AI in China, The Current State of Seed Investing | July 17, 2025

0:15Welcome, everyone, to the information's TITV. My name is Akash Pasrich. It is Thursday, July 17th, and we have got a fun one for you today. We are talking with our Asia Bureau chief, who is here from Hong Kong, to give us a real ground truth view of AI startups in China. I am really excited for that conversation. We're also talking about a big investigation that the information published earlier this week. And we're talking with Slow Ventures general partner Sam Lesson that always promises to be a fun conversation. But before we get started, it has been a busy 24 hours in the information newsroom, and I want to go through some of our reporting with you.

0:51This morning, we published the first look at the org chart for Meta's superintelligence labs. Alexander Wang, who is now Meta's chief AI officer, and Nat Friedman now oversee 3 ,400 people, according to our reporting. Wang reports directly to Mark Zuckerberg. Everyone who used to lead Meta's AI team now reports to Wang, which is quite remarkable to me. And one detail that I've got to include in here from the story, Staffers have to pass through two layers of security and use employee badges with special privileges to gain access to the lab, a sign of its secretive atmosphere. I encourage you to go check that story out.

1:27The other story I want to highlight is we had a story yesterday about how anthropic investors are now interested in backing the company at$100 billion valuation. That is nearly twice the$58 billion valuation that it got just months ago. So the most interesting part of this story for me was that we actually got a hold of Anthropix margins. These are the gross margins and how that compares to OpenAI. And, you know, it really just goes to show, you know, the puck is now going to margins. It's not just a conversation about revenue. And so that was a great story that I encourage you to check out. The last thing I'll mention is that we have a breaking story today about the battle between fintech and crypto and the traditional financial sector.

2:09It's a story about you've got fintech and crypto on one side and JP Morgan and the banks on the other side. This is from our crypto reporter, Uiqi Yang. I encourage you to check that out. And with that, we've got a couple of the headlines for you very quickly here. The House of Representatives advanced their three major crypto bills. This is Crypto Week. We'll talk more about that tomorrow. TSMC raised its revenue growth guidance for 2025 to 30 percent. And then the last thing I'll mention is we've got Netflix earnings tonight. I texted our media reporter, Sihel Patel, to ask him what he's watching for.

2:46And he said to me that I'm watching to see if they acknowledge the growing rivalry between YouTube and Netflix. You've got creator deals and sports league rights in the mix now. It is a fascinating rivalry. Watch for that tonight. I want to bring on a very special person, our Asia Bureau Chief, Jing Yang. She is in New York, all the way from Hong Kong. And just to give you a little bit of context here, Jing has broken stories about ByteDance, about Alibaba. She leads our Asia Bureau in their award-winning coverage. They're coming hot off a victory of what effectively is the Pulitzer Prize of Asia.

3:21Jing, thank you so much for being here with us. We appreciate it. Thank you for having me. So, look, ultimately what I want to get to in this conversation is the ground view of where AI startups in China are at. But very quickly, we've got to start with NVIDIA. I mean, what do you make of this new development that they're going to sell these chips? It's a bit of a cat and mouse game. We know how these regulations go and how they develop new chips around them. What do you make of the news this week and how are you thinking about it? Yeah, since you mentioned the cat and mouse game, maybe I'll just unpack the context for listeners a little bit.

3:57So NVIDIA has been in this cat and mouse game with Washington since about 2022. and basically every time the US export control updated its restriction, NVIDIA will then lower the specification of its GPUs so that it can sell to China again. And as a result, NVIDIA has developed several editions of China-tailored GPUs. And then H20 is the latest in that effort. And before it was banned earlier this year, It was the most powerful GPU that NVIDIA was allowed to sell to China, although it cannot even begin to compare with what NVIDIA is offering to customers elsewhere in its BlackRock edition. And let's just make it clear, what exactly is the fear from the U.S.

4:47government? The fear is that NVIDIA's chips are so powerful and it has been accessed by Chinese military forces. or allegedly, I should say, accessed by Chinese military forces to advance China's military capabilities. And then given fundamentally, the fears are rooted in the geopolitical tensions between the US and China over issues such as the Taiwan Strait tensions and the South China Sea, where China has all these territorial disputes with a range of countries in South Asia. right so then on that is sort of aware then you go from there to why there's like enough evidence and there's enough compelling evidence in Washington that you know Nvidia cannot be allowed to sell those chips to China however I also want to cite Nvidia's side of the story that they repeatedly denied that they don't have any evidence to suggest that their chips have been accessed by Chinese military forces.

5:55That is a state of play. And going back to earlier question about H20 resumption of the sale in China, that definitely surprised a lot of people, even people who I know have been watching this most closely. And then the broader contest in the resumption is that US and China obviously have been trying to reach some kind of agreement, some kind of grand bargain over trade and tech, right? And then this is sort of the first very significant progress that we've seen coming out of talks that have been held in the past months in Geneva, in London, et cetera. And there's a lot of discussion and speculation over what is going to happen next as the vital talks go on.

6:39Right. I mean, that's one question I wanted to ask you with respect to the cat and mouse game. I mean, do you think that this is going to last? I mean, you know, will there be new export bans that come in place or what's your sense of that? I think it remains to be seen, right? So remember, H20 is a part of the Hopper family of the GPUs, which NVIDIA has started to phase out as it tries to promote the latest Blackwell family of chips. And then so I think while H20 has resumed to be selling in China, actually the bigger question is will NVIDIA first develop another China tailorship out of its Blackwell offerings?

7:29And secondly, if it does, will it be allowed to sell those? So this would be a chip that's better than the H20, but conceivably not as good as the chips that would be available here in the U.S. It's actually hard to tell if it will be better than H20, because then again, it depends on whether Washington will update its export control guidelines again. Right, right. Let's talk about, you know, let's come back to the chip story in a second. And I want to move to the view from the ground with AI startups in China. I mean, look, your team was one of the first to report on DeepSeek and some of the developments coming out of that company.

8:16Where's DeepSeek at right now? They're developing their R2 model. They're having some challenges. Tell us a little bit about what you know. Yes. I'll talk about DeepSeek and then move on to other startups. So, after DeepSeek became a global sensation, obviously it has been recognized, it has been elevated to status of national champion, national treasure in China as well. And then we reported earlier this year that DeepSeek also has received, at the same time, the national treasure types of treatment. When I say that with the quotation marks, actually what I'm talking about is that, for For example, some of DeepSeek's key engineers have had their passport taken by the company, saying basically that's an implicit ban on employees traveling overseas at their well.

9:09And then on the other hand, any investors, even very powerful state-owned, state-backed investors that want to meet DeepSeek funder, want to just engage the company's interest, whether they want to receive outside funding, because bear in mind the company all the time grow fully funded by its hedge fund part. Then those meetings, DeepSeek actually cannot make a decision on its own whether they can't even entertain just like initial coffee. The investors have been told that prospective investors have been told that you need to go to to our local government officials to ask them, they will set the priorities about whether and who we should meet.

9:54And then what we see is that then R2, I think DeepSeek never promised a timeline for R2. However, there's been reports, I think initiative and waiters saying that DeepSeek is gonna come out in May, obviously that didn't happen. And then we're already in mid July and it still hasn't happened. We simply don't know. we know that DeepSix CEO and founder, Zamanfong, he's an idealist and professionalist. We reported that he's just not been happy with the performance of R2. And as the expectation keeps building, we're sort of in this really interesting tension where the longer you wait, I think the better you have to become.

10:38And then we also reported that a very important, I think something that people may not actually fully appreciate is that the kind of computational efficiency that Dixie achieved is because these engineers have become absolutely masterful at using NVIDIA's CUDA system. And that means that DeepSeek models are actually, you know, they're trained on NVIDIA chips, written in CUDA, which means that when you deploy the models, it's also the best to be deployed on NVIDIA chips as well. However, we have just talked about the whole chip shortage in China, right? So whenever DPSY R2 comes out, there will be sort of an immediate chip shortage problem that China has to face.

11:26Like when I say China, I mean, you know, cloud providers and any companies that want to access DPSY models. while you can run DeepSeq models with Huawei chips because it's not optimized for Huawei chips. You will lose something. It will not work as well as it's supposed to. I remember your team writing that story. It was very fascinating to me how the export controls were having that impact. DeepSeq is one startup. What about Manus? Manus is a very interesting case. So I would actually take a slight step back So when we're talking about the lay of land of AI startups in China, I want to break down to the foundation of the world of developers, the software application builders and the physical sort of AI or the robotics space.

12:15So we talked about DeepSeq and then in the application software, in the software application space, we see companies like Manus. So what do I mean by like Manus? So you see like they all, you have, we've seen a wave of agentic AI startups, all funded by Chinese engineers. Typically they have, you know, years of experience as working at the tech incumbents like Alibaba, Baidu or Bydance in China. And then they branch out on their own. And then they launch agentic AI applications, typically built on top of anthropic cloud models. However, they are all targeting the market outside of China, so the global ex-China market.

13:04And then the reason that they do that is because they are targeting not enterprise customers but consumer customers. And when you do that, you have to draw the line between, you have to walk the very tight rope because of a very tight censorship speech overall in China. So then they all chose to serve the global ex-China market. So you have this very interesting phenomenon that is bubbling up in the last three months, where Chinese-born and China-trained engineers founding agente AI startups to serve the consumer market in global ex-China. How is that going to shape up? We don't know yet, but it is very interesting.

13:50And there's several companies. Manus, one of them, there's Jenspog, Lovart. I can give you a very long list. They're all still at a very early stage, to be sure. So we don't know how they're actually going to fare. And then last but not least, moving on to robotics space. It's a big one. China has this world-dominating manufacturing supply chain prowess, right? which is why Apple had to make neighbors and its phones out of China before. And then so China has already like a sort of early, sort of a first move advantage if it wants to also become a dominant or important player in human robotics.

14:33And then there is also a lot of interesting development there. There's absolutely what we see in the U.S. in Silicon Valley that a lot of money has got into that space. The same thing has also been happening in China. And I should say, your team, along with Steve Levine, who authors our electric vehicles newsletter, you guys wrote a great story about the humanoid robot view on the ground, I guess, in China. It was a story that we'll link in the show notes. I'm wondering, as it relates to humanoid robotics, I mean, from your view in China, what is the most exciting development? What is the biggest challenge that that sector is facing there locally?

15:16Yeah, so I talked to several funders of China's meeting of other humanoid startups when I was working on that story. I think the biggest challenge is still, in terms of really cutting edge research when it comes to the intersection between AI and the physical application of that, China is still far behind the US. So then the biggest challenge for these Chinese entrepreneurs is that they want to work with either the American academia or talents so they can bridge that gap. And then they also feel like American companies and researchers should come to China as well, because you can do all the exciting research you want, but China is the place where you can execute, where you can see your research come true on an actual robot.

16:14And then the entire supply chain is not entire, but the best and the most economical supply chain is in China. I felt a very strong sense of wanting to collaborate with the U.S., with their counterparts in the U.S. when I was talking to these funders. I think that would be a challenge. Did you see any of the robots during your reporting? Yes, I did. Some of them are actually really impressive. What surprised you? I mean, tell me about the most eye-opening moment. There is one particular robot built by a Shenzhen-based startup. And then the CEO of that startup used to be a tenor, he's Chinese, and he used to be a tenor professor at Ohio State University.

16:59And several years ago, he went back to China to do this. And then the gait of that robot looks really, really good. I mean, I'm not an expert to actually measure all that tiny movements, but just simply by I look in, with my layman's eye, look in the gate performed by some of the best robots in the US. And by that one, I think they are pretty much equally impressive. Wow. I never thought we'd be talking about gate as it relates to robots. It's very fascinating to me. Very quickly, we got to go, but I just want to ask you one last question. What do you make of Jensen Huang's growing role as a political negotiator?

17:46This is something we saw in the news this week. What's your sense of how unprecedented that is, and what's your take on that? It is definitely unprecedented based on what Jensen did before and how he has been this year. Our colleagues Wen Ma and Chen Liu recently wrote a story about how Jensen has been playing an increasingly visible role in the political, in the highest political stages in both D.C. and Beijing. So in D.C., he's very well known, right? He went down to Mar-a-Lago a couple of times. And last week, he also met with President Trump before. And then what we saw as the result was the, on the back of that, was the resumption of a Chinese sale.

18:39However, and then we revealed in that story, by comparison, during the Biden administration, Jensen actually declined an invitation from Biden's team to have a dinner with him. And he instead chose to vacation on one of his homes, lecture homes. And then, and however, in China, the same thing is also happening. Remember several years ago when US-China relations were relatively still amicable, you often see American tech CEOs like Tim Cook and Elon Musk, rough shoulders with senior Chinese leaders, right? But Jensen was nowhere to be seen in those occasions. However, just this year alone, Jensen has already made multiple visits to Beijing that we know of.

19:23The latest one being earlier this week when just one day after NVIDIA announced resumption of Chinese selling China, Jetton delivered a speech at this international supply chain expo that Chinese government was putting on. And when he was doing that speech, he actually took off his iconic leather jacket. He put on a traditional Chinese suit. And then on stage, he was heaping praises of AI models and technology innovation coming out of companies like Alibaba, Tencent, Baidu, Dixie. And later when he was being interviewed by Chinese media, he said, And I quote, I'm very happy to see the development in China.

20:20Wow, I think that's a good place to leave it. Jing, thank you so much for joining us. We appreciate it. And we'll have to convince you to come back on the show, even though it might mean some crazy hours for you. We'll have you on very soon. I'm always happy to be here. Thank you. Thank you very much. For our next segment, I want to bring on our banking reporter, Michael Rodden, to talk about a big story that he published this week about a company called Sandbox AQ. He did an investigation into the company's developments and also how the CEO has been conducting himself at the company. I'm going to let him explain all the details.

21:01The best I can explain the company, he'll do it much better, is that this company sits at the intersection of AI and quantum computing, which means it's already a pretty interesting company. Michael, thank you so much for being here. We appreciate it. Yeah, thanks. So yesterday we published our second story on Sandbox AQ. And so stepping back, you mentioned quantum and AI. Is it two very futuristic technologies? Sandbox is right at the nexus of the two, or at least it's attempting to be. The company is chaired by Eric Schmidt, the former Google CEO. It's backed by an array of Silicon Valley's most powerful billionaires.

21:40So Salesforce CEO, Mark Benioff, early Facebook investor, Jim Breyer, Bridgewater Associates CEO, Paulville founder, Ray Dalio. They put about a billion dollars into this, along with some institutional money and an array of small investors. The whole show is sort of run by Jack Hittery, who's a serial entrepreneur. He's been doing things for about three decades after he dropped out of Columbia. He ran for mayor once. He's done charities. He's done financial services. He's done sort of everything. Last year, we took a pretty close look at the company and we reported that it had sort of embellished a little bit of customer success in its pitches to investors.

22:20And we sort of tried to get under the hood. It's a very complicated business. It's got a few streams of work. So it's got this thing that it calls large quantitative models or LQMs. So generally with AI, big AI companies, we know the LLMs, you know, ChatGPT or, you know, help me out perplexity. You know, you type in, it gives you an answer. That's the sort of consumer facing version of it. With LQMs, it's a little bit different. This is sort of like numbers based. It's like math and equation based. They're trying to help companies or researchers discover new chemicals, new molecules, new materials for manufacturing.

23:06Basically, very complicated, very new, and very deep tech. There's another line of work called AQ navigation, which is sort of using the Earth's magnetic fields to provide an alternative to GPS. So again, very futuristic. And you had some good reporting about who there, if it had any success selling this technology at all. I mean, tell us about the revenue findings you had. Yeah, so coming to this week, I've taken another really good look at the company. And basically, I got a hold of some pretty detailed numbers going into Sandbox's revenue. And so last time we wrote about it, we were a bit like, you know, we weren't really sure how to figure out what's real and what's not and how successful this company is in terms of generating cash, which at the end of the day, it needs to do in order to justify its $5.6 billion valuation.

24:05In the first quarter of this year, about 90 % of its revenue, which was only$10 million, came through entities funded by Sergey Brin. So there was two university contracts to research into these LQMs. Those were grants that were given to sandbox by universities, but they were ultimately funded by Sergey Brin, who was the Google president when Jack Hittery, the CEO of Sandbox, was sort of working with a bunch of researchers inside Alphabet that worked on a bunch of work streams that later turned into Sandbox. So there's a sort of a big question about how much customer demand there is for Sandbox's services and why the money is being funded by someone who's very close to the CEO and has been for a long time.

25:01The rest of the revenue, a very small amount, was made up of sort of trials by a range of customers. So things like airplane manufacturer, Airbus, or pharmaceutical company, Sanofi. But these are quite small in scope and size. So we've been on the hood of that company. I want to jump in here because your story dived into the CEO's spending habits. It also dived into the company's spending habits. You got some details on the company's revenue. I mean, Sandbox AQ is not a company that I have heard of before you started reporting on them. Help us understand what the implication of the story is. I mean, why should people really care about Sandbox AQ and this story?

25:49Totally. no and so it's like on the one side we've got the revenue not being as great as you'd think that the company would hope it to be and then on the other side there's obviously expenses and this sort of revolves around Jack Hulery who's this man about town he's he's flying everywhere he's he's on CNBC he's on Bloomberg he's going to Davos in Switzerland he's he's out and about he's meeting everyone and he's always sort of promoting this company and the image that he's presenting is one of great success. And I discovered that in order to fund that image of success, he's been spending pretty lavishly.

26:24And so we reported that he was investigated by the company after a complaint by an employee for pretty lavish spending on things like travel, entertainment, and including on women who had no relation to the company. So this is sort of like taking them places, whining and dining them. There was an instance where he took one to Wimbledon. And the senior employees at the company were pretty concerned about the way that the company was being managed. And so this all filtered up through a range of processes. And Hidori, we reported, was asked to pay back some of that money. But none of this was ever disclosed to investors.

27:10And I think that if If I were an investor in the company, these are some of the questions that I'd want to be asking about. Right. Well, look, it was a fascinating story. We have to go on to our next guest. But, you know, for those of you who don't follow Michael's reporting, he publishes very in-depth investigations and features like this. So he's definitely a byline to keep tuning into. Thank you so much, Michael, for being here. We look forward to having you on. Anytime. Thanks for having me. Okay. Yeah. For our next segment, we are going to a friend of the show. Sam Lesson is a general partner at Slow Ventures.

27:47The company counts Robinhood, Airtable, Solana and Postmates among its investments. He is a friend of the show because, yes, he is the husband of Jessica Lesson, our founder and CEO. And he joins us. Sam, thank you very much for being here. Excited to be on my first TI TV segment. We are excited to have you. Your face is everywhere, so we're excited to have it here now. Look, people talk a lot about late stage investing and IPOs, you know, the big flashy companies. We don't hear a lot about seed stage investing, which is something that you spend a lot of time doing. I mean, help me understand what the current state of early stage seed investing is right now in Silicon Valley.

28:27Yeah, look, I think the first thing to keep in mind, which is probably unsurprising to anyone listening to this, is it's a tale of two cities, right? There is what's going on in the AI world, which is highly spendy, big valuations, is a lot of energy and a lot of push to get sucked into that. And then there's kind of everything else, which is kind of sleepy, right? And I think that's kind of like what's going on at the seed stage market is, you know, when I talk to a lot of my friends who are seed investors, the kind of grudging conversation is always, like are we going to pay up and do these completely overpriced ai deals um which happen right but there's a pipeline for you know um or how disciplined are we going to be about going out and hunting things that are out of the market right um interesting and can become very good returns so i mean if there's not that much attention and it's sleepy then why isn't it like isn't there opportunity there yeah well that's the thing it's just hard to find i think The reality is seed investing historically, generally, it's very narrative driven.

29:27And if you think about why, it's because companies are born around some idea that people have. There needs to be a capital pipeline in general after them. So if you're going to do a seed investment, you kind of got to know where the next capital is coming from. There was a period, especially post-pandemic, where the story was very exciting about these things because it's so easy to develop software and these markets are interesting. People are going very deep on specific areas of expertise. And by the way, they can monetize quickly. So like maybe the answer is get profitable, right? And go from there.

29:56I think it's a very compelling narrative, but you got to find the right opportunities. It's not easy to do at scale. Then there's the other thing, which is clearly a strategy, which is the later stage investors have made it very clear that they're willing to pay up for AI narratives, right? And so because of that, if you can find the right opportunities early and get into them, you kind of know where your bread's going to be buttered next, right? Why the later stage investors. The problem is, again, it's such a spendy world in the AI space, and it's so easy to tell early stage stories of yada, yada, yada, AI, that one, from a seed investor perspective, you're not really getting paid for the risk you're taking up front, right?