Big Technology Podcast - How China Caught U.S. AI — With Grace Shao

Episode Date: July 29, 2026

Grace Shao is the author of AI Proem. Shao joins Big Technology Podcast to discuss how Chinese AI labs have caught their American rivals despite operating with less advanced computing infrastructure. Tune in to hear how talent, specialization, open-source collaboration, and fierce competition helped models like Kimi K3 approach the U.S. frontier, and what that means for OpenAI, Anthropic, and the business of selling intelligence. We also cover model distillation, China’s compute constraints, the growing importance of AI products, why top researchers are returning to China, and the country’s emerging robotics advantage. Hit play for a clear-eyed look at whether the United States can preserve its AI lead. --- Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice. Watch the full documentary here: https://www.gravitee.io/ai-agent-documentary Want a discount for Big Technology on Substack + Discord? Here’s 25% off for the first year: https://www.bigtechnology.com/subscribe?coupon=0843016b Stop online threats before they become real-world attacks. Visit ⁠ironwall.com/BIGTECHNOLOGY⁠ and request a free Risk Assessment to see exactly how exposed your executives are Learn more about your ad choices. Visit megaphone.fm/adchoices

How does China keep catching up to the U.S. and AI despite having much fewer state-of-the-art resources at its disposal? And what does it mean if USAI Labs permanently lose their lead to China? We'll talk about it with a leading China analyst, Grace Xiao, right after this. In the face of ongoing disruption and opportunity, TMT leaders need to deliver tangible results, not just ideas. When pace and performance matter most, PWC combines market insights and deep sector experience with AI, cloud and emerging tech to accelerate your transformation and drive measurable ROI from strategy to execution. PWC can help you anticipate what's next outpace disruption and compete.

For more information, visit pwc.com. Welcome to Big Technology Podcast, a show for cool-headed and nuanced conversation of the tech world and beyond. Today we're going to talk all about China's rise in AI, which is really now a multi-year phenomenon. And whether the United States can maintain any semblance of a lead over China, now that Kimi K3 has effectively equalled, maybe not the frontier frontier of U.S. AI models, but enough frontier to make us question whether the lead will be maintainable at all.

And then, of course, what the implications are for the AI race. We're joined, of course, by Grace Shao. She is the author of AI ProM on Substack and a leading analyst on China. China's AI efforts and of course the front of the program. Grace, it's great to see you. Welcome to the show. Alex, it's so great to be back. Thanks for having me.

So let me sort of set the stage here. We had this moment last year where Deep Seek was able to produce really great results on reasoning at a cheaper price and the market flipped out. Okay, we know that happened. But even still, you know, despite the fact that people in the U.S. and Europe, maybe all over the globe, have known that China is very capable in its AI research. The release of Kimi K3 largely had this reaction of like, how did they do that? Right. It's confused a lot of people, even though it shouldn't be a surprise necessarily because China's done it before.

You know, it's sort of happened again. And this new model from Moonshot, Kimmy K3, you've talked about on the show, it's doing really well. This is from Nathan Lambert's substack intercom. It comes number two on the VALS AI index. Number three overall on artificial analysis is intelligence index, number one in the front end Cone arena, and it has many more impressive results. So let's just start there.

Should we be this surprised? And how did Moonshot do it? I can't comment exactly how Moonshot did it. But I'll start with a comment. Actually, Nathan even shared with me when we talked about after his China big China trip he visited all the labs. I think a lot of it's in the talent

and people are really shocked by the talent. And this is something we talked about as well a year ago when Deep Seek came out with a whole domestically educated and domestically, you know, trained talent pool. Now, I think that's something really underlooked right now. You know, the talent pool in China right now kind of is because the base is so big. And then there's such a strong STEM education,

which feeds into right now. AI, you know, researcher realm. And then we see, like, the leading researchers really, really a lot of them, if not, you know, maybe 40, 50% of them are of Chinese descent or heritage. Chinese talent right now is kind of having a moment, I think. And then I think within AI research, you know, what I've heard from a lot of labs are saying, they're like, look, it's not really rocket signs, actually. The R&D itself requires a lot of taste and curiosity and test in their air.

but it does take talent and China just has an abundance of very smart mathematicians, physicists and whatnot that are going into AI. Now beyond that I think what's the kind of elephant in the room or the obvious is the obvious constraint-driven specialization or the compute like constraint you must say. If anything in a way,

what you've seen is a lot of these tiny labs faced with compute constraint and in some ways capital constraint, they're forcing. them to specialize rather than compete across every dimension. So, you know, for the sake of, you know, DeepSeek, it's really, really focused on the infrastructure, the innovation and engineering and efficiency and computing efficiency. With a lot of the others like Kimi, it's really, really focused on its agenetic pushout, you know, Minimax previously was more

focused on multimodality and ZAI focus on coding. So in that way, you can see the whole ecosystem is kind of each taking its own pie, not by, I think, design, but because of compute, a constraint. So they have to selectively choose what they do. And then the last thing is, I think it's a shared R&D, and I think this is something we can definitely talk about a bit more beyond just the Kimi breakthrough. It's just that the open source ecosystem has really harness this pretty collegial competition, and there's a lot of, you know, learning and referencing off of each other. Yeah, I don't want to downplay the talent side of things.

And I actually want to read a selection from Lambert about his trip to China where he met the Kimmy team. And of course, he is a researcher at the Allen Institute for AI. So he's got chops in the AI world, not just somebody who would write this if he wasn't impressed by the technical abilities of people. So he wrote, meeting some of the core Kimmy team on my trip to China, it was clear to me that they had incredible culture. Some would say aura and a freedom to express it within the constraints of a GPU limited environment. Where building models is so much of a scaling game, much of the ability to build a good model still comes down to individual execution, motivation, and expression. Having visited them, this result is less surprising. Having visited many AI companies, very few of a culture that you can immediately pick up like this.

And I just want to double click on that. I think both, you know, Deep Seek and Kimi, if not seen as kind of the top two labs right now coming out of China, both founders have openly talked a lot about management of people, you know, removing distractions, really focus on the pursuit of AGI, not kind of getting distracted into consumer applications or whatnot. I think both in a way, you know, represent, I think, this generation of Chinese entrepreneurs where they're not so driven by the media commercialized. but very much driven by a bigger mission. And both of them are very, very much committed to the open source ecosystem. Okay, so I was going to ask you what you mean and what Nathan means by culture.

So is it that just basically like a determination to not be distracted and to just kind of focus on the core science part of this building of AI? Is that what it is? So I can't overgeneralize every single lab, but for sure when you meet a lot of of these researchers from the labs. There is a sense of, I think the nerdiness comes through, but of course that's for every researcher. And then for deep-seek.

Yeah, welcome to research. I think for, you know, Liang Winfong, he's talked a lot about, you know, his philosophy where there's a very low churn rate in the deep-seat lab and the philosophy of committing to open-source technology, open-sorts R&D, the mission to really pursue AGI in his worldview, and his team really committed to that. So because of that, also I think, you know, take it, you know, he might be taking it for granted, but the fact that they don't actually have that much pressure to commercialize

because they have enough money, he says, like, look, we have enough capital. We are not capital constraint compared to maybe other labs. And we are really, really focused on just doing the best we can, given the constraint we're faced with. I think with Kimi, Yang Zhling also openly kind of talked about a lot of about the hardest thing about building this business. It's not the R&D. the hardest thing is managing the organization and finding the right people for the right kind of work and making sure everyone's united i think even ziyai has talked about how tangier who is

the chairman he's actually a he's a professor of yang zirling and many of the chinghua alums um he's currently still a professor at chinghua university he in many ways leans on the fact that he kind of is an o g in the industry and it's able to unite everyone and there's a sense of of unity, a sense of like shared mission. And I think people are undervaluing this case, especially when we're seeing a lot of other labs, maybe having a bit of infighting or, you know, internal mission or value misalignment.

Yeah. Now some people will say, oh, well, this is just like 9-96. So people in China are outworking people in the U.S. where they're doing, you know, 9 a.m. to 9 p.m. 6 days a week. How much of that would you ascribe to it? I think there's 996 everywhere, honestly. I mean, this sounds, this is really controversial. I think if you're passionate about what you're doing, I-996 myself, you know,

but only selectively when there's days you need to grind. Alex, I'm sure you're staying up right now. It's 9.m. in New York, you're doing this recording with me. I'm 9-9-6ing today. Yeah. So I don't think it's a top-down mandate from the company, but I think if you're driven by a mission and you're passionate about what you're doing, people are willing to work.

However, it's actually interesting. The Amelofeng even talked a lot about how he does not believe in overtime for the sake of overtime, phase time for the sake of overtime. FaceTime, which really goes against kind of the stereotype of what people think of Chinese corporates. So it goes down to, I don't think it's just pure grinding, but obviously people are hustling when they need to. Right. Yeah. I mean, still from the, from the, you know, Kimmy K3 did so well on so many benchmarks that it's still, you know, even if the cultural side of things, you know, are where they should be in terms of being focused and trying to get things right. It is

still stunning that they were able to turn in the results they did. And I guess that is, you know, you bring up the specialization thing, and I think that's really important. Like the large language model can get very large and do a lot of things. Like the same model that's going to come up with scientific breakthroughs and help you figure out what you want to order for lunch is also the model that codes. It's the same model, right? And so if you decide, hey, the most valuable thing is getting coding right. And because coding sort of is the foundation for agentic tasks, then you can start to see potentially the results that Kimmy saw when you want to specialize there.

Definitely. And I think on that note, I want to bring up a very interesting phenomenon. I'm sure you saw a couple months ago there was a huge hype around Chinese model, those Chinese companies all pushing out their own frontier models. Like I don't remember, like even like, app, Maituan, you know, hardware companies sell me. And then I think a lot of investors, U.S. investors, were reached out to me and I always ask, like, why are these companies all competing on models?

And one really interesting thing is, I think, on one hand, there is obviously the expansionary nature of Chinese internet companies. We all know Chinese companies love these super apps and they love to get into every single vertical. They can get their hands on. So, you know, the ba'bas and tens of the world do not actually. actually just do what you and I know about them. Like they don't just do commerce or they don't just do social media.

They actually do delivery, ride hailing, map, food ordering, like everything under the sun. So there's that culture in China. I think there's a lot of, it goes back to open source again. I think there's a lot of, I guess, advantage for a lot of these companies to jump into the arena because they already had a lot of open source, open weight research available to them. So the barrier to entry was frankly a bit lower. And then beyond that, what was really interesting is when I speak to the May 2 CFO, which is a leading, like, creative AI company, he was saying, look, we don't really need the most,

like the biggest model, we don't need the best model in that sense, but we need the best model fine-tune for our use case. So we can push our model in every single vertical app we have available. So they really lean in on, you know, fine-tuning open-source models for creative use, it's image or video a generation or even video editing, photo editing, etc. And I just thought that was something that's really embedded and grained it already in a lot of the tech companies in China. Like a lot of companies are already thinking ahead of time, or maybe a few months ahead

of this new mainstream narrative in the U.S. we're seeing where what makes you valuable and is that your proprietary data? And is that proprietary better executed within a smaller model that's better used, created for your use case or is it better that we actually all pay for the biggest largest best model? Right. So having that data combining with a smaller purpose built model can actually deliver similar performance as using the bigger model, maybe with less of your data. So, okay, you mentioned a few times, and I think we should talk about it, the benefit of doing open source, right? And you called it a share R&D sort of effort here. And I remember after Deepseek came out,

I spoke with somebody who knows their stuff in AI, and they were basically like, look, with open source, it's every open source research house working together. When you're building a closed model like Open AI or Anthropic, you know, have, you're basically building on your own. I mean, of course, they can bring in the open source innovations. But it's just, it's basically them against the world where open source, you kind of have the world against everybody. or against the closed models to be more technical about it, more accurate. So just talk a little bit about that because I think that's an important point. Yeah, I think a lot of even just bringing it back to the kind of the narrative around China versus USAI right now, I feel like it's really about open source versus a closed sore at this point, right?

And I think it was really humbling to even see this morning Kimmy released their weights yesterday, like last night in Asia time. And in their opening paragraph, they talked about how like we are still. behind the leading most frontier, but we are inching towards it. Basically something like, I paraphrase, but it's something like that, something in those realms. And I think what it really shows is open source is able to somewhat now play catch up because they are leaning into basically everyone's intelligence or everyone's R&D.

And I think it's really played a large role in propelling China's open source ecosystem. And it's something a lot of the leaders who just talked about like Nangu Pheng or Yang Zhiling have really, I think, embodied as well as rallied behind. It's essentially in the beginning, it was like a branding strategy for a lot of these labs and when I spoke to them because they said, look, if we want more developers on our ecosystem or on our, using our APIs, they need to know what's out there, especially since we are Chinese, frankly. And they're like, if we don't put out our R&D, you're going to have a lot of accusations of this and that.

But we put our papers, we put out our research, you can be the judge of it. So that was the initial kind of starting point of open source and how it got, I think, a lot of users, especially startups that might be more cost constraint and less compliance, you know, worried, getting on these labs. I started getting on these models. And since then, essentially, it's become like a nice virtuous cycle because the more developers on it, the more you learn about, you know, the use cases and whatnot. You can tweak it, you can make it better. So that's really kind of the original, I think, goal was to just sell their models abroad. Now, I think from there, it's really become an unintentional consequence where, you know, the learnings of each lab will now serve as essentially open learning textbook for each of these labs. And they will openly congratulate each other.

I believe, I think when Deep Seep came out with something, ZAI even retweeted them on Twitter, on X, saying, oh, congratulations. this is such like, you know, genius work, blah, blah, blah, like we will incorporate it in our own, you know, our R&D, and our own infrastructure layers. So essentially, you know, that's really what's been driving it from the commercial sense. And I think back on the culture sense, you know, I think a lot of academics and researchers actually really like to be followed, cited, and it's part of that kind of academic loop.

So the people who want to go into private, they need to be cited to go back into academia. and this is something I've spoken to, I've learned from speaking to a lot of the academics. And this allows them to kind of have their work in public. So all of that has garnered a very strong open source philosophy, base in China. Now, a very interesting thing has happened in the ensuing days. And we're talking Monday, July 27th.

This will come out Wednesday on the 29th. But this is all live and this is happening. If you looked at what the advantage of, of let's say US and China were. China, of course, is this open source ecosystem. The U.S. was leading or is leading, I would say still leading, at least when it comes to model intelligence. And the advantage that the United States has had is these closed AI labs,

open AI and anthropic, that have pushed the frontier forward again and again and again. And you would think that if you were thinking, what's the strategy going to be for U.S. companies, it would be almost a fear of open source and a rededication to closed. But the exact opposite has happened this week, where, you know, basically led by Jensen Wong, the Nvidia CEO, seemingly every U.S. company has come out in favor of open source. Even OpenAI has signed on to this letter saying we shouldn't ban open source. Then after like days of silence, Anthropic basically had to come out with a statement said,

hey, hey, by the way, we never called for the banning of open source. So, so Grace, just help us understand. What do you think about the fact that if open source is China's big advantage, how do you then explain what's going on in the U.S., where all these U.S. companies are coming out in full-throated vocal support of open source? Well, I mean, there was a huge 180, right? And I think it was quite interesting. But, you know, going back to what's something we just touched on already,

I think in the last year, low-key, a lot of companies have been building on these open-source models, right? Because essentially, if you self-host these models or you use them through these inference service providers like fireworks, these models become yours or American, if you want to put it, right? So the whole fear-mongering around, does the data or whatever, go to China, that narrative doesn't really work anymore. So then it was very interesting because I think at the day it was kind of like, okay, if we don't open-source and we build the strongest open-source models in the U.S., then, actually this pie is being eaten by someone else anyway. Like it's not like we can stop people from using Chinese open source. And even despite, you know, Kimmy K3 being said is a token hungry model. It's not as cheap as other Chinese models.

The task per, I think, the task per token usage is still quite high. It's still cheaper. It's still cheaper than the most frontier models. And it's inching towards frontier. So it's almost like, I want to cynically say it's a business decision. That's one aspect from the labs. So it's like, well, we now need to come. be an open source, you know, back then the margins were extremely high, right? And I think, you know,

I can't comment exactly how high the margins are because they're not disclosed, but, you know, on the China side, Deep Sikhs people have talked about how they are not revenue driven. So whatever money they make, sorry, they're not profit driven. So whatever revenue they make, they want to put that money into R&D, given that they have the quant fund kind of like, you know, making their money. And I think the philosophy of the founder is not so commercialized. So I think there was, was like a bit of a push for open sourcing. Then beyond that is something going back to what we also touch on earlier, which is why should companies continue to pay for the best intelligence when the intelligence are kind of taking away what makes the company special? So then you see

companies more and more mindfully saying, okay, we shouldn't actually give all our data to Claude and GPT and then in return pay for intelligence twice. That's what the Microsoft CEO said, right? So I think there's a bit of, I think it takes a step back and all understand what where people's perspectives are coming from and what their own goals are, right? That's right. Okay, I actually want to get to what the fact, I want to get to the fact that you can now get open source models to do a similar job as some of the frontier models or close to the frontier models for a cheaper cost, what that does to the AI competition. But before I go there, you know, we're not even 30 minutes in, but we've made it 20 minutes in and I haven't brought up distillation yet.

And so I think a lot of our listeners, if you've made it to this point, have, are probably saying to themselves, well, Grace, these are nice explanations, good culture, and, you know, specialization, and I can buy that at a surface level. But we do have evidence that the labs like Deep Seek and Moonshot have distilled. So basically, taking the essence of the big LLMs from Anthropic and potentially Open AI, and, you know, quote unquote, taking the AIP of these closed labs to build their own models. What's your perspective on that? Yeah.

First of all, I want to say, obviously, no one has come out publicly saying, hey, Grace, I've distilled a model. So I just want to help me to say that. Right. It's more like there's like some research that indicates that it's happening, but like neither of these companies. have said they've done it. Right, right. So I think it's really interesting. And again, just this week, I was listening to the All In podcast.

I'm sure you're like, I think even Freeberg and Sachs, these people who are quite, I wouldn't say anti-China, but had a tough stance on the national security angle on China AI. We're saying, look, distillation is a practice that's been widely used in product iteration and R&D, even, you know, in its Google's days. I think Freeberg was saying that. But beyond that, what's something I found the most enlightening, the framing was provided by Google Deep Minds Yao Shun Yu. He's a researcher at Google.

And he said there's smart distillation and dumb distillation. Dumb distillation is something what, you know, the layman think of when distillation happens. Essentially, you know, I literally take, you know, Model A's answer and plug it into Model B essentially. And then Model A will just spit out what Model B would just say spit out what Model A said. And it's so obvious. It's like copycatting. And I don't think anyone is quite literally doing that because, frankly, it's just too unsophisticated.

Now, there's smart distillation. which is kind of operating in the gray area. It is, again, not IP theft. It's not breaking the law. However, it could be breaking what is considered, you know, your own term services. I think labs should be doing better, you know, KICs. But in that sense, essentially it's like what enterprises are doing in fine-tuning their own models. So say if I'm an enterprise and I self-host an open-source model, I'm going to fine-tune it with my own data.

I'm built harness around it. I'm going to make it the best model for my own. use case. And often I use the frontier model to guide that kind of less frontier model in terms of getting its homework done instead of kind of getting to the right direction or even for synthetic data training. So when you're looking at smart distillation, it's a lot more murkier and it's really hard to say what is right or wrong. And I think it goes back to the mainstream narrative that we're hearing right now, even the happening in the US, what is distillation when we also are hearing potentially, you know, the thinking machines model, distilling on Chinese model.

So it's quite funny where everything's kind of going in a full circle. And it goes back to my point of open sourcing R&D, where it's really open sourcing R&D really pushes the whole industry forward as a unit. And if you really believe in AI is really going to help and transform our economy, how we work, our basic infrastructure, then pushing it forward together, propelling it to. together, it will make more sense. But if you believe this layer of models should be capturing all the value and you should be selling intelligence, then of course you want closed model and you don't want people distilling your models. Yeah, let's read from Nathan Lambert one more time,

just for the heck of it, because I think he has a good perspective here as well. And you've, you highlight this actually on your substack. So I think it was sort of downstream from your finding, Grace. He writes, it is clearly the strongest open model ever. released writing about Kimmy K3. It should be clear looking at this model that if adversarial distillation from the close frontier models in the U.S. contributed, it is at most to a relatively small degree. AI observers who followed the distillation panic and came away with the wrong conclusion. The Chinese AI labs are only producing good models due to IP theft are in for an awakening. The Chinese companies are extremely good at building models in the same way the leading

American companies are, which I would say is even more remarkable given the fact that they can't get the latest Nvidia chips, right? So they have to work off of generations previous. They couldn't get the latest chips and they couldn't get, you know, access to Fable, but the window was too short for when Kimmy K3 came out. And I think, look, Nathan is the technical expert here. If he believes in that, I would really, you know, trust his judgment. I do really appreciate and respect his work. Okay. So, you know, we spoke about this the last time you were on the show last year.

And we got to speak about it again. Because, you know, you mentioned, all right. So like Deep Seek, you know, hedge fund funded, not really interested in the profit. I think that and Moonshot is owned by Alibaba funded by them. They have some backing financial back. Actually, a lot of the labs have a bit of financial backing right now. But we can talk about their financial, like a breakdown. Yeah, so that sort of sparks the question, why are they doing this?

Like, why are they doing open source? It seems like if you're doing all this innovation, right, they are giving away the weights. All these companies are downloading the weights and building their own applications and doing it with like probably USB, you know, in the U.S. doing with US-based consultants, you're not doing it with Moonshots consultants. What is the economic advantage of open sourcing all of this technology? So I do want to start with, obviously, a lot of these labs are actually looking for fundraising because obviously training models is extremely expensive game they're playing or a task

they're trying to complete Pete. You know, ZIA went public earlier this year, mini-max went public earlier. There's both on Hong Kong Stock Exchange. Moonshot is in the pipeline, supposedly removed within the next six months. DeepSeek supposedly also looking at the starboard in China. Now, that out of the way, they need money, right? It's not like they don't need money. However, I think there's a misunderstanding around open source monetization. You're still paying APIs for managed services, and, you know, you're still paying

potentially fireworks for inference service. And if you pay for fireworks, inference service, they usually have a commercial agreement with, like, say the Kimmy provider, where there's come kind of a break as well. Open source is not anti-commercial, but obviously it helps you make less money. Now it goes back to, are you that money hungry, or do you want the technology to proliferate and diffuse more, but you can still make money? And I think this is where people are also realizing actually kind of calling some of the closed frontier labs a bit of hypocrisy right now, because actually labs monetize, you know, like I said, through API access, managed services,

a lot of times, you know, you and I probably will not be buying our own GPU and deploying our own models and running the security debuggy, monitoring. So there's a lot of need for still buying that API. Now, beyond that, I think, you know, if you look at, I think Minimax, CI, I think Z AI's run rate,

ARs already, something like $1 billion now, and then Minimax projected to be $1 billion or to $1.2 billion by the end of the year. So, yes, not as lucrative as maybe anthropic or open AI. They're not not making money. In fact, they're making a lot of money still. Right. There's also this view that like, well, if you think about national competitiveness,

and we've seen governments get involved. I mean, Xi Jinping gave a speech about the virtues of open source. The U.S. government seems to be touching AI every week now. You know, I think in part for a desire to mitigate the harms, but also because they view it as important, you know, from a national strategic perspective to have the lead. So there is this view that like, you know, the companies in China can open source because for the government in China, it's, you know, basically the best possible outcome is to commoditize this like leading industry in the United States. Your thoughts?

Okay. First, I think the argument around subsidization is really funny because I don't know. know if the government is that rich, frankly, just chucking billions and billions at every lab. So definitely I don't think they're like. Yeah. Like they have a lot of money though. There's a lot of subsidization on energy and data centers, but it's definitely if you talk to these labs, they're still, they're private companies. And in fact, most of them actually don't want to take government money because there's

some hindrance as well, right? Like when they go public and et cetera on their structure. Now that side, I think it was a really interesting that President Xi Jinping attended WAC, which is like the world. It's called the World AI Conference that hosted annually in Shanghai. It's been around since 2018, but it honestly didn't really get much traction until maybe last year when post-deep takeoff. And now, like, there's floods of American investors, American policy, think tank people, all going in. And I think what was really interesting is, to your point, I think governments are viewing AI as a very strategic driver.

Now, it could be a driver. I think it's a fewfold, a driver for economic prosperity, a come on. growth, of course. It's a driver for, I think, soft power and diplomacy, of course. And now also obviously a very fundamental point on technological competition. In terms of what she said, I think the highlights was really about openness and inclusivity, which cannot be actually mistaken for quite literally embrace the open source. I think he talked a lot about openness, inclusivity that was echoing what even the previous leader that attended, which is, I think it was the premier that attended WSA last year. It's a lot of the messaging towards

the global south because I think there's a lot of worry around, you know, countries that frankly don't have the talent or compute or even just the raw material, whatever needed to right now participate on the model layer. They don't want to be left behind. And Xi Jinping's message is saying, hey, we will be exporting this along our belt and road, essentially, that you can still participate in the AI boom or the next wave infrastructure upgrade. Yeah. I mean, I'll just say, you know, and then we'll go to break unless you want to comment on it. Like, it's definitely in China's interest for this all to just commoditize.

Even if it's not the direct strategy, I'm sure they're quite happy to see the U.S. industry. after all these billions of dollars I've been put towards the developing of models, at least sweating a bit. So there will have to be, you know, some adjustment on the U.S. side because this is sort of, if you're thinking about it from the closed model standpoint, it's not what you want. It's probably why we saw anthropic spend all that time, you know, waffling, or not waffling, but just not responding to this open source, you know, sort of moment of praise in the U.S. And so that sort of leaves us to what the competitive dynamics of AI looks like, assuming this continues to be the rule that open source, you know, it used to be that the thought was open source was a year behind like the U.S. frontier models. Now it seems like it's just months. So what does the competition look like? We'll cover that when we come back right after this. Hi, everyone, Alex Cantowitz here. I want to tell you about a documentary I've made with gravity

to explore the future of AI agent security. To find out if we're truly ready for autonomous agents, I sat down with MIT Professor Ramesh Rosker, former White House CIO Teresa Payton, Michelin's Group Chief Data and AI officer, Ambika Roger Gopal, and Sharon Guy, a former executive at Alibaba. They each offer unique insights into this evolving landscape.

We conclude with Rory Blundell, CEO of Gravity to discuss the path forward. With gravity leading the way, join us on this journey. You can watch the full documentary at the link in the show notes. This episode is brought to you by Deepel. When I sat down with Deepel's founder,

Yarrukutliovsky on YouTube recently, we got into the case for specialized AI. Deepel voice is what it looks like when the stakes are real-time conversation. And honestly, it's something I wish I'd had for my own cross-border interviews, turning a language barrier, into a non-issue. Deep Bell Voice delivers live translation in over 40 languages for virtual meetings

and in-person conversations, helping people speak in their preferred language without losing flow or nuance. Whether you're meeting with a customer, negotiating with a supplier, or collaborating with global colleagues, it keeps pace with you in real time, easily handling the technical terms, acronyms, and product names specific to your business. So what you actually mean never gets lost in translation. And for the builders listening, Deepel's Voice API, lets you embed real-time speech transcription and translation directly into your products. So go check it out for yourself. You can try DeepL Voice for free at Deepel.com slash try Voice.

That's deepel.com slash try voice. Today's executives are more threatened, more exposed, and more vulnerable than ever before. Corporations spend billions in workplace security. But what happens when a threat finds your executives outside the office? 70% of attacks on executives happen at home or away from the office. and Ironwall understands a terrifying reality. If someone has a grievance against your company, the first place they turn to is Google. It takes them about five minutes to find one of your executive's home addresses online.

And if their personal information is sitting on the open web, they're far too easy to find. The team at Ironwall knows this better than anyone. They've protected some of the most targeted executives and individuals on the planet for almost two decades. Protect your people with continuous personal data removal, proactive prevention tools, and emergency support. So when someone goes looking for your executives, Ironwall ensures that, they hit a dead end. Go to ironwall.com slash big technology, fill in the quick form,

and request your free risk assessment. The team will show you just how exposed your executives are and how to lock it down before a threat reaches their front door. That's ironwall.com slash big technology. Stop online threats before they become real-world attacks. And we're back here on Big Technology podcast with Grace Scha. You should check out her substack. It's A-I-P-R-O-E-M, so it's A-I-P-R-O-E-M dot substack.com.

where she covers this world in great depth and with great clarity, so highly recommend you sign up. Grace, let's talk about what the competition. By the way, feel free to comment on what I shared before the break, but also like, what does the competition in AI look like right now if you have, you know, basically this world, let's say we get to the world where open source commoditizes the closed models and, you know, the whole plan was to sort of have these closed models and sell AGI on a meter. But if you can't do that, then what happens to AI? I think for sure there has been a bit of a global reckoning, I think, on twofold.

One is the need for governance because of how much these AI models can do and the potential risk that's been talked about, especially in the mainstream narrative in the U.S., right? And I think that's really sent kind of fear a bit around the world. Now, given that, I think following up, and we just talked about, I think it is in China's interest. And it was reiterated at WAC, where AI governance will be another focus.

And this ties to, I think, the whole open source thing, because it says, basically they're signaling, let's build around the industry and find guardrails to basically, in some way, control or contain this technology, because they still see this technology as similar to any other technology in that sense. It's not like this new mythical creature that we cannot contain. And then from there, let's export it to the world from a high level. From a business level, I think this is the second point I was going to touch on,

which is it's been really interesting. I think even a year ago when we spoke on big technology podcasts, a lot of companies were really, really gunning for the U.S. market. It was seen as if I want to sell, the U.S. market is always going to be the most lucrative, enterprise companies are willing to pay, blah, blah, blah. You know, if we make it in the U.S., we've made it, right? That was the, like, holy grail. And there has been a complete change in mind recently

when I spoke into quite a few of the leading Chinese products, whether they're on, like, coding agents, what whatnot. It's really focused on potentially going to, you know, Southeast Asia, potentially going to Europe. Because they're saying, okay, first of all, geopolitical headwinds is not, like, it's no joke. going to be easy to sell to U.S. There's obviously a grueling competition in the U.S. domestic market, but there's a lot of desire from other markets that want Chinese technology providers.

They're saying maybe some of them don't want to pay that massive premium from U.S. tech. And maybe some of them are also losing a bit of interest from, you know, the very scary narratives that they're hearing from the U.S. as well. And then on top of that, many of them are looking to build on top of open source models. And they need a support to help them kind of that, you know, infrastructure around it. So it's been very interesting to hear that kind of mentality shift. Yeah, you also, I mean, you put it basically the problem for the U.S. closed source or closed model developers. You put it very clearly. The central risk to the frontier labs is not that their model suddenly become useless. It's that frontier level capabilities, capability becomes

increasingly difficult to monetize at premium prices when open weight alternatives can perform most tasks at a fraction of the cost. I mean, you come out this from a business standpoint. If that becomes the reality, right, what do you even do if you're a close source company, like a close foundation of lab? I think you can still, you can still charge. I think, you know, for certain government agencies, certain companies, you know, Fortune 500 that might have very strict regulation compliance or rules, whatever. I think for certain sensitive sectors, if you were to want to use American tech stack, it still makes a lot of sense because maybe the money does not is not a main fact considering factor. Does that make sense? But for a startup for SME, like every penny

matters and you're going to want to find the best model for your ROI. So I do think at the day, majority of the world actually runs in a very pragmatic lens because you got to pay your bills and you kind of make sure your business is generating money. So when you are buying for intelligence, that intelligence needs to make sense and justify the cost of it. And I think what we saw a couple months ago was a sudden awakening or realization that a lot of the token maxing wasn't making investment sense because, you know, spending a million dollars per person on token usage is mental when their salary is maybe like 200k and their revenue generation is even lower than that. Do you know what I mean? Like it doesn't make any business sense. So I do think businesses will

look at this very differently and it's putting pressure on the closed models but I think people who are working at the most frontier or even like I'm just pulling a name out of my head but like a Jane Street if you're going to spend $2 million per head but you're going to generate like

$20 million on each you know bet like on each investment then that money is justified to Jane Street probably but I think you know it goes back to how do you justify that cost right and I think you also, I mean, you effectively have the answer in your piece, or maybe a different answer,

or maybe an answer that expands upon this in your piece that I think is, you know, sort of says it all, right? So you say, and this is your reaction to Kimmy K3, you say K3, but you can basically say this about this entire open source moment. You say it's the latest evidence that the AI frontier is becoming more contested, more global, and potentially less proprietary. And I think we talked about this a little bit last time also what happens when the AI intelligence itself becomes more proprietary becomes less proprietary the thing that matters is the product right so i asked earlier what do the AI labs do and we've been on this a bit on the show recently if you're if you're if the intelligence that you've developed becomes less proprietary um if it becomes more of a

commodity and less of something that you can hoard your products are are just what matter and we we saw this with Deepseek and we're seeing it again here. I think that if there was a belief that you could build a trillion dollar company just by selling intelligence from the API alone, not going to happen, most likely. But what could happen is if you're the developer of the intelligence, you can build products with that intelligence and sort of return to your investors that way. Your thoughts? Yeah, definitely. I think, you know, on Deepseek, first of all, I don't think they would push out products because like I said, I don't think they're trying to commercialize or. No, no.

I'm talking about open AI and anthropic. I think the deep seeks and the Kimmy K2s of the world are thrilled to build just the models. But I do think I agree with you. I think there's a lot of stickiness when in the products. And I think, you know, Claude, co-work, these products like are still very, very sticky. And there's still going to be a lot of potential for them to continue to grow. Like I recently spoke to designers where I thought maybe like do you use Claude Design or do you still switch back to Figma because FIGBA isn't going to add on AI.

Surprisingly, the answer was that if I'm going to do everything within Cloud already, I'm already using it as my thinking partner. I'm already using it for coding. I'm already using it for all the other tasks beyond my actual day-to-day design work. Then I will also use Clod for design. So it's almost adopting a super app kind of mentality. It's like I want to capture all. Now, however, I have the reverse kind of like the counter argument as well.

So I went to visit Alibaba recently. and I was really fascinated with a product that's just really, really not in the radar. Frankly, I don't even think their own company is really valuing it. It's called Axio. And it's actually just a very simplistic agent interface built on top of their supply network. And why it really amazed me. It was because the know-how and the actual edge they have is not as intelligence. Because, like we said, frankly, finding a supplier for, I don't know, a glass or a lipstick or a microphone,

is not difficult, right? But what their actual proprietary strength or their know-how that no-one else can replicate is their 20 years of like kind of doing business in finding suppliers, matching with merchants, and then helping merchants sell it. So they're a 2B2C business,

and now they basically built an agent on top of the 1688, which is their wholesaler website. And like I was talking to their people, like their representatives, and they're showing me the interface. It looked like co-work a bit glossier, like prettier because you know how Chinese apps love to have like a zillion different buttons. So there it's colorfully colded and it has all these buttons. And if you are just an SME or like a drop shipper or like a Shopify brand owner,

all you need to do is go on the platform and say, I'm looking for insert 8. Like you can make merch for big technology. And I want merch for big technology. This is my, these is people I've interviewed. This is my vibe. I don't know what color I want. I don't know what size. I want something universal.

Help me think it through. And they will literally go through their database and try to find you their right optimal kind of product. And then they will source it and they'll talk to the supplies for you. Find the best price for you. Then give you the three to five options. And then you can pick. So for me, I was like, wow,

the actual edge of these products are not an intelligence. It's opposite of clot, what clot is doing or GPG is doing. They're not going into every vertical. They are simply doing this one thing, but doing it very well. And it's kind of similar to what I talked about inmate tour as well, which is the creative industry. They're like, we build these models, and they're going to be the best models for you to use to optimize your e-commerce product placement or to filter your jaw of face and make your jawline more shizzled or whatever. But they're not for you to build, you know, use to create the next, you know, Hollywood blockbuster or whatever. So I think there's more and more awareness around that now.

It's so interesting. What you're describing is almost a flip of what we've seen in the consumer internet up until this point where China had the super app and the US had a bunch of disparate apps. And I think maybe because of China's embrace of open source and because of the US closed AI model, we might end up seeing a proliferation of individual AI apps in China while the US goes to super apps. It's like the craziest dynamic. That's so funny. We just coin something. Alex, We need to IPE this. We're called it first. That's good.

That's good. We'll do a co-byline on substack. That'll be fun. Okay. We have to talk more about compute because, yes, you know, you could say Moonshot was able to design Kimmy K2 or K3. I keep calling it K2 because I get the mountain stuck in my head, K3. And there was a Kimmy K2. But with the compute they had.

But, and you talked about they get paid when people want to, you know, use the model off their API. The issue is that they couldn't really sustain a lot of demand, right? They had to take the model off, basically, or not they had to limit new signups because of their compute constraints. And so if I'm, let's say, let's say, I'm going to try to channel like Greg Brockman from OpenAI. You know, he might say, China can go and ship all the parity AI they want. If they can't deliver it with enough compute, it doesn't matter. And compute is going to be the thing that makes open AI win in this race. What do you think about that?

Honestly, I think compute is the obvious constraint. I don't think anyone's hiding from that. None of the lab leaders are hiding from it. And the fact that, like you said, Kimi literally posted on Xing that they needed to, you know, reassess basically who they serve. They're not the first to talk about. I think last year around Chinese, or this year around Chinese New Year, GLM face something similar, Deep SIP has faced something similar when demand is literally higher

than supply, which is funny because, you know, the argument in mainstream is always about, is there enough demand for AI? There is. Now, on the China side, is there's not enough supply for AI? So I, like, as in the intelligence side, I like the compute side. So I think it's, it's not a secret that China's trying to figure out how to build their, like, self-reliant tech staff, It's not a secret. Huawei is working very closely with Deep Seek on how to optimize hardware and software and try to figure out this.

Now, I am not a semi-expert, so I can't comment too much more on the technicalities, but I think even Elon Musk recently came out during a economist interview saying he said something in the lines of, I believe China will figure out bloggery and it's closer than we think. So I wouldn't give a number on it,

but when I hear from other experts, like I've spoken to Paul Triolo, who is a expert in semi-semi-semiconductors and especially China's supply chain on the space. He also believes, like, China's going to find probably solutions, you know, it might not be the smallest chip, you know, but they could potentially find other ways to optimize the chip, even if it uses more energy. And it goes back to what we talked about even last episode when we spoke, it was like China's energy infrastructure side is not a problem. So if you have enough electricity and power to be powering these chips, even if you need to use double amount of energy, that's not a kind of a bottleneck for the AI compute side. However, obviously, the arguments then how sustainable is it in the long run for the environment and everything.

But I think it's an ongoing R&D and potentially we'll see more breakthroughs coming in the next few months or years. Now, I'm going to ask you a question that you've called ridiculous in your writing, but I feel like we shouldn't. let it go unaddressed, which is, you know, we started this conversation about, you know, you talked about how China has a lot of homegrown AI talent. And that is true. But the wrinkle in the Kimi K3 story is the Moonshot CEO, Yang Zilin, did a lot of his graduate work, or his graduate work at Carnegie Mellon. And in fact, it's like Carnegie Mellon advisors were celebrating his breakthroughs on X in the days after the release. So the question that you've called ridiculous is why didn't he stay in the U.S.?

I think for, you know, audience outside of China, it is an interesting question. And so I'm just going to ask it to you. Why did he leave and decide to do this elsewhere? I can't exactly tell exactly what he said, obviously. But I think multiple star researchers have reached. turn to China over the years. There's obviously various different layers of this. At a very high level, people love to say, obviously, geopolitical headwinds is not making easier for Chinese national or Chinese ethnic people. I think the rise of racism, frankly, even during COVID, made people feel uncomfortable

at a personal level. That we don't know. That could potentially attribute it to it, right? But then there's also just the personal reason I think is the main driver for most people, honestly. Like, I've spoken to researchers where hate to overgeneralize, but their wives or spouse or whoever are maybe teachers, lawyers, you know, healthcare professionals in China. And those are not very transferable skills, like not very transferable credentials. You know, previously what we saw in a scene in immigration or immigrant families is that people who got like who were doctors will maybe go to the US and have to retrain, redo their residency or even, you know, frankly not be able to practice anymore. So there are a lot of personal reasons driving a lot of researchers and tech professionals saying, hey, I would actually

rather be close to my home. And then I can have my spouse or family do whatever they want and they can still build their own career. There's number one. Then there's obviously the family kind of point. I think, you know, again, people really want to be close to family. I think it's not that hard to understand, right? There's that. And then on top of that, I think, is the more nuanced thing where I get some hate from. But look, I was raised in Canada. I was educated in U.S. I think a lot of my peers around me are similarly like that. And I think there is a choice for people, frankly, like us, and it's a privilege we have. And when you think about it, where you want to stay, there is obviously the visa requirement. But then there is also the quality of life and your own

native culture, right? So where I sit in Hong Kong, I can natively be both Chinese and Western. I think it's very actually accepted. I think someone like Yang Zhling or, you know, Yao Shui at Tencent, who's a former Open AI researcher, You know, their native language is Chinese. Their native culture is Chinese. And then it goes back to the quality life. I'm sure people love to say, oh, but you know, why would you want to stay in Asia, blah, blah, blah. The fact is, I think, 30 years ago, for any average Chinese immigrant to go to North America, it is a no-brainer.

Because the quality of life in almost any major city is going to be higher than a major city in China. But that kind of, I'm not talking in politics. I'm talking about individual. haste and individual lifestyle, the day-to-day me buying coffee and living in a nice apartment, things like this, the quality of life is no longer justifiable. And I think it's really a trade-off people decide. So just to bring it back to myself, my parents immigrated to North America more than 30 years ago, that was a very easy decision for them.

Back then, they were educated in North America. They stayed on, right? No one actually questioned why would you return to China or anything like that. And but now actually if you meet very, very talented researchers, finance professionals, journals, whatever, right? People want to return to their home country because of familiarity, but also because of very high quality life. And then on top of that, I think Yang Jaling, knowing that he is a genius that he is, probably had some more, you know, relationship in China to help him build this out and his own peers, his own team to build out. imagine, you know, moving to a new country in your 20s and trying to build that kind of relationship and build up that reputation.

Totally. Okay. Grace, my last question for you, I don't want to let you out of here with out addressing what the next step in this is going to be. And I think we all know it's robotics. You know, we spoke about it last, last time that you were here, that China has this advantage because they've been building a lot of hardware and now they have AI development. So just give us like a quick look at how robotics has progressed in the last. last year in China. I recently saw, I think it was in Shanghai, a bunch of robots fighting

each other in a UFC. I don't know if they were like telecontrolled or autonomous, but it does seem like there's some progress being made. And obviously, it's a place we got to pay attention to. So give us the update on that. Yeah, there's definitely a rise of so-called tech investor tourism in the Xinjiang GBA area, which is the greater Bay area that connects to Ohio Shinjin in Hong Kong, because that's where all the manufacturing of robotics happen right now. So for sure, I think China right now has a huge advantage in the supply chain of robotics because given the last 30 years of being the manufacturing hub of essentially everything under the sun, any robotics company that will have some kind of a supply and touch point in China

and likely from that GBA area to start with. Then beyond that, I think a lot of the speed is an advantage when I speak to people on the ground, production line can roll off almost 50% faster than other countries. Even if you look at like an EV car or something, when I spoke to people in the EV industry, they're saying a Zika car can roll off the production line within a year and a half from design to production versus maybe three to five years for traditional OEM.

That kind of Shinjin speed transfers into robots. It's significantly cheaper. It's said that a lot of these robots, whether they're humanoid or industrial robots, the raw kind of hardware itself can be like at least 50% cheaper than produce elsewhere. So all of these have really kind of emboldened China's hardware space or robotic space.

And beyond that, what's really interesting is you're seeing a lot of companies in the EV space or in other areas that are autonomous driving. They're now inching towards and expanding their range into humanoid robots. So it's very interesting. However, that said, I will preface saying, I think, you know, human or robots right now, it's still in a very nascent state. Of course, the dexterity and the mobility has improved significantly from what we've even seen last year, or 10 years ago significantly. But the real life use cases are very, very minimal. Because if you think about how much it takes for us to even like lift up an arm like that, like I look like all the T-Rex, but, you know, it's not easy.

And what is the real use case of this? But this takes training. And then the real, real, real bottleneck for all these companies, whether you're talking about autonomous driving or industrial robotics or human robotics in terms of AI integration is that they don't have enough physical data. And I think that's where the world models come in and we're seeing a lot of competition going in there right now. Yeah, that makes me relieved. I think, you know, I have appreciated the fast progress that we've seen over the past couple years.

But if we had a robotics intelligence explosion alongside, this like LLM explosion that we're happening right now. I don't know if we could handle it. I mean, I'm sure we'd figure it out at a certain point. And they will get there. The researchers will get there on robotics. But kind of, I don't know. I think it takes more time.

You know, I just spoke to Pony AI CEO a couple weeks ago. And he basically was, you know, when Silicon Valley said, okay, autonomous driving is going to reach us in next three to five years. And it took him 10 years. So I think he found out of the company, 2016. Now 2026, they're deploying hundreds. of cars in their fleets. He was saying, I was like, what about robot?

He's like, look, it's the same thing. People love to hype about it. They're saying, look, human or robots are going to be deployed in next two to five years in the mass market. He gave me a rough number of 10 years again. And I believe in because I've seen a lot of these robots. They frankly can't do much. And then it's also the, again, ROI.

Like, if you're going to buy a robot and help you restock bottles on your convenience store. shop, those cost about 700Ks, almost 100K USD. In markets, especially like Asia, across Asia, where labor definitely does not cost 100K for a year, how do you justify that? And they do extremely so and often make mistakes, where you can actually create employment. I don't think they're taking anyone's jobs anytime soon. I think industrial robots in manufacturing warehouses where actually there is already a labor

shortage globally, they will see more use cases and more mass adoption, but this is not going to be seen by the consumer eyes. In fact, they're already being deployed, right? Like the six axes arms, the things that lift things, logistical use cases where autonomous vehicles like that look like, I think the company was called Nealex, they look like little boxes are already on wheels and shoveling things and taking things and putting things back in shelves and warehouses. These are, I think, use these cases where they're actually complementing the current workforce because a lot of times these are not really fun jobs. These are your laborers jobs that people don't. don't want to do already. And frankly, the intelligence of it is very low. So they can just repeat,

program it. Yeah, I believe in the potential for these things, but right now nothing makes me happier than seeing a humanoid robot brought on stage for a demo and just falling and totally collapsing. I mean, I have great joy when that happens. So, but eventually, get it right eventually. Just like, let's take our time on that one. All right, Grace, the website, I should say it again, A-I-P-M-Substack.com. P-R-O-A-I-P-R-O-E-M-Substack.com. Grace, you've done it again. You know, two years in a row,

great opportunity to speak with you and help us understand everything going on with the Chinese AI movement, which I think will only get more interesting from here. So thank you so much for coming on. Thanks for having me again, Alex. All right, we'll have to do it again soon. Thank you, everybody, for listening and watching,

and we'll see you next time on Big Tech. Technology Podcast.