Another unfiltered conversation: the bloodline politics of AI talent; open source as strategy? token-maxxing; why OpenClaw hit China harder than Silicon Valley; future predictions
Would you have any recommendations for reading on how China funds deep tech development and commercialization - both from govt grants or private funding.
Would you say the gap between Chinese open-source models and top three US labs has been widening?
What will actually trigger the consolidation? It seems to me that the independent Chinese model companies will find it hard to generate enough revenue to sustain themselves - or to raise US-style rounds that would allow them to compete with Alibaba, Tencent, ByteDance..
Very insighful. Thank you Afra for this wonderful interview. I found several of her points both highly illuminating and deeply convincing.
First, AI is being embedded into two fundamentally different state-capital-industry systems. As a result, the same underlying technology is likely to evolve into two very different political-economic forms in China and the United States.
Second, the value created by the same amount of token consumption is not the same in the two countries. The U.S. is structurally better positioned to form closed loops around enterprise use cases with high conversion and clear ROI. China is different. It might be more likely that Chinese AI will first find value through consumer applications, entertainment, manufacturing, and industrial efficiency gains.
Third, Chinese market is likely to keep generating AI applications that Silicon Valley did not even anticipate, and China’s manufacturing depth will allow these applications to diffuse faster and penetrate more deeply.
That said, I do have reservations about several parts of the framework.
First, the “China OS vs. America OS” framing risks making the contrast feel too essentialized. It can leave the impression that the United States is naturally capital-driven, optimistic, and decentralized, while China is naturally local-government-driven, coordinated, and anxious. Reality is more complicated. The United States in recent years has also become more state-shaped, more industrial-policy-driven, and more security-oriented. I wrote two months ago that America itself is moving in a more state-capitalist direction (leonliao.substack.com/p…). China, meanwhile, is not driven only by local government competition. Its AI race is also full of market incentives, speculation, platform rivalry, and commercial calculation, and in my view market-driven forces are far stronger than local-government competition alone. At the same time, China’s local fiscal pressures, industrial policy priorities, compute support, outbound expansion considerations, and security concerns do shape the direction of corporate decision-making. Chinese companies are certainly profit-seeking, but in a strategic sector like AI, profit maximization is not the only objective.
Second, I strongly agree with Afra’s emphasis on manufacturing as the most important long-term strength of Chinese AI, but she does not yet fully develop the question of global spillover. Manufacturing scenarios can generate a large number of high-value applications, but whether these applications can evolve into global standards, global brands, and global profit pools on the scale of internet platforms is a different question altogether. China may be exceptionally good at embedding AI into industrial processes, but that does not automatically mean it will dominate the global software layer, platform layer, or protocol layer of AI.
Third, on open source, Afra argues that Chinese open-source AI still looks more like a competitive strategy than a fully validated long-term business model. I am less convinced on this point. Based on recent developments, the 2025 annual reports just released by MiniMax and ZP AI already suggest that open-source strategy is becoming an extremely explosive commercial model. In particular, OpenClaw has driven a surge both in API calls and in user growth for open-source models whose token costs are only about one-tenth those of closed-source models. That increasingly suggests open source is not merely a strategic weapon in competition, but also a model with real commercial force behind it.
Du Lei's explanation for why OpenClaw exploded in China is an interesting angle I hadn't considered. It makes me think of Dan Wang's argument in Breakneck: that 40 years of technology-driven prosperity has made Chinese society more receptive to new tech. China adopts the tech with less hand-wringing. Could that be another driver for OpenClaw's popularity in China? Does this make China "definite optimists" about technology (but pessimists about the world) in Thiel's 2x2?
Thank you, Afra, for bringing us yet another insightful article!
"The people benefiting from the AI boom, from this round of capital appreciation and market gains, are a small minority — small enough that they barely register as a statistically meaningful share of society. For the vast majority, this rally has no intersection with daily life. They haven’t actually made money."
This 100% highlights the current challenges China faces. After integrating AI into logistics, medical devices, and the automotive industry, it's important to ask the question: will this truly benefit the 90% or just the 10%? Specifically, will we see a rebound in employment rates and per capita disposable income? I believe this is the turning point that will determine whether China shifts from "definite" to "indefinite."
I’m glad you and Hua Han approached the viral spread of Openclaw in China from this particular angle. People need to realize that just a year ago, the concept of an AI Agent was still very foreign to the average person in China; most people’s understanding of AI was limited to chatbots.
However, the AI Agent concept didn't blow up overnight either. A key milestone (also mentioned in the article) was the Doubao Phone. Despite being short-lived, as a high-profile AI smartphone under ByteDance, it did a great job of introducing the concept of general AI to mainstream users. For many Chinese customers, it was the first time they realized AI could directly help them organize emails, build webpages, or even write code—and these features were easily accessible and ready to use.
The explosion of OpenClaw also owes much to its promotion on RedNote and Douyin (Chinese TikTok). These two massive social media platforms helped reach millions of users, allowing non-tech-savvy folks to grasp the concept of an AI Agent.
First of all Afra, congratulations on another compelling work of journalism.
I feel like a pessemistic ludite civilian (in a good way) and I haven't got my hands on a ten-thousand-GPU terminal either. Although, there was a $20,000 dollar display model for DIY model making at a retail store on which I clicked around on the Windows desktop like an idiot.
As Hua Han points out, "whose (economic) recovery is this?...people benefiting from the AI boom... are a small minority... For the vast majority, this rally has no intersection with daily life."
I remember the dot-com bust here in Silicon Valley. You would drive down Central Expressway and see all the burnt out derelict office buildings. Well, I think you would have noticed that recently it looks very deserted. One notable exception is Atlasian, for which the parking lot is full.
This just goes to show that the AI bubble here is very different than Web 1.0 or Web 2.0 which filled all those buildings with people sitting at their terminals.
I think its also worth mentioning that while Web 1.0 may seem outdated, it has only very recently been embraced by the government here in the US, even in Silicon Valley. Afra, you mention, "America continues to rely on email and webpage systems for government services." I think "continues" would be generous. Most of the junk software that the government buys bareley even works, assuming the intended audience is compentent enough to log on.
All that being said Web 1.0 still works. Im using it for this long comment post right now.
Web 2.0 on the other hand looks like it may finally come to a long overdue death, read NMDOJ. This may seem off topic but, there are so many parallels in this article to past tech bubbles that can't be ignored.
Du Lei brought up the example of Didi and Uber China, "The idea that one company swallows the whole AI landscape the way Didi swallowed ride-hailing overstates the parallel." For this, I think its necessary to go back to the begining.
David Pogue who recently released the book The First 50 years of Apple makes the case that the iPhone gave birth to Web 2.0 in the interview on All Things Considered.
NPR's Chloe Veltman: "...And the App Store."
"That single gesture launched entire industries — Uber, DoorDash, Tinder, Airbnb," said David Pogue
What he doesn't mention is that it wasn't the technology but the Silicon Valley funding machine that made that happen. There is an anecdote in the book Super Pumped by Mike Issac in which VC John Doerr recalls his pitch to Steve Jobbs for the first VC "App" funding. This is essentialy what started the App bubble which led to widespread adoption of the smart phone and then web 2.0. This is what Du Lei is refering to when hes says the VC's have, "concentrated authority to effectively determine who wins in a given vertical."
Another book that really gives us a visceral understanding of Sillicon Valley is Silicon Boys by David A Kaplan. I grew up around Bubb Rd. in Cupertino in the 1980's and we would see the way Apple employees dressed. Even though Steve Jobs may have resembled a hippy, and there were tie-dye shirts and Birkenstocks on Bubb Rd., the greed and hustle have always been here. Kaplans book with excerpts by Tom Wolfe, brings it all the way back to Fred Terman, who did at Stanford in the early 1900's what you describe with the Yao Class.
Kaplan's book was published in the late 1990's at the height of the bubble. That was Web 1.0 or what's now known as the era of the "dot-coms." At that time the big fish was Microsoft (Google is not a thing yet). The main competition was for internet browsers and portals. The news making the headlines was the anti-trust case brought against Microsoft for forcing Windows '95 users to use IE rather than Marc Anressen's Netscape Navigatior (now Firefox).
Turns out that the claims made then were very similar to what we're talking about here. Netscape was a paid product whereas IE came with the OS; MS could use it to gain market share by giving it away for free. I think that was the first big example in tech for what China's calling the Anti-Unfair Competition Law. The next obvious example was Uber's subsidies as mentioned by Du Lei, "By August 2016, Uber China was finished — its brand, operations, and data absorbed entirely by Didi." The book Super Pumped goes into in this in excruciating detail.
Hua Han said that China "encourages brutal internal competition as a mechanism for driving down costs and shaking out weak players." This sounds very uncivilized, like a bunch of Seagulls in the parking lot with a loaf of bread.
You describe "AI brain fry," as AI tools leaving "people more anxious, depleted, and cognitively frayed," and mockingly, "The cohort apparently most afflicted are Silicon Valley’s tech investors." Interestingly enough it seems that the whole AI bubble was spawned in psychotic hedge fund environments and the tech continues to be persued by people with no particular intrest in electronics. A great example of this phenomenon is represented in Ray Dalio of Bridgewater. Rob Copeland formerly with WSJ and now NYT published "The Fund" in which he chrolnicles the founder's manic persuit of AI with his Principles OS system. Incidentally, Bridgewater also had strong in-roads to China and even modeled its company's management techniques after the Politburo. Dalio's fortune can be traced back to the Vanderbuilt family, "blood-lines" indeed.
Very fun to see my hometown (Huangshan) gets mentioned! It is one of the best places for digital nomads in my "unbiased" view.
I've always been interested in the potential of open-source AI infrastructure, and China definitely has a higher probability of pulling it off and making full use of it. The question then becomes: what will the US do if that happens? Will we see a future where individuals and companies are essentially forced to pay an 'AI tax' and AI becomes public utility?
“You cannot talk about AI without talking about the nation-state’s body, its operating system, the way blood moves through it.”
Insightful.
Same with the Internet, the Internet is a Federation that organically arose from the organic and so Constituted Federation of America.
Would you have any recommendations for reading on how China funds deep tech development and commercialization - both from govt grants or private funding.
The most up-to date, intelligent, expert AI industry conversation online. It's long because it covers a lot of new ground.
thanks for the generous comment 🥹
Would you say the gap between Chinese open-source models and top three US labs has been widening?
What will actually trigger the consolidation? It seems to me that the independent Chinese model companies will find it hard to generate enough revenue to sustain themselves - or to raise US-style rounds that would allow them to compete with Alibaba, Tencent, ByteDance..
Very insighful. Thank you Afra for this wonderful interview. I found several of her points both highly illuminating and deeply convincing.
First, AI is being embedded into two fundamentally different state-capital-industry systems. As a result, the same underlying technology is likely to evolve into two very different political-economic forms in China and the United States.
Second, the value created by the same amount of token consumption is not the same in the two countries. The U.S. is structurally better positioned to form closed loops around enterprise use cases with high conversion and clear ROI. China is different. It might be more likely that Chinese AI will first find value through consumer applications, entertainment, manufacturing, and industrial efficiency gains.
Third, Chinese market is likely to keep generating AI applications that Silicon Valley did not even anticipate, and China’s manufacturing depth will allow these applications to diffuse faster and penetrate more deeply.
That said, I do have reservations about several parts of the framework.
First, the “China OS vs. America OS” framing risks making the contrast feel too essentialized. It can leave the impression that the United States is naturally capital-driven, optimistic, and decentralized, while China is naturally local-government-driven, coordinated, and anxious. Reality is more complicated. The United States in recent years has also become more state-shaped, more industrial-policy-driven, and more security-oriented. I wrote two months ago that America itself is moving in a more state-capitalist direction (leonliao.substack.com/p…). China, meanwhile, is not driven only by local government competition. Its AI race is also full of market incentives, speculation, platform rivalry, and commercial calculation, and in my view market-driven forces are far stronger than local-government competition alone. At the same time, China’s local fiscal pressures, industrial policy priorities, compute support, outbound expansion considerations, and security concerns do shape the direction of corporate decision-making. Chinese companies are certainly profit-seeking, but in a strategic sector like AI, profit maximization is not the only objective.
Second, I strongly agree with Afra’s emphasis on manufacturing as the most important long-term strength of Chinese AI, but she does not yet fully develop the question of global spillover. Manufacturing scenarios can generate a large number of high-value applications, but whether these applications can evolve into global standards, global brands, and global profit pools on the scale of internet platforms is a different question altogether. China may be exceptionally good at embedding AI into industrial processes, but that does not automatically mean it will dominate the global software layer, platform layer, or protocol layer of AI.
Third, on open source, Afra argues that Chinese open-source AI still looks more like a competitive strategy than a fully validated long-term business model. I am less convinced on this point. Based on recent developments, the 2025 annual reports just released by MiniMax and ZP AI already suggest that open-source strategy is becoming an extremely explosive commercial model. In particular, OpenClaw has driven a surge both in API calls and in user growth for open-source models whose token costs are only about one-tenth those of closed-source models. That increasingly suggests open source is not merely a strategic weapon in competition, but also a model with real commercial force behind it.
thank you so much for the thoughtful note. so many great points on the shortcomings of this piece.
Du Lei's explanation for why OpenClaw exploded in China is an interesting angle I hadn't considered. It makes me think of Dan Wang's argument in Breakneck: that 40 years of technology-driven prosperity has made Chinese society more receptive to new tech. China adopts the tech with less hand-wringing. Could that be another driver for OpenClaw's popularity in China? Does this make China "definite optimists" about technology (but pessimists about the world) in Thiel's 2x2?
LOL
Thank you, Afra, for bringing us yet another insightful article!
"The people benefiting from the AI boom, from this round of capital appreciation and market gains, are a small minority — small enough that they barely register as a statistically meaningful share of society. For the vast majority, this rally has no intersection with daily life. They haven’t actually made money."
This 100% highlights the current challenges China faces. After integrating AI into logistics, medical devices, and the automotive industry, it's important to ask the question: will this truly benefit the 90% or just the 10%? Specifically, will we see a rebound in employment rates and per capita disposable income? I believe this is the turning point that will determine whether China shifts from "definite" to "indefinite."
I’m glad you and Hua Han approached the viral spread of Openclaw in China from this particular angle. People need to realize that just a year ago, the concept of an AI Agent was still very foreign to the average person in China; most people’s understanding of AI was limited to chatbots.
However, the AI Agent concept didn't blow up overnight either. A key milestone (also mentioned in the article) was the Doubao Phone. Despite being short-lived, as a high-profile AI smartphone under ByteDance, it did a great job of introducing the concept of general AI to mainstream users. For many Chinese customers, it was the first time they realized AI could directly help them organize emails, build webpages, or even write code—and these features were easily accessible and ready to use.
The explosion of OpenClaw also owes much to its promotion on RedNote and Douyin (Chinese TikTok). These two massive social media platforms helped reach millions of users, allowing non-tech-savvy folks to grasp the concept of an AI Agent.
First of all Afra, congratulations on another compelling work of journalism.
I feel like a pessemistic ludite civilian (in a good way) and I haven't got my hands on a ten-thousand-GPU terminal either. Although, there was a $20,000 dollar display model for DIY model making at a retail store on which I clicked around on the Windows desktop like an idiot.
As Hua Han points out, "whose (economic) recovery is this?...people benefiting from the AI boom... are a small minority... For the vast majority, this rally has no intersection with daily life."
I remember the dot-com bust here in Silicon Valley. You would drive down Central Expressway and see all the burnt out derelict office buildings. Well, I think you would have noticed that recently it looks very deserted. One notable exception is Atlasian, for which the parking lot is full.
This just goes to show that the AI bubble here is very different than Web 1.0 or Web 2.0 which filled all those buildings with people sitting at their terminals.
I think its also worth mentioning that while Web 1.0 may seem outdated, it has only very recently been embraced by the government here in the US, even in Silicon Valley. Afra, you mention, "America continues to rely on email and webpage systems for government services." I think "continues" would be generous. Most of the junk software that the government buys bareley even works, assuming the intended audience is compentent enough to log on.
All that being said Web 1.0 still works. Im using it for this long comment post right now.
Web 2.0 on the other hand looks like it may finally come to a long overdue death, read NMDOJ. This may seem off topic but, there are so many parallels in this article to past tech bubbles that can't be ignored.
Du Lei brought up the example of Didi and Uber China, "The idea that one company swallows the whole AI landscape the way Didi swallowed ride-hailing overstates the parallel." For this, I think its necessary to go back to the begining.
David Pogue who recently released the book The First 50 years of Apple makes the case that the iPhone gave birth to Web 2.0 in the interview on All Things Considered.
NPR's Chloe Veltman: "...And the App Store."
"That single gesture launched entire industries — Uber, DoorDash, Tinder, Airbnb," said David Pogue
What he doesn't mention is that it wasn't the technology but the Silicon Valley funding machine that made that happen. There is an anecdote in the book Super Pumped by Mike Issac in which VC John Doerr recalls his pitch to Steve Jobbs for the first VC "App" funding. This is essentialy what started the App bubble which led to widespread adoption of the smart phone and then web 2.0. This is what Du Lei is refering to when hes says the VC's have, "concentrated authority to effectively determine who wins in a given vertical."
Another book that really gives us a visceral understanding of Sillicon Valley is Silicon Boys by David A Kaplan. I grew up around Bubb Rd. in Cupertino in the 1980's and we would see the way Apple employees dressed. Even though Steve Jobs may have resembled a hippy, and there were tie-dye shirts and Birkenstocks on Bubb Rd., the greed and hustle have always been here. Kaplans book with excerpts by Tom Wolfe, brings it all the way back to Fred Terman, who did at Stanford in the early 1900's what you describe with the Yao Class.
Kaplan's book was published in the late 1990's at the height of the bubble. That was Web 1.0 or what's now known as the era of the "dot-coms." At that time the big fish was Microsoft (Google is not a thing yet). The main competition was for internet browsers and portals. The news making the headlines was the anti-trust case brought against Microsoft for forcing Windows '95 users to use IE rather than Marc Anressen's Netscape Navigatior (now Firefox).
Turns out that the claims made then were very similar to what we're talking about here. Netscape was a paid product whereas IE came with the OS; MS could use it to gain market share by giving it away for free. I think that was the first big example in tech for what China's calling the Anti-Unfair Competition Law. The next obvious example was Uber's subsidies as mentioned by Du Lei, "By August 2016, Uber China was finished — its brand, operations, and data absorbed entirely by Didi." The book Super Pumped goes into in this in excruciating detail.
https://www.hsfkramer.com/en_US/notes/crt/2025-07/china-releases-the-2025-anti-unfair-competition-law 17 July 2025
Hua Han said that China "encourages brutal internal competition as a mechanism for driving down costs and shaking out weak players." This sounds very uncivilized, like a bunch of Seagulls in the parking lot with a loaf of bread.
You describe "AI brain fry," as AI tools leaving "people more anxious, depleted, and cognitively frayed," and mockingly, "The cohort apparently most afflicted are Silicon Valley’s tech investors." Interestingly enough it seems that the whole AI bubble was spawned in psychotic hedge fund environments and the tech continues to be persued by people with no particular intrest in electronics. A great example of this phenomenon is represented in Ray Dalio of Bridgewater. Rob Copeland formerly with WSJ and now NYT published "The Fund" in which he chrolnicles the founder's manic persuit of AI with his Principles OS system. Incidentally, Bridgewater also had strong in-roads to China and even modeled its company's management techniques after the Politburo. Dalio's fortune can be traced back to the Vanderbuilt family, "blood-lines" indeed.
Very fun to see my hometown (Huangshan) gets mentioned! It is one of the best places for digital nomads in my "unbiased" view.
I've always been interested in the potential of open-source AI infrastructure, and China definitely has a higher probability of pulling it off and making full use of it. The question then becomes: what will the US do if that happens? Will we see a future where individuals and companies are essentially forced to pay an 'AI tax' and AI becomes public utility?
I would love these in podcast/audio form to listen to on commute/work!
edit: Using the substack listen mode but would definitely prefer the real voices!
> So when OpenClaw arrived in China, it compressed an entire year of product evolution into a single moment.
I feel like there must be something more to this...
can you elaborate?
It feels post-rationalization. It answers what OpenClaw did do, but not 'Why OpenClaw', and not anything before, or later.
This salary range is wrong by a laughable margin: “a senior associate at a major New York law firm earns between $80,000 and $110,000 a year”
Cravath-scale, industry-standard big law starting salaries for first year associates are well north of 200k per year in 2026.
oh wow...thanks for the note!
Great comment.