Odd Lots

China's AI push driven by state industrial policy and vertical integration, not just market forces

48 min Listen to the original Episode page
China TechIndustrial PolicySemiconductorsAI HardwareHuaweiEVsMarket Structure

Summary

Jill Weisenthal and Tracy Alloway host Kyle Chan, a Brookings Institution fellow, to discuss how China's tech ecosystem is reshaping around state-directed industrial policy. Chan explains that Chinese firms like BYD and Xiaomi achieve deep vertical integration by leveraging shared expertise across sectors, contrasting with the US model of outsourcing hardware. The conversation highlights Huawei's efforts to build an AI chip ecosystem independent of NVIDIA's CUDA, despite significant bottlenecks in lithography and high-bandwidth memory. Speakers note that US export controls have accelerated China's 'Sputnik moment' for semiconductor self-reliance, prioritizing long-term sovereignty over short-term consumer welfare or GDP growth. Finally, they address why Chinese stock markets underperform despite industrial progress, citing intense competition, low barriers to entry, and the commoditization of intelligence.

Key takeaways

  1. 1
    Huawei attempted to compensate for lower individual chip efficiency by bundling more chips together into 'superpods' to provide sufficient compute capacity for large pre-training runs.

    Jill Weisenthal · 20:33

Episode sections

  1. 0:00 – 5:35

    Intro, Sponsors, and China's Manufacturing Prowess

    Jill Weisenthal and Tracy Alloway discuss the trend of Chinese tech companies achieving vertical integration and self-reliance, noting that physical products are increasingly likely to be built in China. They touch on the tension between efficient manufacturing and low profit margins due to overcapacity.

  2. 5:35 – 17:05

    Kyle Chan Joins to Discuss Industrial Policy and Tech Ecosystems

    Kyle Chan, a fellow at the Brookings Institution, joins the podcast to discuss the nexus of firms, government, and industrial policy in China. He explains how companies like BYD and Xiaomi leverage overlapping expertise across sectors like EVs, batteries, and robotics.

  3. 17:05 – 21:20

    Huawei's AI Chip Capabilities and the CUDA Ecosystem

    The hosts discuss Huawei's Ascend chips, noting the historical difficulty in competing with NVIDIA's CUDA ecosystem. They highlight a shift where Chinese AI companies like DeepSeek and Z.ai are now co-designing hardware with Huawei to close the software gap.

  4. 21:20 – 23:23

    Production Bottlenecks: SMIC, Lithography, and HBM

    Kyle Chan details production constraints for Huawei, citing reliance on SMIC and limited lithography equipment. He mentions the 'Bring Your Own HBM' (BYOH) bottleneck and efforts by Chinese firms like CXMT to produce high-bandwidth memory domestically.

  5. 23:23 – 26:40

    Impact of US Export Controls and the 'Sputnik Moment'

    The speakers agree that US export controls have galvanized China's push for semiconductor self-reliance. This has led to a full-stack effort to replicate the global supply chain, including lithography machines, driven by priority status in China's five-year plans.

  6. 26:40 – 29:52

    Tension Between Frontier Performance and Long-Term Sovereignty

    A tension exists between Chinese AI firms wanting NVIDIA chips for immediate performance and the government prioritizing long-term supply chain sovereignty. The government discourages reliance on NVIDIA, pushing firms to use domestic alternatives despite the performance trade-off.

  7. 29:52 – 33:54

    Beijing's Tech Self-Reliance vs. Consumer Welfare

    Jill Weisenthal argues that Xi Jinping prioritizes high-tech self-reliance and future industrial dominance over traditional GDP growth or consumer stimulus. The government views the 'new quality productive forces' and complete industrial systems as the primary goals, with economic health for households being secondary.

  8. 33:54 – 37:02

    AI as an Enabler for Biotech and Manufacturing

    The speakers discuss whether AI will replace dedicated biotech efforts. Kyle Chan suggests Beijing sees AI not as a replacement for other sectors but as a turbocharger for drug discovery, healthcare, and manufacturing, aiming to infuse the physical world with robotics and autonomous systems.

  9. 37:02 – 40:45

    Huawei's Chip Production Challenges

    Jill Weisenthal asks why Huawei might fail in chip production. Kyle Chan compares catching the tech frontier to chasing a bullet train, noting that while Huawei improves, competitors like TSMC are also advancing. The main bottleneck is lithography equipment, with workarounds like logic folding and multi-patterning facing yield and complexity issues.

  10. 40:45 – 43:00

    Chinese Stock Market Underperformance

    The hosts discuss why Chinese stock investors rarely profit despite industrial progress. Kyle Chan attributes this to low barriers to entry, government financing, and intense competition leading to overcapacity and thin margins, making sustainable competitive moats rare except in specific cases like DJI or CSMT.

  11. 43:00 – 48:26

    Commodification of Intelligence and Market Limits

    Jill Weisenthal notes the CSI 300 index's poor performance, dominated by Maotai and CATL. She argues that AI's value may be limited by how much intelligence the economy can absorb, suggesting that while frontier models have margins, the broader market for intelligence may become commodified with little profit potential.

Assets discussed

AssetKindStanceContext
NVIDIANVDAequityneutralEarlier, we recorded an episode about the whole NVIDIA mega speed stuff and trying to wrap our heads a little bit around the um, I guess, ecosystem of data centers that serve Chinese companies, many of them based
AppleAAPLequityneutralFrom an American perspective, think thinking about big global industrial competition. the Xiaomi story I find to be a little bit blackpilling from the American perspective. Because when it when Xiaomi came out, that was when, you know, when we first heard about this company, Xiaomi, the iPhone was like in ascendant. But in the 2010s, at one point, Apple was looking for the
AmazonAMZNequityneutralI mean, we know like, you know, Amazon, they're they have their own chips that they design, train training, them.
MicrosoftMSFTequityneutralMicrosoft designs chips, OpenAI wants to design chips.
AlphabetGOOGLequityneutralAlphabet obviously has the self-driving cars, et cetera.
AlibabaBABAequityneutrallike Alibaba, which you know, you can kind of almost like level by level match up each layer of the

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Samsung Galaxy S26 Ultra features 'Super steady videos' and 'nitography' for low light picks.

Unidentified speaker (UNKNOWN) ·0:16

How to verify: Product specification sheet for Samsung Galaxy S26 Ultra; Bell promotional materials. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

BYD started as a battery maker.

Unidentified speaker (speaker_1) ·8:36

How to verify: BYD corporate history; Bloomberg Businessweek profile of BYD. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Huawei's Ascend chips lacked a CUDA-equivalent software library, creating a chicken-and-egg problem.

Unidentified speaker (speaker_1) ·18:47

How to verify: NVIDIA CUDA documentation; Huawei Ascend software development kit (CANN) release notes; industry reports on AI chip ecosystems. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Huawei is working with SMIC, China's premier foundry, which is limited by lithography equipment.

Unidentified speaker (speaker_1) ·21:43

How to verify: SMIC annual reports; semiconductor industry analysis on Chinese foundry capacity. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Chinese company CXMT is trying to move from DRAM into HBM production.

Jill Weisenthal ·22:45

How to verify: CXMT investor presentations; semiconductor market research reports on HBM supply. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Huawei told customers it could make NPUs but required them to 'Bring Your Own HBM' (BYOH).

Jill Weisenthal ·23:05

How to verify: News reports on Huawei chip supply constraints; industry analysis of Chinese AI hardware supply chains. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

US export controls had a 'galvanizing effect' on China's push for semiconductor self-reliance.

Jill Weisenthal ·24:02

How to verify: US Department of Commerce export control lists; Chinese government policy documents on semiconductor self-reliance. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

There is a rumored EUV lithography machine prototype in development in Shenzhen.

Jill Weisenthal ·26:03

How to verify: TechCrunch or Reuters reports on Chinese lithography development; patent filings by Chinese entities. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Chinese AI companies like DeepSeek and Z.ai are eager to buy NVIDIA H200 chips.

Unidentified speaker (speaker_1) ·29:26

How to verify: NVIDIA earnings call transcripts; news reports on NVIDIA chip sales to China. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

The Chinese government discourages reliance on NVIDIA GPUs to promote domestic chipmakers.

Jill Weisenthal ·29:32

How to verify: Chinese government statements on tech self-reliance; policy directives from MIIT. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Xi Jinping does not care much about the economy or consumers in the traditional sense, prioritizing high-tech self-reliance over GDP growth or stimulus.

Jill Weisenthal ·31:04

How to verify: Check Xi Jinping's speeches and policy documents from 2026-2026 for emphasis on 'high-tech self-reliance' vs. 'consumption stimulus'; compare with official GDP targets and real estate policy shifts. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

China's real estate collapse was a deliberate 'popping of a bubble' and a pivot towards a different economic model.

Jill Weisenthal ·31:25

How to verify: Review Chinese government statements and economic white papers from 2026-2026 regarding real estate regulation and the 'three red lines' policy. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The term 'new quality productive forces' is a buzzword used to describe the push for high-tech industries and complete industrial systems.

Jill Weisenthal ·32:04

How to verify: Search Chinese government policy documents and state media reports for the term 'new quality productive forces' and its definition. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Huawei is attempting 'logic folding' (advanced chip packaging) as a workaround for EUV lithography restrictions.

Unidentified speaker (speaker_1) ·39:11

How to verify: Review technical analyses of Huawei's chip packaging technologies and patents filed between 2026-2026; compare with industry reports on advanced packaging vs. lithography. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Multi-patterning techniques lower yield rates and cause more errors in chip manufacturing.

Unidentified speaker (speaker_1) ·39:51

How to verify: Consult semiconductor manufacturing textbooks or industry reports on multi-patterning challenges and yield impacts. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

DJI is a singular exception in the drone market with a sustainable competitive moat, unlike other Chinese tech sectors.

Jill Weisenthal ·41:22

How to verify: Analyze market share data for Chinese drone manufacturers and entry/exit rates in the sector from 2026-2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Maotai is the largest member of the CSI 300 index.

Unidentified speaker (UNKNOWN) ·43:00

How to verify: Check the current composition and weighting of the CSI 300 index on financial data providers (e.g., Bloomberg, Wind) as of late 2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

CATL is the number two component in the CSI 300 index.

Jill Weisenthal ·43:16

How to verify: Check the current composition and weighting of the CSI 300 index on financial data providers (e.g., Bloomberg, Wind) as of late 2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

The Shanghai Composite index has been essentially flat (horizontal line) since 2000.

Unidentified speaker (UNKNOWN) ·43:33

How to verify: Plot the Shanghai Composite Index performance from 2026 to 2026 using historical market data. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

China is investing heavily in developing its own lithography machines, specifically DUV immersion and potentially EUV.

Unidentified speaker (speaker_1) ·39:11

How to verify: Review Chinese government funding announcements for semiconductor equipment and reports on domestic lithography development by companies like SMEE. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify Huawei's Ascend chip performance metrics and software ecosystem maturity against NVIDIA's CUDA via independent benchmarks and developer documentation.
  • Check SMIC's yield rates and capacity utilization reports for advanced node production to assess the feasibility of domestic AI chip manufacturing.
  • Review CXMT's production capacity and yield data for High-Bandwidth Memory (HBM) to evaluate the viability of the 'Bring Your Own HBM' model.
  • Analyze Chinese government five-year plans and recent policy directives on 'new quality productive forces' to identify priority sectors for state funding.
  • Examine financial filings of BYD, Xiaomi, and Huawei for vertical integration metrics, including in-house component manufacturing percentages and R&D spending.
  • Assess the impact of US export controls on Chinese semiconductor equipment imports using data from the Bureau of Industry and Security (BIS).
  • Compare the valuation multiples and profit margins of Chinese tech firms against US peers to understand the 'commoditization of intelligence' claim.
  • Research the performance of the CSI 300 index and key holdings like Maotai and CATL to validate claims about market underperformance despite industrial progress.
Full transcriptexpand

0:00Jill Weisenthal: Finance is increasingly being built on digital rails. But as the adoption of virtual assets, tokenization, and stable coins grows, can regulation not only keep pace, but help accelerate it. Find out more later in the podcast.

0:16Unidentified speaker (UNKNOWN): Oh,

0:16Unidentified speaker (speaker_1): group shot. I'll take it. No, we're good. There's one in every group: the notographer. Best intentions, worst shots. But there's hope. The Samsung Galaxy S26. Ultra on Bell. Super steady videos and perfectly framed shots with my fan cam. And nitography for stellar picks even in low light.

0:34Jill Weisenthal: I want that. Yep.

0:35Unidentified speaker (speaker_1): Postworthy with just a click. All on Bell. Canada's fastest 5G plus network. Visit a Bell store or bell.ca slash Samsung for details. Bell. Connection is everything.

0:47Tracy Alloway: Bloomberg Audio Studios.

0:50Unidentified speaker (speaker_2): Podcasts. Podcast Radio News

1:03Jill Weisenthal: Hello and welcome to another episode of the Oddlots Podcast. I'm Jill Weisenthal.

1:09Unidentified speaker (UNKNOWN): And

1:09Tracy Alloway: I'm Tracy Alloway.

1:10Jill Weisenthal: Tracy, we're recording this on November 11, 11th, 2026. Earlier, we recorded an episode about the whole NVIDIA mega speed stuff and trying to wrap our heads a little bit around the um, I guess, ecosystem of data centers that serve Chinese companies, many of them based in uh Southeast Asia specifically. But like, you know, as we sort of we talked about it a little bit in the conversation that we did, is like there there's also this question of like, okay, for now, Chinese AI companies like they want to get access to NVIDIA GPUs one way or another. Is that always going to be the case? Or eventually will China, you know, we're seeing a little bit of, you know, China has Huawei in a sand. Like, eventually could it just be that this race to get American GPUs is not as intense as it used to be?

2:00Tracy Alloway: Yeah, I think when you look at China's technological development, I mean the long arc of history has kind of pointed towards vertical integration and self-reliance. Self-reliance sounds like a communist party slogan. Technological self-reliance. But I think it's true, right? And the thing with AI that we keep hearing out of China is this idea of diffusion.

2:21Unidentified speaker (UNKNOWN): Yeah.

2:22Tracy Alloway: Right. And so it's really interesting to see some companies that you wouldn't necessarily expect, they're not just like developing their own models, but like they're also developing their own components.

2:37Jill Weisenthal: Yeah. And this is so you know, yeah, because we were in uh Hong Kong earlier this year. I remember we did that episode with the CF of uh Baidu.

2:45Tracy Alloway: Yeah.

2:46Jill Weisenthal: You know, I don't think Baidu like has their own uh LLM. It's called Ernie. You don't hear about it very much. But like

2:53Unidentified speaker (UNKNOWN): I didn't know

2:53Tracy Alloway: it was called Ernie. I

2:54Unidentified speaker (UNKNOWN): forgot

2:54Tracy Alloway: about that.

2:55Jill Weisenthal: But like you don't hear about it that much. It's not one of the ones that gets like that much attention. But like here is a classic example of like, you know, they have a chip design business, they have a data center business, they have the search engine business, they have the LLM model, they have self-driving cars, this sort of uh full stack integration, so to speak. You know, when you said like the arc of history, the the what the thing that I think of is like if there is something that gets built, if it's a built product, there's a good chance that in the future it'll be built in China. Yeah, well,

3:28Unidentified speaker (UNKNOWN): you know what

3:28Jill Weisenthal: I'm saying? Like this seems to all anything that is physical, there's a good chance that either currently or at least other trajectory the world is going in right now, it will have been built in China.

3:37Tracy Alloway: But that also reminds me. so one of the main tensions when it comes to China manufacturing or technology production is like, okay, yeah, they build things really efficiently

3:48Unidentified speaker (UNKNOWN): and

3:48Tracy Alloway: cheaply and maybe better than a lot of other people do, but often they don't make that much money by doing so, right? Because you have the classic overcapacity problem where like the Chinese government will say, We want you all to go

4:01Unidentified speaker (UNKNOWN): develop your own chips

4:02Tracy Alloway: and then suddenly there's like a billion chip developers or EV manufacturers or whatever. And then also it seems like Chinese investors, I don't know if they've just gotten comfortable with like lack

4:14Unidentified speaker (UNKNOWN): cluster returns.

4:15Jill Weisenthal: I've read so many galaxy brain takes about why, like, or I don't know if they're galaxy brain takes, but takes about how you know all these companies are doing so well, but the stocks never go anywhere.

4:25Unidentified speaker (UNKNOWN): And

4:25Jill Weisenthal: it's like, you

4:26Unidentified speaker (UNKNOWN): know

4:26Jill Weisenthal: what, maybe I'll. just give up trying to understand that question. Can I say one more thing about China's manufacturing prowess? So uh I bought these headphones.

4:33Unidentified speaker (UNKNOWN): So

4:34Tracy Alloway: you need to do your China maxing

4:36Unidentified speaker (UNKNOWN): disclaimer.

4:37Jill Weisenthal: No, no, no. This is uh

4:38Tracy Alloway: your I went to China once and

4:40Unidentified speaker (UNKNOWN): now

4:40Tracy Alloway: I'm totally China Tech AI pills.

4:42Jill Weisenthal: I was gonna say the opposite, which is I bought these headphones. They broke recently. They seem to be as good as beats. Like they were like, oh, these are as good as beats. so they cost me like the equ in whatever the equivalent of $25. They broke.

4:55Tracy Alloway: That's interesting. I don't

4:57Unidentified speaker (UNKNOWN): I

4:57Tracy Alloway: have a bunch of random China made tech, and it's doing pretty I I showed you that back massager with the little hands. That thing, I use it like every night before I go to bed, and it's great. Oh,

5:08Jill Weisenthal: I have an eye massager.

5:10Unidentified speaker (UNKNOWN): Wait,

5:10Jill Weisenthal: I'm gonna say another thing too, actually. Did

5:12Unidentified speaker (UNKNOWN): your

5:12Tracy Alloway: eye massager break?

5:13Unidentified speaker (UNKNOWN): No,

5:13Jill Weisenthal: that still works, but you know, I bought um a pair of uh leaning running shoes.

5:18Unidentified speaker (UNKNOWN): Oh

5:18Tracy Alloway: yeah. They

5:19Unidentified speaker (UNKNOWN): sort of

5:19Jill Weisenthal: squeak.

5:20Tracy Alloway: Oh no, you said this

5:21Unidentified speaker (UNKNOWN): before.

5:21Tracy Alloway: Yeah.

5:22Jill Weisenthal: So it's like,

5:23Unidentified speaker (UNKNOWN): I

5:23Jill Weisenthal: don't know. I'm like, I think maybe I don't know. It's like maybe like, maybe, you know, maybe I should still just stick to some of these American brands.

5:30Unidentified speaker (UNKNOWN): All right, let's not. We've turned this into a product

5:32Tracy Alloway: review episode, which we didn't mean to do.

5:35Unidentified speaker (UNKNOWN): Let's get to the perfect guest. Let's talk

5:36Jill Weisenthal: to we really do have the perfect guest, sort of tell talking yourself. to some pretty high stake questions actually about future Chinese manufacturing chips, all of that. Someone whose work we've admired for a long time and have wanted to have on the podcast for a long time. And now it's timely. We're going to be speaking with Kyle Chan. He's a fellow at Brookings. He also has a newsletter called High Capacity, talks about all of these things. So, Kyle, thank you so much for coming on the podcast.

6:00Unidentified speaker (speaker_1): Great to be on.

6:01Jill Weisenthal: Just for those who don't know, we followed your work. for a while. Tell us a little bit about the type of work that you do. Why are we talking to you today?

6:09Unidentified speaker (speaker_1): Good question. Um, so I focus on Chinese technology and industrial policy. And I think um I started doing this before it was cool, I would want to say.

6:21Unidentified speaker (UNKNOWN): But

6:21Unidentified speaker (speaker_1): now it is very much in vogue.

6:23Unidentified speaker (UNKNOWN): Yes.

6:23Unidentified speaker (speaker_1): And the question here is not just sort of like looking at individual. technologies. I think we're all familiar with the rise of Chinese EVs or Chinese AI models

6:33Unidentified speaker (UNKNOWN): that

6:33Unidentified speaker (speaker_1): look almost as good as some of the American ones. But there's a bigger story underlying all this, which is sort of this nexus of the firms that are, you know, pioneering a lot of this technology, the uh government that is trying to promote this, the local governments and the central. government and the industrial policy activist like strategy to to make a dent in some of these areas and then to tie it all together. That's sort of like the the big picture question that I have, which is don't just see these as individual sectors and individual companies, but how do they actually have all these spillovers? Are those spillovers deliberate? How do the EV companies feed into the robotics industry? Or into the

7:15Jill Weisenthal: semiconductor industry and vice versa?

7:17Tracy Alloway: So, on this Nexus point, in some of your writings, you have these epic Venn diagrams. I don't know, do they even qualify as Venn diagrams when you have like 30 circles overlapping each other? But the point is that you have all these different industries or tech businesses in China that on the surface seem to be doing specific. things. So, you know, maybe one of them is doing autonomous vehicles, another one is doing EVs, another one is doing robots, semiconductors. But when you start to look at the relationships between them, you can see that they're actually overlapping. And I guess they're drawing on expertise in one area to develop other types of technology. Walk us through that particular ecosystem. as your very complicated Venn diagrams as you see it.

8:07Jill Weisenthal: So for those who are not looking at this very messy complex diagram, it has a bunch of these circles, right? A typical Venn diagram might have two. This has like more than a dozen. And they're different technologies, like electric vehicles, batteries, robotics, smartphones, consumer electronics, AI, blah, blah, blah. And they all overlap. So the point here is that they're companies. They're literally specific firms that sit at the intersection of these different sort of circles. There is BYD, for example,

8:36Unidentified speaker (speaker_1): who I think we mostly know as an EV company, but you know, they sort of got started as a battery maker, actually. And that's partly why they became such a powerful EV company.

8:45Jill Weisenthal: And

8:46Unidentified speaker (speaker_1): they also make their own chips. They're getting into autonomous driving.

8:49Jill Weisenthal: So

8:49Unidentified speaker (speaker_1): they're sort of one part of it. Also robotics as well. Xiaomi,

8:53Unidentified speaker (UNKNOWN): this

8:53Unidentified speaker (speaker_1): is the smartphone company. I

8:54Unidentified speaker (UNKNOWN): think to a lot

8:55Unidentified speaker (speaker_1): of people who are familiar with them, known for cheap smartphones and cheap consumer electronics, household appliances, stuff like that. Air purifiers. I used to

9:03Unidentified speaker (UNKNOWN): have

9:04Jill Weisenthal: a Xiaomi air purifier. That was, it was okay. It might have squeaked too. But then they jumped into electric vehicles. And, you know, some people are like, wow, you

9:13Unidentified speaker (UNKNOWN): know,

9:13Jill Weisenthal: Ladron, Lajan, the founder of Xiaomi. is such a visionary. He took this big bet. But also, he was drawing on a kind of supply chain foundation and a sort of pool of expertise and resources that already existed in China, partly due to these other

9:31Unidentified speaker (speaker_1): sectors that have, you know, uh flourished Flourish for for a while now.

9:34Jill Weisenthal: From an American perspective, think thinking about big global industrial competition. the Xiaomi story I find to be a little bit blackpilling from the American perspective. Because when it when Xiaomi came out, that was when, you know, when we first heard about this company, Xiaomi, the iPhone was like in ascendant. But in the 2010s, at one point, Apple was looking for the next thing, and they're like, oh, maybe we'll make cars. And then they didn't. And then fizzled. And then Xiaomi not only makes cars, but seemingly makes very good cars. And so you think like Apple could not go from phones to cars, but Xiaomi can, apparently all these other things. Now, just real quickly, like I'm looking at your Venn diagram, and American companies are kind of getting into like are they're not that different right now. I mean, we know like, you know, Amazon, they're they have their own chips that they design, train training, them. Microsoft designs chips, OpenAI wants to design chips. I think Anthropic wants to design chips, et cetera. Alphabet obviously has the self-driving cars, et cetera. How much of this would you confidently say this is a Chinese tech thing versus a in 2026, all big tech companies are kind of gonna have to compete at every level or want to compete at every level?

10:50Unidentified speaker (speaker_1): It's a bit of both for sure. Because as you point out, we see some of this happening in the US tech sector. Um you have like Google going into autonomous driving, robotics, making their own TPUs,

11:02Jill Weisenthal: you know,

11:02Unidentified speaker (speaker_1): doing this full stack approach. And some people compare them

11:04Unidentified speaker (UNKNOWN): to

11:04Unidentified speaker (speaker_1): like Alibaba, which you know, you can kind of almost like level by level match up each layer of the

11:10Unidentified speaker (UNKNOWN): stack.

11:10Unidentified speaker (speaker_1): What I think is different in the Chinese case are two things. One is just the number of companies that are doing this, right? It's not just like a small handful that are

11:19Unidentified speaker (UNKNOWN): uh

11:20Jill Weisenthal: doing this this full stack approach. It seems like even food delivery companies are jumping into making their own foundation AI models. And you have uh smartphone companies like Xiaomi making their own chips and foundation models as well. So that's one thing that's different. The other thing that's really different is that these are kind of like more full stack in terms of the actual manufacturing of the hardware itself, right? So ultimately, when Meta

11:47Unidentified speaker (speaker_1): makes their VR goggles, I think a lot of that is outsourced. The actual manufacturing is outsourced, including partly to China. Whereas some of these other companies in China, they're controlling the factories,

11:59Jill Weisenthal: they're

11:59Unidentified speaker (speaker_1): building even further further down to the components. So I think that's another um

12:04Jill Weisenthal: yeah

12:04Unidentified speaker (speaker_1): difference.

12:05Tracy Alloway: What would you argue are the driving forces for Chinese companies doing this? Is it just a sort of, I guess, natural progression? This reminds me very much of the Ricardo Hausman monkey

12:17Unidentified speaker (UNKNOWN): swimming

12:18Tracy Alloway: from trees thing, where you have like one industry and then it leads to the next industry and it sort of builds on itself. But then in China, you also have to your professional expertise, a lot of industrial policy where the ruling authorities will come out and they'll say, Hey, we don't want so many like digital internet companies or such an emphasis on them. We want everyone to move into making like actual hardware. And you get very strong signals from the Chinese government about what matters to them at any given point in time.

12:49Jill Weisenthal: Absolutely. Yeah. The way I would break it down is kind of demand versus supply side factors. By supply side, I mean like the resources, the supply chains, and the talent that exists to be able to jump into one of these sectors, especially if you are already in an adjacent sector. So that's your uh houseman monkeys jumping across different trees. I think that's a really apt analogy for what we see with Chinese tech companies themselves. They're jumping from tree to tree, not making a huge leap into quantum or something like that, but into kind of adjacent um overlapping industries. And then on the demand side, right, there are reasons to want to move away from a low margin industry like EVs and try to tap into maybe the next big thing, maybe robotics or some other fast faster growing market. And then of course, you know, as as you're pointing out, there is the government itself, right? Like I have a line where it's like, for a lot of things in China, including in China tech, it's the government stupid, right? That is such a crucial factor in all of this. And to ignore that would be to kind of miss, you know, the bigger political economy, which is really crucial in China. And so here you have signals from the government, both positive and negative. And I think um the positive signals include identifying priority areas that

14:09companies, that local governments, that ministries and agencies should focus on. There's a five-year plan that comes out and it literally lists out uh what some of these priority areas are, whether they're There, chips, AI, robotics, also sort of industries of the future, quantum fusion, biotech, brain computer interface, whatever Elon is up to, it's probably on that list. So you have sort of those positive signals, and that that tells the firms if you go in this direction, that will unlock a whole bunch of resources for you, whether it is local government support or whether it is special state-backed investment funds targeting those sectors. So you kind of want to swim with the current rather than against. And there's also negative signals, too, right? So a negative signal is a crackdown on the hedge fund industry in China. And in some ways, this is a bit of a tenuous link, but some people hypothesize that one reason why DeepSeek even started to go into the whole AI space was partly because they felt pressure being a traditional financial institution.

15:26Unidentified speaker (UNKNOWN): What's

15:26Jill Weisenthal: next for virtual assets? I'm Jacob Grease, host of Economy Scale, a series exploring growth in key global sectors. And in this episode, we're diving into the digital forces shaping finance. Sean McHugh from Debye. Dubai-based virtual assets registry authority has this to say. We're

15:42Unidentified speaker (speaker_1): seeing a lot of institutional players come in, digital asset dedicated hedge funds, high frequency traders, and that sort of thing, and folks who service institutional clients.

15:50Jill Weisenthal: As more capital enters the market, adaptive regulation is likely to become increasingly key. Merrick of Ripple says a lot of regulators globally would like to keep the industry at arm's length. That's the opposite here. I think also what you'll find is that regulation and frameworks can come out, but some of them are maybe a slight iteration on some 80-year-old policy and not actually in line with what the technology can do and offer. And again, that's very different in Dubai. Listen to the full episode wherever you get your podcasts. This series is brought to you by Bloomberg Media Studios and the Dubai Department of Economy and Tourism.

16:27Unidentified speaker (UNKNOWN): Oh,

16:28Unidentified speaker (speaker_1): group shot. I'll

16:29Jill Weisenthal: take it.

16:30Unidentified speaker (speaker_1): No, we're good. There's one in every group: the notographer. Best intentions, worst shots. But there's hope. The Samsung Galaxy S26 Ultra on Bell. Super steady videos and perfectly framed shots with my fan cam. And nitography for stellar picks even in low light.

16:45Jill Weisenthal: I want that.

16:46Unidentified speaker (UNKNOWN): Yep,

16:46Unidentified speaker (speaker_1): postworthy with just a click. All on Bell, Canada's fastest 5G plus network. Visit a bell store or bell.ca slash Samsung for details. Bell, connection is everything.

16:56Unidentified speaker (UNKNOWN): And

16:57Jill Weisenthal: I'm just looking at your uh Venn diagram ByteDance. Are they gonna do they have a smartphone? They were working with actually ZTE on this uh agenc smartphone. It had a little bit of controversy though, because um it was supposed to be able to operate apps on your phone as like a normal person would, but but

17:13Unidentified speaker (speaker_1): then and this is like a classic Chinese. story, they got blocked by a lot of the major Chinese platforms because they wanted their own control and they wanted

17:22Unidentified speaker (UNKNOWN): their own agents to

17:22Unidentified speaker (speaker_1): kind of take over like 10 cents.

17:24Jill Weisenthal: This is gonna be a big deal with all agentic stuff, which is no one wants the I mean, we've already there's already a knife fight happening with agentic commerce where no one wants to just be a Of passive door for someone. else.

17:34Unidentified speaker (UNKNOWN): This is like

17:35Tracy Alloway: the walled garden

17:36Unidentified speaker (UNKNOWN): stuff all

17:36Tracy Alloway: over again.

17:37Jill Weisenthal: There's gonna be a lot of that. Okay, so I'm looking at your megavend diagram, and in the middle is what like the one, like the s the beating heart of and that is in almost everything is uh Huawei. And obviously it's interesting that they're in basically everything that you have listed here, but it's particularly crucial when we're talking about AI. We know that one of the ingredients is chips, and we wouldn't even be talking about a company called Mega Speed getting NVIDIA chips. We wouldn't be talking about chip smuggling. We wouldn't be talking about export controls or any of this if Huawei had enough capacity to make good enough chips to um compete with NVIDIA. Why do you give us, like right now, your read on where Huawei is today, both in terms of their capabilities and their capacity to produce them and their scale?

18:27Unidentified speaker (speaker_1): Yeah. So up until now, the main problem was getting Huawei's AI chips, the Ascend series, to be reliable enough and functional enough to be used by Chinese AI companies who had grown up building on NVIDIA and the NVIDIA Kudo E Kuda ecosystem. So there was a long process where you did not have that entire sort of library of software that had been optimized for CUDA. You didn't have that equivalent for Huawei's sort of ecosystem. And they had been trying to build this up, but they had this sort of chicken and egg problem,

19:02Jill Weisenthal: which is that they didn't have the user base to contribute and build that up. And because they didn't have the libraries, they couldn't attract the user base. So they had this chicken and egg problem. But now, in part thanks to the export controls on um and restricted access on NVIDIA GPUs, there's been a push, especially from Beijing, to close the loop on this relationship and get the Chinese AI companies to work with the AI chipmakers, especially Huawei, to

19:30Unidentified speaker (UNKNOWN): sort

19:31Jill Weisenthal: of almost co-design future hardware. And in particular, you have a really tight partnership between DeepSeek DeepSeek

19:38Unidentified speaker (UNKNOWN): and

19:39Jill Weisenthal: Huawei, where they literally have personnel situated at each other's companies. They have a lot of intensive sort of technical back and forth trying to figure out how to

19:51Unidentified speaker (UNKNOWN): have

19:51Jill Weisenthal: these chips run, at least for inference for DeepSeek models. And you see actually now increasingly other Chinese AI companies announcing that their latest models, like um Z.ai's latest models, are you know on day one ready for up running on Huawei chips as well as other domestic AI chips in China. And then more recently,

20:11Unidentified speaker (UNKNOWN): we

20:11Jill Weisenthal: have this push for the training side, right?

20:13Tracy Alloway: So

20:13Jill Weisenthal: up until now, the Huawei Ascend chips were good enough for inference, but we're not quite good enough for training, especially for these really massive two, three trillion parameter models. And Huawei have been trying to improve not just their capabilities at the single chip level, but try to leverage their expertise as a networking technology company to bundle more of these chips together, to tightly connect them together, and to kind of make up for what they lacked in quality with quantity. So these chips were less efficient, they were less powerful on an individual chip basis, but you could bundle them together and you could start to build up these sort of superpods. And the goal there was to build up enough sort of compute capacity that was concentrated that you could run a large pre-training run successfully. And so that's the tipping point I think we're starting to see with the Chinese AI models. And there's uh rumors. that the next DeepSeek model, for example, will be fully pre-trained

21:19Unidentified speaker (UNKNOWN): on Huawei chips. Okay,

21:21Unidentified speaker (speaker_1): so that's sort of

21:21Unidentified speaker (UNKNOWN): like on the

21:22Unidentified speaker (speaker_1): capability side.

21:23Unidentified speaker (UNKNOWN): But then

21:24Jill Weisenthal: there's

21:24Unidentified speaker (UNKNOWN): a

21:24Unidentified speaker (speaker_1): production capacity

21:25Jill Weisenthal: side,

21:26Unidentified speaker (UNKNOWN): which

21:26Unidentified speaker (speaker_1): is another story altogether. Because basically,

21:29Unidentified speaker (UNKNOWN): even if you're trying

21:30Unidentified speaker (speaker_1): to do

21:30Unidentified speaker (UNKNOWN): this sort of quality for

21:31Unidentified speaker (speaker_1): quantity

21:31Unidentified speaker (UNKNOWN): trade-off, you do need

21:32Unidentified speaker (speaker_1): enough

21:33Unidentified speaker (UNKNOWN): quantity, you do

21:34Jill Weisenthal: need enough

21:34Unidentified speaker (UNKNOWN): volume

21:35Unidentified speaker (speaker_1): of chips in

21:35Unidentified speaker (UNKNOWN): order to

21:36Unidentified speaker (speaker_1): satisfy

21:37Unidentified speaker (UNKNOWN): all that demand. And at this point, Huawei is constrained. So Huawei works a lot with SMIC.

21:43Unidentified speaker (speaker_1): SMIC is

21:44Unidentified speaker (UNKNOWN): sort of like the

21:45Jill Weisenthal: premier foundry in China for chipmaking. They are kind of like the analog of TSMC, and they are limited by their equipment, by the lithography machines that they have, and by their own supply chains. So Huawei has been working closely with them, but they have limited capacity and a lot of their capacity. is actually allocated towards Huawei. Huawei itself has been leading its own efforts to build kind of a separate network of fabs, also trying to supply its own sort of chip needs. And that is still sort of ramping up as well. And so the big question is how quickly can you get that production scale up over time? And how limited is that by not just the number of fabs you have, but the equipment that goes into them by, for example, lithography machines, especially DUV immersion machines that largely at this point still come from ASML in the Netherlands. And then there's also the HBM bottleneck as well. You have Chinese companies like CXMT trying to move from DRAM into

22:45HBM. And while they still lag behind, you know, the big three, Heinrich Heinek, SAMS. Samsung, and Micron, they're trying to, in the next two to three years, trying to get to at least good enough HBM to then pair up with Huawei chips. But at this point, there's even been reporting that Huawei is saying to customers, like, we can make the NPUs, the neural processing units, but it's sort of B BYOH. Bring your own HBM. Oh. You're gonna have to source that from NPUs. elsewhere. So that's another bottleneck in the near term. Yeah. So that's sort of where we stand in terms of the shift in capabilities, but then also the bottlenecks on the production side.

23:23Tracy Alloway: This is sort of a tangential question, but on the other hand, I think it's actually a very fundamental question. When you look at the export controls around chips, um the US export controls, there was an argument when these first started being applied that this was gonna I think people called it the sort of like three body effect, that it was going to have the effect of speeding up China's own technological development and like lending even more of a sense of urgency to develop their own chip competitors. When you look at what's happening with Huawei chips now, does it feel like that's happened? That China has like had its kind of sputnik moment, I guess?

24:02Jill Weisenthal: Yeah, absolutely. So I think that US export controls had a massively galvanizing effect on China's push for semiconductor self-reliance. People who argue against that point to the fact that China has had a longer effort predating US export controls to try to develop their own sort of national chip industry. You have um national semiconductor plants, you know, going back into the 2010s. But what's really different here is sort of this full stack effort to build not just the chips, but the equipment, the materials, the photo resists, the etch and deposition equipment, to build up the entire supply chain. China's almost trying to replicate the entire global semiconductor supply chain in China, which I don't think any country has ever done before. And some indicators of like how big a switch this has been is um with. lithography. Lithography was one area that China had really deprioritized for a long time. They had not really invested in building up that capability, I think largely because they expected to be able to have access to lithography machines, you know, for the foreseeable future. And when that switch happened, when, for example, during the first Trump administration, you got the effective ban on EUV. lithography machines from ASNL to China. That was a huge wake-up call because that basically capped the ability of

25:21Chinese chipmakers to make chips of a certain process node, of a certain kind of scale and degree performance that was it was a huge moment. It basically kind of limited China's chip emissions as long as they were reliant on the rest of the world. And so from then on, you see this long-term effort, and building lithography machines in particular requires a very long-term sustained effort. But you have this long-term effort, and especially pioneered by Huawei, who, you know, not ironically was the main target of a lot of these expert controls from the US, to figure this whole thing out by hook or by crook, by trying to recruit engineers from all around the world, including from ASML, by investing. in uh the RD, in a couple of different sites. There's um one rumored site in Shenzhen, where there's supposedly an EUV lithography machine prototype in the works. So all these efforts are a sign that China's kind of like switched from, yeah, we should probably make our own chips to we must at all costs. And it's priority number one

26:23on the five five-year year plans.

26:25Tracy Alloway: You're You like set your sights on. being able to match global scale, and eventually you just grind and grind and you get there. And China has obviously done that in uh several areas and beyond just getting to the frontier, racing past the frontier batteries is one of the obvious categories that just comes to mind, just far ahead of the frontier. When it

26:44Jill Weisenthal: comes to AI models themselves, you know, I understand this impulse. It's like, okay, like use the Huawei chips and let's work together and build up this ecosystem. But at the same time, like that could potentially entail falling a little bit behind the frontier. And this is a brutal fight for the frontier that is literally measured in months in many cases. And we talk about, you know, the next best models being six months behind or three months. behind or whatever it is. Is there a tension with AI specifically where the Chinese model, and here I'm not talking about AI, but the Chinese model of like, let's work on our self-reliance, let's build this network, does have the risk of like, yeah, Uh, but what if we do that and now we're nine months behind? Or what if we do that and now we're 12 months behind and these sacrifices of not just trying to get the NVIDIA chips by Hooker by Crook are not worth making, even if it's good on a national basis.

27:47Unidentified speaker (speaker_1): Absolutely. The trade-offs here are really sharp. And there are trade-offs all over the Chinese industrial policy landscape. It's like one of the big themes that I focus on. And in this specific case, you have this trade-off between today. can you close the gap with the US AI frontier? Can you narrow that gap? Or are you focused on tomorrow being able to build a fully self-reliant semiconductor supply chain that may not be completely on par with what TSMC and NVIDIA can produce, but is good enough and will basically allow China to continue to make AI progress, regardless of what Washington decides. And I think here what's interesting is the tension between the firms and the government in China. And even across industry in China, there's different views, right? So there are the Chinese AI companies like DeepSeek or Z.ai or Alibaba who they they want to build better and better models. They would probably prefer to just keep building on NVIDIA chips if they could. They are very familiar with that. It doesn't require much extra work. So they're sort of optimized on running on that platform. But Beijing doesn't like that. And for the Chinese central government, the goal is not just to have really incredible models

29:06today. The goal is to build up that kind of independence and sovereignty, essentially, in supply chains in the long run. So you see this tension where. you have the Chinese AI companies clamoring to get the H200s that Trump allowed to be exported to China. They seemed very eager. They were putting in their orders, they were trying to uh negotiate how much they could get.

29:28Jill Weisenthal: And you have Beijing itself saying, hold on, hold on, hold

29:31Unidentified speaker (UNKNOWN): on. Like

29:32Jill Weisenthal: we don't want so many of these NVIDIA GPUs. We want to focus, we want you guys to focus on partnering. with domestic chipmakers in China to close that loop and to help them make their chips better. And of course, there's this agreement within the industry in China because the AI companies want to run faster on the models. But of course, all the chipmakers, you know, they love the export controls and they love Beijing's push for self-reliance because they have seen an astronomical surge in demand for their chips that didn't exist before.

30:01Tracy Alloway: Can you talk a little bit more about I guess, the long game here or what Beijing would regard as a success? Because it's very easy to imagine a scenario where China becomes very good at a bunch of new technologies, partly by integrating a lot of its existing firms or doing this sort of vertical integration that we've been discussing. But margins, you know, remain pretty thin. Maybe households don't feel particularly wealthier than they did in the past, which is something we've been talking about on the show as well. Like, would that count as a success um for Beijing if all those products were, I guess, integral to global supply chains?

30:44Jill Weisenthal: Yeah. I mean, I'll throw out a controversial take, which is Xi Jinping does not. really care that much about the economy or about consumers or about households in the sense that, you know, in the in the traditional sense that we might think about in the US or in the West. I think up until now, you have all these calls for an adjustment of, you know, imbalances, you know, why isn't China boosting uh consumption? Why aren't they offering, if not um, handouts to consumers, some kind of stimulus? measures to stimulate demand, right? There's a big demand slump, especially after the real estate collapse, which was a deliberate sort of popping of a bubble and a pivot towards something else. And the answer to all that is if you start with a premise that the goal is not GDP growth in and of itself, the goal is high tech self-reliance self reliance of sort of like winning the industries of the future. If you start with that premise, everything starts to make a lot more sense. So those who have been wondering, like,

31:45well, you know, is the stimulus coming? Are they gonna finally bail out, you know, whoever it is that needs to be bailed out in order to get demand going? That I think is ultimately seen as secondary downstream to this high tech push. And the idea is there's this very clunky term that has become sort of like the buzzword, which is the new quality productive forces. If you have these particular high tech industries, growth sectors, uh, industries of the future, if you can start to win in those areas and also have the supply chains, sort of the complete industrial system, which is like a term in Chinese that's very common, to back all that up, then everything else will kind of follow. You will end up having the kind of robust. uh economic growth. You will have jobs, you will have high-tech jobs, right? You know, and this shift away from sort of blue-collar factory jobs, you might have purple-collar jobs, jobs that require a more advanced degree and have more technical uh know-how, maybe uh maintaining and repairing robots, for example, rather than actually sort of making widgets on the factory floor. So to take a step back, basically, I see that there is this. sort of all-in push to try to become, if not the dominant, a dominant global player in a number of these key industries. And everything else is sort of like, you know,

33:04if if that works, that's great, but that's not really the main goal right now.

33:25Unidentified speaker (UNKNOWN): Oh,

33:25Unidentified speaker (speaker_1): group shot. I'll take it. No, we're good. There's one in every group: the notographer. Best intentions, worst shots. But there's hope. The Samsung Galaxy S26 Ultra on Bell. Super steady videos and perfectly framed. shots with my FamCam and nitography for stellar picks even in low light.

33:43Jill Weisenthal: I want that.

33:43Unidentified speaker (speaker_1): Yep, postworthy with just a click. All on Bell, Canada's fastest 5G Plus network. Visit a Bell store or bell.ca slash Samsung for details. Bell, connection is everything.

33:54Jill Weisenthal: So we did a few episodes a couple years ago about um biotech in China and how that's growing and expanding. But there is a view. in Silicon Valley that there won't be a biotech industry, that there'll be an AI industry and that the AI will find all the biotech. And that actually there might it may even be stupid to try to like invest in quote, what we call biotech per se today, that you should really just be building more advanced AI models because those models are going to be what finds all the molecules of the future, et cetera. And I'm sort of curious the degree to which either Xi or people in China have this view that's like, forget biotech, forget all this other stuff, forget quantum, just build the great AI model, and then the AI model will solve these other things in science.

34:43Unidentified speaker (speaker_1): Yeah, that's interesting. Cause at first I was gonna say Xi Jinping is for sure AI pilled. That is a really key priority. And maybe at the top of the list of sort of these more general purpose

34:54Jill Weisenthal: technologies that they think are gonna sort of fuel their growth and development going forward. But the way that you phrase it was really interesting because I don't think that

35:04Unidentified speaker (UNKNOWN): um

35:04Jill Weisenthal: he or the powers that be in Beijing see this as the be all end all. The

35:09Unidentified speaker (UNKNOWN): last

35:09Jill Weisenthal: technology. That's right. That's right. I think they want to integrate this into everything. else. They want it to ch turbocharge drug discovery and healthcare innovation and

35:21Unidentified speaker (UNKNOWN): manufacturing.

35:22Unidentified speaker (speaker_1): And they want to infuse sort of like the physical world with physical AI and have robotics and autonomous delivery systems everywhere. But whether that will just completely eclipse the traditional, you know, it's funny to have this term, like the traditional high high-tech technology. industry.

35:38Unidentified speaker (UNKNOWN): I know, I know

35:39Jill Weisenthal: it's

35:39Unidentified speaker (speaker_1): the future. But yeah, basically I I don't see that happening. And in fact, when you see the kinds of policy documents, the blueprints, the plans going

35:49Jill Weisenthal: going ahead, these different technology sectors are still laid out as like critical, you know, so-called future pillar industries, basically.

35:57Tracy Alloway: Yeah, but actually biotech is an interesting example

36:00Unidentified speaker (UNKNOWN): here

36:00Tracy Alloway: because like I I think it tells you something about how China actually approaches a lot of these technological development projects because like the original big biotech push that was mostly from an agrarian perspective, right? That was like we're gonna develop new technologies in order to boost food production.

36:17Unidentified speaker (UNKNOWN): Yeah, yeah, right.

36:17Tracy Alloway: And then it morphed into like biomedical stuff,

36:21Unidentified speaker (UNKNOWN): like

36:21Tracy Alloway: actual pharmaceuticals. And so you can envision now that like they probably don't think that models are gonna, you know, solve the entire biomedical thing, but like there absolutely will be an adaptation of like biopharma with AI. I'm

36:36Jill Weisenthal: sure Dario thinks the models will just come away

36:39Unidentified speaker (UNKNOWN): with it.

36:39Jill Weisenthal: Yeah, yeah.

36:40Tracy Alloway: Well, I mean, we've had interesting examples. So Citrini, who we had on the show, um has been working with one of the models to develop an anti anti-venom venom for scorpion biomes. bites, which apparently didn't really exist before. This was like for all different types of

36:55Unidentified speaker (UNKNOWN): things. And they're showing some

36:56Jill Weisenthal: progress on it.

36:57Tracy Alloway: Yeah, I think it's going into some sort of clinical trial. So, you know, it it's an interesting one for sure.

37:02Jill Weisenthal: Let's go back to chips for a second. You know, I sort of at this point still have this view that eventually, like the trajectory of the world is that if something is built, it'll be built in China. Now that may change and there's a lot of evidence. efforts to like fight against this, I don't know, extropy or whatever it is, but that is the trend. I'm curious, like in five years, if Huawei is not at the frontier of chip production and scale, et cetera, why not? What would what would be a reason to think that maybe they are not going to be able to do in chips what they've been able to do in many? other industries.

37:40Unidentified speaker (speaker_1): Yeah. So it depends on what you mean by sort of like not succeeding. Yeah. Because on the one hand, there is like not catching up to the frontier and sort of keeping pace. And in that way, I kind of compare

37:53Unidentified speaker (UNKNOWN): the

37:53Unidentified speaker (speaker_1): whole chip industry to like trying to catch a bullet train because you like trying to run as fast as possible and accelerate as fast as possible, well, but you're the the the goal that you're trying to catch or the frontier you're trying to catch up to is itself accelerating.

38:07Unidentified speaker (UNKNOWN): Yeah.

38:08Unidentified speaker (speaker_1): So you know, while Huawei is improving and while some of the other Chinese shipmakers are also sort of improving in and they're trying to catch companies like TSMC, which is just in investing like an incredible amount, and they're already at the frontier and they're operating with the best technology, the best equipment in the world from Japan and the Netherlands and uh Germany and the US. And so when you put that all together, it's that's a formidable sort of, there's not like a finish line that

38:31Unidentified speaker (UNKNOWN): you'll reach and you'll

38:32Unidentified speaker (speaker_1): say, like, I got to the global frontier. That finish line is also running away from you. But if you're asking about, you know, will they be able to have sort of like good enough chips that have decent enough supply where Chinese AI can sort of continue roughly at the pace that it's going at now? What would be the major barriers to that? Probably then it would be bottlenecks in the equipment in particular lithography. And so there is a huge sort of range of uncertainty around how effectively China can develop its own lithography, especially DUV immersion and then potentially EUV, or whether China might be able to bypass some. of that technology, right? There's been some hype around Huawei's efforts to do this sort of logic folding where they're arguing, you know, some people just say it's fancy chip packaging, but you know, they're trying to rather than just drive down the process node further and further, as if you had an EUV machine, they're trying to basically double back and and sort of repack re pat use use better packaging so that you can build. connections sort of within the chip itself. And that's supposed to be another workaround. There's other sort of techniques like multi-patterning where you kind of like redo the design multiple times. But that can get messier in terms of

39:51lowering your yield rate and causing more errors along the way. So there's sort of like workarounds. But ultimately I think the the key bottleneck is the lithography machines and whether China can can get there. And then those right? This is the thing that's like so wonderful and also so frustrating about the chip business, which is you can pick any one of these parts of the of the chip stack, like lithography, and it in and of itself has supply chains numbering in the hundreds, if not thousands. And so you know, for every ASML uh EV machine, there's going gonna to be a uh Trump lasers company, no relation to um the president. There's gonna be special uh lenses made by Zeiss that are the best in the world. And you know, there's always the stats about how they're so perfectly level. But every single one of those in and of itself is another choke point essentially in the whole sort of complicated global chip supply chain.

40:45Jill Weisenthal: What's your answer for why stock investors never seem to make any money in John? And actually we haven't talked about much on the podcast. I mean, it's been a very bad year for Chinese.

40:53Unidentified speaker (UNKNOWN): Yeah, it

40:54Tracy Alloway: really has.

40:54Jill Weisenthal: It's actually been a pretty brutal year. Surprise, surprise. Stock investors, they're like, no, this is the year. It's finally. What's your answer that you would give for why like all the progress, et cetera, stock investors never seem to make any money?

41:08Unidentified speaker (speaker_1): A big part of it is it's hard to make profits in in in China. It's hard to have a sustained competitive moat. There are some exceptions to that. So uh, you know, one company that I always find kind of interesting is DJI. Oh,

41:22Jill Weisenthal: yeah. Why

41:22Unidentified speaker (speaker_1): aren't there like a bazillion Chinese drone companies? How how did that market get so concentrated? You know, you have like CSMT, as I mentioned earlier, with uh with with memory and it has carved out a special space within that that sector. But in general,

41:36Jill Weisenthal: wait, do you have an answer for why DJI is singular? within the drone space, like the the way they are?

41:41Unidentified speaker (speaker_1): I actually don't know. It could just be a quirk of timing and we may actually start to see the flood coming. Right. So um there there's going to be new players emerging. But at least at this point, it's sort of uh you know it it's in in its own category. But more broadly you do have this phenomenon where it's hard to have a sustainable moat. You have other companies jumping into the mix that all also have access to, frankly, some of the best engineers, including your own engineers who can jump ship. They have access to financing from

42:09Jill Weisenthal: local governments. So the barrier to entry for a lot of these industries is actually relatively low. And so they end up getting kind of flooded with, you know, overcapacity, uh redundancy, but then also building these overall very, very globally competitive industries that once China starts to export to the rest of the world. well, good luck to the rest of the world. So I think that's that's part of the reason why why we see this effect. If you don't have long-term profits or don't expect that to really be sustainable, it's hard

42:36Unidentified speaker (speaker_1): to justify rising or uh high valuations.

42:40Unidentified speaker (UNKNOWN): All right,

42:40Jill Weisenthal: Kyle Chen, great chatting with you. Great to finally have you on the podcast. Let's stay uh let's stay in touch.

42:46Unidentified speaker (speaker_1): Absolutely. Thanks you guys

43:00Unidentified speaker (UNKNOWN): You

43:00Jill Weisenthal: know, Tracy, the uh the Chinese stock market sell-off is so bad that once again, and I I think it this was not the case, but once again, the Mao Tai company is the biggest member of the CSI 300 three hundred index. That's

43:14Tracy Alloway: a everyone's drinking. because stocks.

43:15Unidentified speaker (UNKNOWN): Everyone's drinking, yeah.

43:16Jill Weisenthal: No, like, you know, the number two is cattle. So that's

43:19Unidentified speaker (UNKNOWN): uh that's

43:20Jill Weisenthal: the um the battery company.

43:21Tracy Alloway: Yeah.

43:22Jill Weisenthal: And then there's some other but st we're back to the world where the uh the Baijo company

43:26Unidentified speaker (UNKNOWN): is the uh it is

43:28Tracy Alloway: amazing if you look at like a long term chart of the Shanghai composite since two thousand

43:33Unidentified speaker (UNKNOWN): You just don't make

43:34Jill Weisenthal: any money there.

43:34Tracy Alloway: it's like it's pretty much an uh almost a horizontal line. Yeah. I mean, there are some pretty big ups and downs, but like the overall direction of travel is

43:45Unidentified speaker (UNKNOWN): sideways. Sideways and better.

43:46Tracy Alloway: Sideways, yeah.

43:47Jill Weisenthal: And every once in a while, like there's a hype cycle. But I thought that was really interesting. This idea that like AI strikes me as a real tricky thing for the Chinese model of development. Because intuitively, it's like, yes, I suspect that like on enough time, absolutely, you would, you know, it seems that wow, we would get as good as NVIDIA and eventually you have to scale to produce as much as TMC on the long enough timeline. But in the meantime, the pace of improvement in the model layer is so fast. How long can you afford to go? with the homegrown technology if it's not good enough? And that's a very interesting question to me.

44:32Tracy Alloway: The other thing, we didn't even really get into sort of self-recursive improvement for model development. But that also seems like cliff isn't the right word, but like if the Western models get there before China, yeah. Like

44:48Jill Weisenthal: You can't afford yeah.

44:49Unidentified speaker (UNKNOWN): Like if

44:49Jill Weisenthal: if that's a thing. and we don't even know if it's a thing. Yeah. But if that's a thing, then it's like that really, I think, puts a limit on the degree to which any model maker can really it's like we're gonna go with the indigenous technology. Yes, it'll slow us down, but it's good for the nation. If there is this day in which you can get to like RSI and then it's off to the races, if that's real.

45:11Tracy Alloway: But the counterpoint to all of that is basically like, well. what are these AI models for? Right. Like

45:17Unidentified speaker (UNKNOWN): the

45:18Tracy Alloway: AI is supposed to help us like run our businesses more efficiently, produce things more efficiently, maybe make some technological breakthroughs here and there. And so

45:27Unidentified speaker (UNKNOWN): like

45:27Tracy Alloway: the focus on actual diffusion throughout the economy makes a lot of sense from that perspective. Whereas in in like in the US, it very much feels like AI is the sort of like be B the model itself is the be all and end all of development.

45:43Unidentified speaker (UNKNOWN): No,

45:43Jill Weisenthal: totally. I was talking to some folks the other day in finance. And they're saying like, look, like, there is a point where AI is good enough for us. As in, there is a point where we cannot absorb more intelligence per se. And at that, once you hit that point, we've you could can like you could make the models 10 times smarter We can't use this. Yeah.

46:04Unidentified speaker (UNKNOWN): And

46:05Jill Weisenthal: then what are you doing it for? And then the list of things that can actually absorb that intelligence starts

46:09Unidentified speaker (UNKNOWN): to shrink. You're a

46:10Tracy Alloway: widget manufacturer.

46:11Unidentified speaker (UNKNOWN): You're not going

46:12Tracy Alloway: to be like solving Navier Stokes

46:13Unidentified speaker (UNKNOWN): or something like that. You just

46:15Tracy Alloway: want to like automate a few things.

46:17Jill Weisenthal: Right. And this is if you were to make the bull case for the commodification of intelligence, it would be some version of most economic economically productive things in the economy can only absorb so much intelligence, unquote. And that the American closed source labs, like, yes, they might have big margins on the very frontier, but they're that's not that big

46:44Unidentified speaker (UNKNOWN): of a market. The

46:46Jill Weisenthal: market is smaller, and then the rest of it gets really commodified and there's no money. That's a very plausible way that I don't know that that's gonna be, it's just plausible.

46:54Tracy Alloway: All right, shall we leave it there?

46:56Jill Weisenthal: Let's leave it there.

46:56Tracy Alloway: This has been another episode of the All Thoughts Podcast. I'm Tracy Alloway. You can follow me at Tracy Alloway.

47:02Jill Weisenthal: And I'm Joe Weisenthal. You can follow me at the Stalwart. Follow our guest, Kyle Chan. He's at Kyle I Chan. Follow our producers, Carmen Rodriguez at Carmen Armin Dash. Show O'Bennett Bennett at dash Dashbot, bot Kale Brooks at Kyle Kale Brooks. and Kevin Lozano at Kevin Lloyd Lozano. And of course, you can chat about all of these topics 24-7 in our Discord, discord.ggodlots. And for more content, you can uh check out our daily newsletter. Please subscribe. That's over at Bloomberg.com slash oddlots.

47:30Tracy Alloway: And if you enjoyed this conversation, if you like it when we talk about complicated Venn diagrams, then please leave us a positive review on your favorite podcast platform. And if you're watching on video, then please like or leave a comment or better yet, subscribe. And

47:43Jill Weisenthal: remember, if you're a Bloomberg subscriber, you can listen to all of our episodes absolutely ad-free. Just go to the uh Bloomberg channel on Apple Podcasts and follow the instructions there.

47:53Tracy Alloway: Thanks for joining.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights