I appreciate you picking this up—and the nuanced critique. I’ll give a brief response to your three points of pushback, but I’d be happy to discuss this further.
First, you call out the distinction between industrial policy creating the risk of bubbles versus the bubble itself being the policy. I agree that there is an important difference. As you note, and as I mentioned in my article, the anti-involution push (including the 2026 Government Work Report) clearly shows that policymakers recognize this as a real problem.
However, my claim is not that the government is trying to create a bubble to achieve its goals in the AI industry. Instead, I’m arguing that overcapacity and involution, while unintended side effects of the policy, can still be an important part of the causal mechanism through which China builds industries rapidly and at scale. Policymakers do not need to want the downsides of involution for these dynamics to be instrumental in accelerating industrial development, adoption, and application.
Second, you state that I oversimplify the government’s role in directing capital flows. Again, I agree. For brevity, I did leave out significant nuance. (I have a forthcoming paper that unpacks some of these details, so I am painfully aware that I didn’t unpack the complexity of how the state and the private sector interact in the capital allocation process.) However, my claim is not that the government chooses individual ventures, but that it plays an active role in shaping the ecosystem through capital allocation mechanisms. Even though the end decision-makers are often private venture capitalists, the state has a significant impact on the decisions that these private VCs make.
Third, you point out the ways in which NEVs and AI aren’t comparable. While there are significant differences between both the underlying technologies and the relevant industrial policies, I still believe there are many parallels that we can draw on. For one, while pricing doesn’t directly correlate with a corresponding uplift in competitiveness, it does have an effect.
Another point you raise is that companies don’t necessarily look for overseas customers when excess capacity emerges. Here, I disagree. While, as you note, China does have an enormous market for industrial AI applications, if domestic competition is strong enough and the barriers to entering new markets are low enough, it’s only logical for companies to go abroad. Leaving aside the government’s support for AI companies entering global markets, from a purely profit-driven perspective, companies have every incentive to seek international customers if they can make more money doing so.
Finally, I agree with your conclusion that if China wants to truly innovate at the frontier, it will need long-term capital that bears the innovation risk. However, to increase exports and push the AI+ adoption agenda, the current policy seems to be working pretty well.
Again, I appreciate you taking the time to respond. Hope to continue the conversation.
Every capital-intensive technology gets overbuilt on somebody's balance sheet. What's unusual here is that the paperwork decides whose, before the cycle even starts.
Worth asking how long founders keep signing once that is common knowledge.
Industrial AI policy is a parts-and-power story. The scarce edge is deciding which AI-adjacent bottlenecks still deserve capital when demos outrun supply.
Right, and the sequel is already on the record. Beijing spent 2025 running an anti-involution campaign to unwind the price wars its own solar and EV subsidies created, so I'd expect the AI version to end in forced consolidation rather than a market correction.
Ceci est lié à l'économie chinoise. La base de population est importante, et la poursuite des intérêts de l'investissement. Une fois que les fonds gouvernementaux tomberont, tous participeront, ce qui entraînera la situation où tous les employés s'emparent de la balle, atteignant un multiplicateur élevé en peu de temps, formant une bulle. Il n'y a pas de croissance durable et de développement stable.
I have a more general question. Let's start with Hayek's general observation that without a true price system (which precludes a state owned banking system) it is impossible to bring about an efficient allocation of capital. If this is true, then why does not China's huge state-sponsored investments in all sorts of spectacularly advanced technologies represent a gross mismanagement of the Chinese people's limited life savings, upon which they are depending for support in old age?
good hayek question, and the token dai program out of guangzhou this month is a partial answer. bank of china isn't lending purely on political priority there, the underwriting inputs are token output and consumption, compute contract value, compute receivables, actual paying-customer activity, not a central planning target. that's closer to a synthetic price signal than blind subsidy. doesn't solve what you're really asking though, who audits the reported burn, and burn gets easier to pad the moment it becomes collateral. still a genuinely different allocation mechanism than the equity guidance funds most people write about.
Thanks for your replying Yuzu Xu. I can't honestly say I understand what you are saying, but that's on me. (I'm really old!) In any case, I have great hopes for China and the Chinese people, but worry about their near-term future too. Here's a piece for the people of China : https://shorturl.at/Xi1kj
no need to apologize, i probably buried the point the first time. simple version, a bank in guangzhou started accepting how much ai compute a company burns as loan collateral, the way it'd normally take a house. it's a new kind of proof that a company is real and growing, not paper. appreciate you reading, and for subscribing
closer to the second. it's not property changing hands, it's collateral in the receivables sense, future compute revenue the bank can claim against if the loan goes bad. a house has resale value on its own. compute burn only has value while the company using it keeps operating, so it's really a solvency signal dressed up as security
Tsinghua's Institute for AI International Governance translated and responded to my article here: https://mp.weixin.qq.com/s/k5mC15yNN_0soB2qbjl-7w
I'm responding to their response below:
I appreciate you picking this up—and the nuanced critique. I’ll give a brief response to your three points of pushback, but I’d be happy to discuss this further.
First, you call out the distinction between industrial policy creating the risk of bubbles versus the bubble itself being the policy. I agree that there is an important difference. As you note, and as I mentioned in my article, the anti-involution push (including the 2026 Government Work Report) clearly shows that policymakers recognize this as a real problem.
However, my claim is not that the government is trying to create a bubble to achieve its goals in the AI industry. Instead, I’m arguing that overcapacity and involution, while unintended side effects of the policy, can still be an important part of the causal mechanism through which China builds industries rapidly and at scale. Policymakers do not need to want the downsides of involution for these dynamics to be instrumental in accelerating industrial development, adoption, and application.
Second, you state that I oversimplify the government’s role in directing capital flows. Again, I agree. For brevity, I did leave out significant nuance. (I have a forthcoming paper that unpacks some of these details, so I am painfully aware that I didn’t unpack the complexity of how the state and the private sector interact in the capital allocation process.) However, my claim is not that the government chooses individual ventures, but that it plays an active role in shaping the ecosystem through capital allocation mechanisms. Even though the end decision-makers are often private venture capitalists, the state has a significant impact on the decisions that these private VCs make.
Third, you point out the ways in which NEVs and AI aren’t comparable. While there are significant differences between both the underlying technologies and the relevant industrial policies, I still believe there are many parallels that we can draw on. For one, while pricing doesn’t directly correlate with a corresponding uplift in competitiveness, it does have an effect.
Another point you raise is that companies don’t necessarily look for overseas customers when excess capacity emerges. Here, I disagree. While, as you note, China does have an enormous market for industrial AI applications, if domestic competition is strong enough and the barriers to entering new markets are low enough, it’s only logical for companies to go abroad. Leaving aside the government’s support for AI companies entering global markets, from a purely profit-driven perspective, companies have every incentive to seek international customers if they can make more money doing so.
Finally, I agree with your conclusion that if China wants to truly innovate at the frontier, it will need long-term capital that bears the innovation risk. However, to increase exports and push the AI+ adoption agenda, the current policy seems to be working pretty well.
Again, I appreciate you taking the time to respond. Hope to continue the conversation.
thank you for including the Chinese phrases
Thank you, very interesting!
where else can we see the engineering principles of economic sector management akin to what they're doing now with AI?
real estate! china just soft landed its realestate bubble like a lunar lander by reducing lending to that sector.
and redirecting that lending capacity to the industrial and green sectors. increasing that by a huge multiple.
so instead of being rich by owning a speculative apartment people can be rich by having great jobs and everything they need being cheap.
the Chinese miracle that we have seen is nothing compared to what's happening starting now.
https://substack.com/@piglistener/note/c-289239443?r=48i0n
Every capital-intensive technology gets overbuilt on somebody's balance sheet. What's unusual here is that the paperwork decides whose, before the cycle even starts.
Worth asking how long founders keep signing once that is common knowledge.
Industrial AI policy is a parts-and-power story. The scarce edge is deciding which AI-adjacent bottlenecks still deserve capital when demos outrun supply.
https://paretoinvestor.substack.com/p/humanoid-robots-stocks-2027
Right, and the sequel is already on the record. Beijing spent 2025 running an anti-involution campaign to unwind the price wars its own solar and EV subsidies created, so I'd expect the AI version to end in forced consolidation rather than a market correction.
Ceci est lié à l'économie chinoise. La base de population est importante, et la poursuite des intérêts de l'investissement. Une fois que les fonds gouvernementaux tomberont, tous participeront, ce qui entraînera la situation où tous les employés s'emparent de la balle, atteignant un multiplicateur élevé en peu de temps, formant une bulle. Il n'y a pas de croissance durable et de développement stable.
I have a more general question. Let's start with Hayek's general observation that without a true price system (which precludes a state owned banking system) it is impossible to bring about an efficient allocation of capital. If this is true, then why does not China's huge state-sponsored investments in all sorts of spectacularly advanced technologies represent a gross mismanagement of the Chinese people's limited life savings, upon which they are depending for support in old age?
good hayek question, and the token dai program out of guangzhou this month is a partial answer. bank of china isn't lending purely on political priority there, the underwriting inputs are token output and consumption, compute contract value, compute receivables, actual paying-customer activity, not a central planning target. that's closer to a synthetic price signal than blind subsidy. doesn't solve what you're really asking though, who audits the reported burn, and burn gets easier to pad the moment it becomes collateral. still a genuinely different allocation mechanism than the equity guidance funds most people write about.
Thanks for your replying Yuzu Xu. I can't honestly say I understand what you are saying, but that's on me. (I'm really old!) In any case, I have great hopes for China and the Chinese people, but worry about their near-term future too. Here's a piece for the people of China : https://shorturl.at/Xi1kj
no need to apologize, i probably buried the point the first time. simple version, a bank in guangzhou started accepting how much ai compute a company burns as loan collateral, the way it'd normally take a house. it's a new kind of proof that a company is real and growing, not paper. appreciate you reading, and for subscribing
Like a new form of property then? Or just evidence that you are probably solvent? I still don't understand.
closer to the second. it's not property changing hands, it's collateral in the receivables sense, future compute revenue the bank can claim against if the loan goes bad. a house has resale value on its own. compute burn only has value while the company using it keeps operating, so it's really a solvency signal dressed up as security
and what about deepseek funding?
Wow