting phase in the US-China AI race.
0:07:22 I should just mention before I pass it back to you, Alice, that, you know,
0:07:24 there needs to be a little bit of nuance on this.
0:07:34 Not all US LLMs are closed source, and not all Chinese large language models are open source.
0:07:36 So it’s a mixed picture.
0:07:43 You know, we’ve got Lama from Meta, we’ve got Fi from Microsoft and Gemma from Google.
0:07:46 All of those are open source.
0:07:50 And of course, it’s famous that many of the big and successful Chinese models,
0:07:55 such as DeepSeq and Q1, as you already mentioned, are open source too.
0:08:02 But I just want to, I just want to make that clear where, you know, it’s, it’s a variegated picture.
0:08:07 It’s not the case that everything done in China is open source and everything done in the US is closed source.
0:08:12 I just want to leave you with one extraordinary statistic.
0:08:24 And that is that the latest Chinese AI sensation, which is called Kimi K2, cost just a fraction of what it would cost to train an AI model in America.
0:08:29 It cost about 4.6 million US dollars.
0:08:37 That is just a tiny fraction of what open AI has been investing in, in its large language models over in the US.
0:08:39 But how do you see this, Alice?
0:08:42 Do you, do you think that, you know, it’s too early to call?
0:08:47 Or do you think that the American model is looking bubbly and the Chinese model is looking lean and mean?
0:08:50 I think that’s a great way to phrase it.
0:08:52 Bubbly versus lean and mean.
0:09:10 I think I’m more in the camp of that kind of narrative, although with the caveat that, as we’ve previously talked about, James, the US has thrown billions of CapEx into the data center rollout, into leading edge LLMs, into the frontier labs and hyperscalers.
0:09:19 And as a result, I think is probably going to be further along in the AGI, artificial general intelligence race or super intelligence, as we previously mentioned.
0:09:29 But the Chinese model, to your point, James, is leaner and meaner in the sense that, you know, it is doing these models more efficiently, much cheaper to train.
0:09:39 What was interesting for me to notice, and you already mentioned it, James, is that not only are they cheaper on the inference costs, but also on the training costs.
0:09:49 I remember meeting, I believe, the father of AI in China, who wrote that famous book, AI Superpowers, which seems like a lifetime ago, but I highly recommend that read, I believe, in 2018.
0:09:52 So before this really ramped up as a serious issue.
0:10:03 And he had said at the time a year and a half ago that China may not be able to compete on training costs because the US has dedicated a huge amount of CapEx towards training.
0:10:10 And it has some of the best data sets and training models in the world, but China can rapidly decrease the inference costs.
0:10:13 Now it seems like they’ve decreased both the inference and training costs.
0:10:24 To your point, the Kimi K2 open source model is just $4.6 million, as you say, that’s 1,500 times less than what open AI spent on R&D this year.
0:10:29 And again, just to follow up on your point, James, so that listeners can appreciate this.
0:10:33 I think it’s a bit of a misnomer to say that China is leading the way on open source.
0:10:48 I’ve been talking to some AI scientists more recently who’ve said that actually the majority of the quote-unquote open source models are really open weight, meaning that the underlying code and data set are not freely available, but the weights or the training parameters are.
0:10:51 That’s the difference between an open weight versus an open source.
0:10:59 But the fact of the matter remains that China is going through what I see as a Cambrian explosion in terms of different LLMs.
0:11:08 And we’ve talked about this theme of involution, of price wars, of excess capacity, of hyper-competition.
0:11:16 I think this is already starting to be played out in the Chinese context, where you have a lot of different LLMs in the race or in the ring.
0:11:23 Kimi just overtook of QN and DeepSeek very recently in terms of its parameters and performance.
0:11:35 I think the fact that China is being used, its LLMs are being used by American companies, is a pretty telling and a compelling sign that consumers are driven by price.
0:11:44 And the fact that these LLMs can be cheaper than their American counterparts, I think, is making them more desirable, even amongst some of these Western companies.
0:11:57 Which is why, you know, I find that the recent talk about Chinese LLMs being competitive is coming at a time in which a lot of people watching Wall Street are very worried about the AI bubble in the US.
0:12:00 I think the two of them coexisting as narratives is not a coincidence.
0:12:05 One question I do have for you, James, is have you actually used the different models?
0:12:13 I’ve been playing around a little bit across the different models from the US and China, but I want to get a sense of what you’ve experienced.
0:12:15 Yeah, I have actually.
0:12:16 I’ve used DeepSeek quite a lot.
0:12:18 I continue to use DeepSeek quite a lot.
0:12:24 Obviously, I’ve used ChatGPT quite a bit and I’ve just started to play around with Kimi as well.
0:12:31 You know, to be honest with you, I’m using it at such a low level that I don’t really notice that much of a difference.
0:12:35 I’m sort of using it as a dictionary, really, and as a database.
0:12:38 So, yeah, I mean, they’re all pretty good.
0:12:49 I have noticed that the DeepSeek results are a little bit skewed away from politically sensitive topics that, you know, you might expect from a Chinese LLM.
0:12:53 But I haven’t noticed anything too egregious, I must say.
0:12:58 But there is this question about security, which is hanging in the background.
0:13:05 And I think that a lot of people are concerned by the question of security with regard to Chinese large language models.
0:13:18 I should cite an article by the Financial Times just recently that says that Alibaba provides tech support for the Chinese military against targets in the United States.
0:13:20 Now, I need to be very careful about this.
0:13:29 The claims, which couldn’t be independently verified by the FT, were rejected very forcefully by Alibaba.
0:13:35 They said that this information leaked from your source is complete nonsense.
0:13:37 So that’s a very strong denial.
0:13:45 But there is always going to be this issue of security hanging over the Chinese large language models, I think, to users in the West, at least.
0:13:47 Yeah, I completely agree.
0:13:49 And I think that we should watch the space.
0:13:55 I think it’s a matter of time before these LLMs get politicized in the same way that semiconductors have been.
0:13:58 I actually think that that is a likely outcome.
0:14:03 The more that we talk about Kimmy and DeepSeek and the more that we worry about an AI bubble in the U.S.,
0:14:08 I think the more political backlash there will be against some of these Chinese LLMs.
0:14:14 But just to round up this discussion, I thought it was interesting as I looked into Kimmy’s architecture.
0:14:21 One of the reasons that it is more efficient than the American LLMs is that it uses this thing known as the mixture of experts design,
0:14:25 whereas ChatGPT and some of these other LLMs are using a dense model.
0:14:33 This mixture of experts design only utilizes certain parameters when created, and a dense model uses them all.
0:14:39 And apparently, the Kimmy K2 is over a trillion parameters, but it will only use about 32 billion at a time.
0:14:50 And this points to, I think, one last note that I’ll end on, is it points to the example in which Chinese AI is, at a research level, evolving in a very, very different way.
0:14:55 And so the more that I think about this space, and I’m trying to read up very quickly, because it’s changing so quickly,
0:15:04 I see that sort of whole evolution of two very, very different species of AI that I think is happening right now between China and the U.S.
0:15:07 Okay, we’ll be back with more after a quick break, so stay with us.
0:15:15 Support for this show comes from Odoo.
0:15:21 Running a business is hard enough, so why make it harder with a dozen different apps that don’t talk to each other?
0:15:23 Introducing Odoo.
0:15:26 It’s the only business software you’ll ever need.
0:15:30 It’s an all-in-one, fully integrated platform that makes your work easier.
0:15:34 PRM, accounting, inventory, e-commerce, and more.
0:15:35 And the best part?
0:15:40 Odoo replaces multiple expensive platforms for a fraction of the cost.
0:15:43 That’s why over thousands of businesses have made the switch.
0:15:44 So why not you?
0:15:47 Try Odoo for free at odoo.com.
0:15:50 That’s O-D-O-O dot com.
0:15:56 Saks Fifth Avenue makes it easy to holiday your way.
0:16:02 Whether it’s finding the right gift or the right outfit, Saks is where you can find everything.
0:16:08 From the perfect Chloe bag for your hard-to-shop-for sister to a Prada jacket for a fancy holiday dinner.
0:16:15 And if you don’t know where to start, Saks.com is customized to your personal style, so you can save time shopping.
0:16:20 Make shopping fun and easy this season, and find gifts and inspiration to suit your holiday style.
0:16:24 Well, welcome back.
0:16:27 Tensions are well and truly boiling in East Asia.
0:16:36 China and Japan are locked in a fiery diplomatic clash over Taiwan, with threats, insults, and even references to beheading now making the headlines.
0:16:45 And it all started when Japan’s new Prime Minister, Senai Takeichi, said a Chinese attack on Taiwan could trigger a Japanese military response.
0:16:49 This is a rare explicit stance from a sitting leader.
0:17:03 And Beijing responded swiftly, summoning ambassadors, issuing warnings of a crushing defeat if Japan intervenes, and lashing out in state media, calling Takeichi a quote-unquote evil witch, and warning Chinese citizens to avoid travel to Japan.
0:17:06 The spat highlights more than just the issue of Taiwan.
0:17:13 It also involves historical grievances, territorial disputes, and Japan’s shifting security posture.
0:17:20 Japan, America’s key ally in the region, may increasingly act as a deterrent while the U.S. pulls back.
0:17:27 And Takeichi’s hard line could boost her actual standing at home and abroad, even with conservative American allies.
0:17:33 James, I think that this is giving me 2010 deja vu all over again.
0:17:39 When we had the Senkaku Island incident, we had the Rare Earths embargo for a couple months in Japan.
0:17:47 We had a lot of Chinese, everyday Chinese citizens boycotting Japanese goods, burning Japanese goods, looting Japanese stores.
0:17:53 And I’m not surprised, frankly, that China has reacted in a pretty aggressive manner.
0:18:00 And I wouldn’t be surprised if everyday Chinese boycott Japan entirely, as they’ve done in the past.
0:18:05 Anytime there has been an anti-Japan or hyper-nationalist movement.
0:18:19 It’s no surprise that Chinese airlines announced over the weekend that they would allow all passengers to cancel or change Japan-bound flights free of charge, after Beijing basically urged the citizens to avoid visiting Japan.
0:18:28 This comes at a time, I think, where Trump is clearly trying to pull back from U.S. commitments in the region.
0:18:37 Japan, I think, is sitting there pretty worried about the state of affairs, given that it relies so heavily on the deterrence that the U.S. offers.
0:18:41 And that Japan is not a nuclear power, which we’ll get into in just a bit.
0:18:48 I certainly would be nervous sitting in this region at a time where politically both sides want to play hardball.
0:18:52 But what’s your take on this, James, having lived in the region for quite a bit of time?
0:18:59 Yeah, I mean, I’ve lived in Japan and China and Taiwan and, you know, the countries that are in play.
0:19:02 It’s amazing to me how quickly this has escalated.
0:19:08 It’s gone from zero to 60 or zero to 120 in what feels like a very short period of time.
0:19:21 You know, I woke up on Sunday morning to see, on my social media feeds, three Chinese warships steaming towards the Senkaku Islands, which are territory disputed by Japan and China.
0:19:23 It really was quite dramatic.
0:19:34 And I think it’s important for the West as well, particularly for the United States, because, of course, Japan is America’s leading ally in Asia.
0:19:41 Japan also plays host to several U.S. military bases, and these are really major.
0:19:49 The Yokosuka Naval Base, just near Tokyo Bay, is the home port for the U.S. 7th Fleet.
0:19:52 And there’s a large number of U.S. ships there.
0:19:58 And it also has the nuclear-powered aircraft carrier called the USS George Washington.
0:20:02 This is one of the, you know, the premier ships in the U.S. Navy.
0:20:13 So, any kind of a flare-up between Japan and China, especially if it has a military dimension, which this one very clearly does, is a big issue for Washington.
0:20:20 And I imagine it’s quickening pulses all over the U.S., just as it is for you and I.
0:20:25 Of course, the Japanese side is now trying to de-escalate.
0:20:35 They have sent over Masaaki Kanai, who is head of the Foreign Ministry’s Asian and Oceanian Affairs Bureau.
0:20:40 They sent him to Beijing to see if he can sort of try and patch things up.
0:20:47 But, in the meantime, as you’ve already alluded to, this is hammering some Japanese stocks.
0:20:53 You know, we’ve seen the Isetan and Mitsukoshi department store owner.
0:20:58 That stock price has fallen over 10%.
0:21:04 And also, the Tokyo Disneyland operator, that’s lost nearly 6%.
0:21:05 Japan Airlines is down.
0:21:11 This is all because these companies benefit hugely from Chinese tourists going to Japan.
0:21:19 And as you’ve said, Alice, once nationalist fervor grips the Chinese people, as it is clearly starting to do,
0:21:26 then you immediately see a drop-off in the willingness of Chinese to buy Japanese stuff and to travel to Japan.
0:21:28 How do you see this?
0:21:30 I mean, there is obviously historical context here.
0:21:38 I mean, why is it that Chinese feel so triggered, as it were, by Japan at these moments?
0:21:43 Well, I mean, China had a harrowing experience in World War II at the hands of Japan.
0:21:49 Japan occupied parts of China, even getting up to Shanghai, where my family is originally from.
0:21:57 And the rape of Nanjing or even the systemic wiping out of populations in China are deeply, deeply remembered
0:21:59 within the Chinese collective.
0:22:08 And I think that that probably intensifies a lot of already, I think, strategic rivalries and issues at play.
0:22:11 So there’s a deep set historical mistrust.
0:22:17 But at the same time, I think when we look at it from a real politique standpoint,
0:22:24 Japan is deeply worried about an ascendant China and what that means for the Asia-Pacific region.
0:22:36 If the U.S. decides that it ultimately doesn’t want to be a regional and global hegemon, Japan is in a very, very, I think, isolated state because it has no nuclear deterrence of its own.
0:22:47 It cannot, according to the treaty that it signed after World War II with the Americans, it cannot have an offensive military capable of warfare.
0:22:55 And it has the three non-nuclear principles, meaning that it cannot possess, produce or allow the introduction of nuclear weapons onto its territory.
0:23:00 So I think Japan, from a real politique, realist standpoint, is in a very desperate situation.
0:23:07 And I’m not surprised that Takeichi is trying to draw into question the three non-nuclear principles.
0:23:12 That is another reason why China is worried, because, and this is according to people close to her,
0:23:19 she apparently wants to call to question the strict adherence of the no introduction of nuclear weapons principle,
0:23:25 because ultimately she wants to allow U.S. nuclear-armed vessels to sit in Japanese ports
0:23:32 to basically have a credible nuclear deterrence against China in a potential showdown over Taiwan or in the region.
0:23:41 So this is, I think, all driven by the tectonic shifts of power, mainly in terms of the U.S.-China relationship in the region.
0:23:47 And I think, ultimately, I wouldn’t be surprised, even if we do see a de-escalation in the short term,
0:23:54 that there is an increasing rivalry between Japan and a mistrust between Japan and China,
0:24:00 and that Japan gets closer with, say, Korea, India, Australia,
0:24:06 some of these other countries that could be on the opposing side of China if there is a showdown in the region.
0:24:10 But just very quickly, James, having lived in Japan,
0:24:15 do you think that there has been a shift in Japan’s attitude towards China in the last few years?
0:24:22 Well, I mean, I think that certainly Prime Minister Takeichi comes from the wing of the Liberal Democratic Party,
0:24:27 the LDP, which is known to be more hawkish towards China.
0:24:32 She’s seen as a little bit of a successor to Shinzo Abe, the previous prime minister,
0:24:36 who was, you know, considered to be fairly hawkish towards China.
0:24:40 And I think that is already playing out in the minds of the Chinese
0:24:48 and making China extra anxious about what the next few years of Takeichi’s tenure is going to produce.
0:24:57 So I think that this flare-up that we’ve seen at the moment could set the tone for however long Takeichi is prime minister,
0:25:01 even if there are attempts by the Japanese, as I’m sure there will be, to de-escalate.
0:25:07 Like you, I see this as a, well, maybe not a continuous source of tension,
0:25:12 but, you know, a general undertow of tension between the two countries.
0:25:14 Well, James, we’re going to take a quick coffee break.
0:25:17 So stay with us as we wait for the next segment.
0:25:29 Support for this show comes from the Audible Original, The Downloaded 2, Ghosts and the Machine.
0:25:32 The Earth only has a few days left.
0:25:38 Roscoe Cadulian and the rest of the Phoenix Colony have to re-upload their minds into the quantum computer,
0:25:43 but a new threat has arisen that could destroy their stored consciousness forever.
0:25:48 Listen to Oscar winner Brendan Fraser reprise his role as Roscoe Cadulian
0:25:52 in this follow-up to the Audible Original blockbuster, The Downloaded.
0:25:58 It’s a thought-provoking sci-fi journey where identity, memory, and morality collide.
0:26:03 Robert J. Sawyer does it again with this much-anticipated sequel that leaves you asking,
0:26:06 what are you willing to lose to save the ones you love?
0:26:10 The Downloaded 2, Ghosts and the Machine.
0:26:12 Available now, only from Audible.
0:26:21 Well, welcome back.
0:26:26 This is my favorite section of this podcast, not just because I am a caffeine addict,
0:26:29 but also I find this story quite astonishing.
0:26:34 China’s coffee wars are growing up and Starbucks is taking cover, it seems.
0:26:37 After nearly 30 years at 8,000 stores,
0:26:42 America’s Starbucks is selling a majority stake in its Chinese business to bore you your capital,
0:26:44 keeping just 40%.
0:26:49 Local rivals like Luckin and Koti are surging ahead, opening thousands of outlets,
0:26:53 cutting prices, and serving drinks, very bizarre drinks, I would say,
0:26:58 tailored to local tastes, from pork drizzled lattes to alcohol-infused specials.
0:27:04 Starbucks’ premium strategy and slow product rollout just couldn’t keep up in the end.
0:27:09 This joint venture gives Starbucks local insight, faster expansion, and digital integration,
0:27:14 critical in a market that remains dominated by WeChat Pay and Alipay.
0:27:18 Now, other U.S. brands like Burger King and Decathlon are following suit,
0:27:24 selling their steaks to local partners to survive amid China’s hyper-competitive consumer landscape.
0:27:30 Now, the big question is, is China still a land of opportunity or have homegrown brands
0:27:32 completely rewritten the rules?
0:27:37 James, I love this story for many, many reasons, as already alluded to.
0:27:44 But also, I had noticed from a friend who sent me a message earlier in the year that they had picked up a coffee,
0:27:46 a pineapple cold brew from Luckin’s New York store.
0:27:52 And I realized then that Luckin, a coffee brand that I’d long associated just with China,
0:27:55 has become very global in the last few years.
0:28:00 And it was funny for me to see this picture because I still remember during COVID went bankrupt.
0:28:05 And prior to that, I had cases of financial fraud.
0:28:10 So I was surprised to see that they not only have become profitable again,
0:28:16 but they are trying to relist in the NASDAQ and now have at least two stores that I know of in New York,
0:28:21 where apparently the regular priced ice latte costs as little as $5.75.
0:28:26 Now, I’ve compared that to some of my colleagues living in New York who say that at Starbucks
0:28:32 and other places you could get an ice latte for $6 to $8 US dollars, but listeners should definitely correct us.
0:28:39 I do know for a fact that people who download the app as first-time users can have a coffee for as little as $1.99.
0:28:46 So this is a strategy that they’ve definitely kept alive that has been successful in the Chinese context.
0:28:52 But I am a little bit skeptical that they can do a reverse Starbucks and be as successful in America
0:28:55 because I do think the Chinese market is quite different
0:29:00 and that they’d have to burn a lot of cash to really survive in America.
0:29:05 But it’s interesting to see these hungry executives in Luckin,
0:29:08 but also another tea company that I really love in China called Chaji.
0:29:11 They do specialty iced teas.
0:29:16 These executives are really trying to jump over the Pacific and be as successful in America as they were in China.
0:29:20 I’m a little skeptical, but James, please refute me if you disagree.
0:29:22 I love that.
0:29:23 I mean, I hadn’t heard that story.
0:29:27 Luckin coffee going to America and selling coffee for $1.99 a cup.
0:29:31 Yeah, I mean, that’s really quite something.
0:29:33 It’s going to help Nandani make New York more affordable.
0:29:35 That’s right.
0:29:36 Well, it certainly will.
0:29:41 I mean, I certainly wouldn’t have any qualms about, you know, buying a coffee for $1.99 if it tasted OK.
0:29:49 But, you know, this story has quite a lot of resonances for me, too, because I was in Beijing when Starbucks arrived in 1999.
0:29:55 I reported on the first Starbucks shop, which was in the World Trade Center.
0:30:08 And, you know, at that time, we all, I certainly wasn’t the only one, or the journalists there were all writing about how Starbucks had the chutzpah to try to sell coffee to a country with thousands of years of tea culture.
0:30:12 There was a lot of quotes from people saying, this will never catch on.
0:30:15 And then look how wrong we were.
0:30:18 You know, there’s literally thousands and thousands of stores.
0:30:20 So that was one thing.
0:30:31 But the other really fascinating aspect of this story to me, which I hope you’ll be able to help me with, because I think you’re much more of an expert than I am, is what it says about Chinese tastes.
0:30:41 You already mentioned that the halo effect of foreign brands, not just Starbucks, but plenty of others in China, is beginning to slip.
0:30:45 We’ve seen Apple giving ground to local competitors.
0:30:48 We’ve seen Estee Lauder giving ground to local competitors.
0:30:52 LVMH in the luxury sector doing the same thing.
0:31:04 So the other aspect of it is the localization, which maybe includes a better understanding by Chinese companies of what Chinese really want to drink or to eat.
0:31:09 And I don’t know, maybe this is a bit outlandish, Alice, please tell me.
0:31:19 But I found something online about a coffee shop that opened in southwestern China that had cooked pork intestines, like on a skewer on top of the coffee.
0:31:28 And then there was also – actually, this was Starbucks earlier on, I think it was the end of last year – putting like pork fat drizzle onto their coffee.
0:31:41 I don’t know if this is the sort of the outlandish edge of this, but maybe there’s something more mainstream to say about Chinese taste and the ability of Chinese companies to understand Chinese taste better than foreign companies.
0:31:43 Do you think there’s anything in that?
0:31:46 Well, firstly, I think that those are probably a bit more outlandish.
0:31:55 Distinctly remember, now that you’re talking, last year, during the summer, what was viral at the time was people putting Malta in coffee.
0:32:04 So Malta, for listeners who don’t know, is China’s sake, although many people who don’t like Malta would say it probably tastes like a poor man’s sake.
0:32:06 It’s a type of white spirit.
0:32:10 And so they were trying to revive the Malta brand and mix it with coffee.
0:32:19 And apparently it was really viral online, and especially Gen Z, young Chinese people were going out and drinking these Malta iced coffees.
0:32:26 But the reason I tell this story is that, I mean, the Chinese marketplace is so quick moving and evolving, and they love fads.
0:32:27 They love new trends.
0:32:33 And so, in a way, this sort of hyper-experimentation fits within the Chinese model.
0:32:44 But the other thing that I think that the Chinese model shows that I think will be more well-received, rather than these crazy pork fat lattes in the U.S. context, is the technology platform.
0:32:52 So in China, when you go to a Lakin or a Chaji or any of these chains, almost nobody orders at a till.
0:32:55 In fact, if you do, people kind of stare at you strangely, and you don’t pay with cash.
0:32:59 Almost everyone is paying in-app or with a QR code, so they pick up.
0:33:05 It’s a hyper-efficient business, and when you see the people move, it’s actually quite a sight to behold.
0:33:16 And I think bringing some of that technology and operation system to the U.S. context could be, I think, quite well-received, and just make the whole operation more efficient.
0:33:24 I’m not so sure about these crazy tastes, and already when I see the Luckin Coffee menu in New York, it’s a bit more tame in terms of what’s available.
0:33:35 Right now, they’re going on the Thanksgiving theme with these caramel popcorn ice lattes, these cinnamon pumpkin ice lattes, so more within the U.S. Thanksgiving theme.
0:33:45 But certainly, the way you were telling that story, James, about how coffee really did take China by storm, because it changed the culture, I think is an important lesson.
0:33:54 Because I don’t think anybody looking at Starbucks in the early 90s could have foreseen the fact that coffee really took over China.
0:33:59 And the flip side of this is that Starbucks has already reached peak China.
0:34:05 Its market share in China fell from 34% in 2019 to just 14% as of last year.
0:34:15 And this is, again, another sign, to your point, James, that as tastes have been westernized within China, they’ve gone back to finding Chinese alternatives.
0:34:22 And on future episodes, I’d love to explore that with different brands in luxury, in retail, in food.
0:34:27 There’s a story about gold here to tell as well, with Lowepul being Chinese Cartier.
0:34:35 I think that this is the ongoing trend, is that there’s going to be these Chinese alternatives that cater to change to palate over the last few decades.
0:34:43 But yeah, certainly, I’m excited to head over to New York in a week or two and try some of these New York-made Luckin coffees.
0:34:45 Let us know. Let us know what you think.
0:34:52 Yeah. And before we end, James, I have to ask you, what is your favorite or preferred coffee and choice?
0:34:56 Oh, goodness me, I’m incredibly boring. I’m just Americano with a dash of milk. That’s it.
0:35:00 Okay. Well, I’m going to try that when I go to New York and tell you.
0:35:09 I’m very much, I would say, I love a matcha latte right now, but I’m very partial to a flat white or a cortado, heavier on the coffee.
0:35:15 All right, James, this is my favorite time of the episode, prediction time.
0:35:19 What’s your prediction as you look at your crystal ball for the near future?
0:35:26 Okay. I’ve got something about humanoid robots. This is one of my favorite topics.
0:35:31 I mean, we have mentioned humanoid robots before, but I’ve got a specific prediction.
0:35:39 So Wang Xingxing, who is the head of Unitary, that’s the company that makes one of China’s most advanced humanoid robots,
0:35:49 said that humanoid robots in China will reach their, what he called, chat GPT moment within three years.
0:35:58 Now, he defined the chat GPT moment for robots as when a robot will be able to walk into a room that it’s never been in before,
0:36:04 find something like a glass of water that it’s been asked to find, pour the water into a glass,
0:36:08 and then bring it back to its host and give it to them.
0:36:11 So at the moment, this is not how humanoid robots work.
0:36:15 Humanoid robots are pre-programmed to do specific tasks.
0:36:21 What he’s talking about is where a humanoid robot can assess its surroundings, think what it needs to do,
0:36:24 find a solution, and then deliver on the solution.
0:36:27 He says that should happen within three years.
0:36:30 I reckon it’s going to happen sooner than that.
0:36:38 I reckon that by the end of 2027, we will see the chat GPT moment for humanoid robots,
0:36:40 and it will happen first in China.
0:36:41 That’s my prediction, Alice.
0:36:42 What about you?
0:36:47 To my Luddite ears, a horrifying eventuality.
0:36:49 I’m very scared of robots, I have to confess.
0:36:56 But as you were talking, I was thinking, gosh, the Chinese have definitely prioritized a different part of the AI stack,
0:37:01 whereas the Americans have gone in and poured a ton of money into LLMs.
0:37:06 I don’t think it’s a surprise then that you have Meta’s AI head,
0:37:10 Yan Lecun, basically leaving and going into world models,
0:37:15 so trying to be more in the space of embodied AI, humanoid robots, real world AI data.
0:37:21 So again, definitely watch this space, but I would agree with you that China is leading
0:37:25 when it comes to this humanoid AI robotic space.
0:37:30 So mine is also AI-focused, and I’m going to piggyback off of what we first talked about.
0:37:37 It honestly surprises me that the Trump administration hasn’t gone out yet and banned Chinese AI models.
0:37:39 I feel like that’s a pretty low-hanging fruit.
0:37:42 There was speculation earlier in the year that they could ban DeepSeek.
0:37:49 I think this is one of the policy areas in which you have an alignment between the Silicon Valley tech companies
0:37:56 that are competing potentially with Chinese LLMs, so Meta, Google, etc., OpenAI,
0:38:04 and then an alignment between the Silicon Valley tech entrepreneurs and the national security hawks in Washington,
0:38:13 who, to your point, James, are very worried about the security implications of companies and even officials using Chinese LLMs.
0:38:21 So I think in 2026, we may see more politicization around this issue that might ultimately push the Trump administration
0:38:27 to issue some kind of entities listing for these Chinese AI models, effectively banning them from being used.
0:38:33 Maybe starting first with ministry officials and then spreading them out.
0:38:35 All right, that’s all for this episode.
0:38:38 Thank you so much for listening to China Decode.
0:38:40 This is a production of Prof G Media.
0:38:43 Our producer is David Toledo.
0:38:46 Our associate producer is Eric Janikas.
0:38:49 Our video editor is Ness Smith-Savidoff.
0:38:52 Our research associate is Dan Shallon.
0:38:55 Our technical director is Drew Burrows.
0:38:57 Our engineer is William Flynn.
0:39:01 And our executive producer is Catherine Dillon.
0:39:08 Make sure to follow us wherever you get your podcasts so you don’t miss an episode and talk to you again next week.
0:39:22 Support for this show comes from Odoo.
0:39:28 Running a business is hard enough, so why make it harder with a dozen different apps that don’t talk to each other?
0:39:30 Introducing Odoo.
0:39:33 It’s the only business software you’ll ever need.
0:39:37 It’s an all-in-one, fully integrated platform that makes your work easier.
0:39:41 CRM, accounting, inventory, e-commerce, and more.
0:39:42 And the best part?
0:39:46 Odoo replaces multiple expensive platforms for a fraction of the cost.
0:39:50 That’s why over thousands of businesses have made the switch.
0:39:51 So why not you?
0:39:54 Try Odoo for free at Odoo.com.
0:39:57 That’s O-D-O-O dot com.
0:40:02 Support for this show comes from Volkswagen.
0:40:13 As the U.S. gets ready to host soccer’s biggest moment on a worldwide stage, Volkswagen is helping people discover new turfs and new ways to play the beautiful game right here in the U.S.
0:40:22 From deaf and power wheelchair soccer, to beach and futsal, Volkswagen is actively supporting all the communities and teams within the U.S. soccer ecosystem.
0:40:32 They’re supporting talent from across the U.S. soccer extended national teams and are focused on helping to give these less widely known forms of soccer a platform moving forward.
0:40:38 From the pitch to the sand and everything in between, welcome to our turf.
0:40:38 you
In this episode of China Decode, hosts Alice Han and James Kynge break down how China is quietly building the “Android of AI” while the U.S. is pouring billions into the ultra-premium iPhone equivalent. As American firms chase ever-bigger, pricier models, Chinese competitors are going lean, open-source, and dirt-cheap — and U.S. startups are already switching to them. They unpack why Chinese models are suddenly dominating Hugging Face, how an AI price war could spark a market correction, and whether U.S. export controls are backfiring.
Plus, a diplomatic firestorm between China and Japan is raising tough questions about the future balance of power in East Asia. With Tokyo taking an unusually hard line on Taiwan — and Beijing responding with fury — Alice and James examine what’s driving the escalation, what it means for U.S. strategy, and how historical grievances still shape the region’s security map. Finally, China’s coffee wars are heating up — and Starbucks is blinking. After losing ground to aggressive local rivals like Luckin and Cotti, Starbucks is selling off majority control of its China business. They explore why Western brands keep struggling in China’s hyper-competitive consumer market — and whether Starbucks can claw back relevance.
Learn more about your ad choices. Visit podcastchoices.com/adchoices
Leave a Reply
You must be logged in to post a comment.