Are Export Controls Working? Field Notes from China’s AI Frontlines
What this covers
May 8, 2026 — This episode of China's DeepSeek Moment series from the Asia Society Policy Institute's Center for China Analysis (CCA) examines how export controls are reshaping China’s semiconductor and AI industries, drawing on fresh, on-the-ground insights from recent study trips across Beijing, Shanghai, and Silicon Valley. The discussion explores how U.S. restrictions are shaping Beijing’s industrial policy, Chinese AI firms’ strategies, and the broader innovation ecosystem — its strengths and weaknesses, choke points, and emerging sources of leverage — and whether these measures are working as intended. As Washington and Beijing approach the Trump–Xi summit, the episode offers an inside-out view of the key levers each side holds in a high-stakes phase of tech competition. Moderated by Lizzi C. Lee, CCA fellow on Chinese Economy, it features Scott Kennedy, trustee chair in Chinese Business and Economics at the Center for Strategic & International Studies; and Paul Triolo, CCA honorary senior fellow on Technology. (51 min., 53 sec.)
Subscribe for more videos like this: http://AsiaSociety.org/YouTube --- Support Asia Society today: http://AsiaSociety.org/Donate --- Subscribe to our newsletter to stay connected: https://asiasociety.org/email-subscriptions --- Facebook: http://facebook.com/AsiaSociety Instagram: http://instagram.com/AsiaSociety Twitter: http://twitter.com/AsiaSociety LinkedIn: http://linkedin.com/company/asia-society/
#asiasociety #asiasocietypolicyinstitute #centerforchinaanalysis #china #aichips #uschinarelations
Source description (no synthesized summary yet).
US export controls on semiconductors and AI were designed to protect national security but have evolved into geopolitical tools that are now counterproductive, driving mutual escalation with China while achieving unclear strategic goals and causing widespread economic damage across global supply chains.
- Export controls have shifted from narrowly-targeted WMD prevention to broad technology dominance and industrial policy tools applied to rapidly-evolving technologies with static rules
- Both the US and China are now trapped in a tit-for-tat escalation spiral over critical technologies (semiconductors, AI, rare earths) with no defined acceptable level of interdependence or mutual understanding of limits
- The choke-point targeting model has reached the limit of its utility, imposing massive economic costs on both sides while achieving minimal measurable national security gains
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
China has developed technological strengths in critical minerals and other areas and has since 2020 quickly moved up the learning curve on how to implement export controls as an offensive and defensive tool, mirroring US approaches.
“the Chinese also have technological uh, strengths um, in critical minerals and other areas and they also now have to write regulations and and use these tools and and China has quickly moved up the learning curve of how to do export control since their law was issued in 2020.”
Export controls were originally designed as non-negotiable tools to prevent weapons of mass destruction and other harmful weapons from falling into the hands of terrorists and adversaries, but have been expanded to cover broader digital technologies with dual-use capabilities.
“the original idea of export controls is is that they were non-negotiable tools to protect the world from technology related to weapons of mass destruction and other really harmful weapons falling into the hands of terrorists uh and and adversaries that shouldn't have them. Um but over time, um we have expanded uh the coverage of of export controls uh to to be broader.”
The US approach of imposing controls unilaterally and threatening measures against allies if they don't comply has proven to be a disaster over the last 4 years and remains a major source of contention between Japan, the Netherlands, and the US.
“the US approach has been to impose controls and then and then and then threaten various measures of what against allies if they don't comply. Um and that has proven to be a disaster over the last 4 years. Um and and and and you know, it's still a source of major contention between uh for example, Japan and the Netherlands and the US.”
Informal letters (isinformed letters) from the US government to semiconductor producers have become a mechanism to quickly push manufacturers to restrict their technology use to specific potential customers, representing a significant innovation in export control enforcement during the Biden administration.
“it's existed before, but it's now much more common and Paul referred to this is the informality of this with these it is informed letters that go to uh, producers to very quickly push them to res- uh, restrict their technology use often to specific uh, potential customers.”
The foreign direct product rule represents a key US innovation in asserting extraterritorial jurisdiction over allied and other nations' technologies when those technologies are based on or derived from American tools or manufacturing equipment.
“because these technologies are available and not only in the hands of the US, the US has begun to assert extraterritorial jurisdiction over um our allies and others whose own technologies are in part based on using American tools. And so, I think people have probably now started to know this word foreign direct product rule, which is one ex- uh way in which extraterritorial jurisdiction is imposed.”
Chinese innovation in pharmaceuticals has followed a clear progression from providing raw chemical inputs (APIs) and contract manufacturing to conducting clinical trials to developing and marketing genuinely innovative drugs globally, demonstrating capability for frontier innovation.
“China's gone from uh and this is as you have written, um really you know, providing APIs and doing clinical trials and making generics to to now being a leader in in new innovative drugs that are developed in China uh and then marketed globally in partnership with with with uh global uh pharmaceutical companies.”
Both the US and China are now in a tit-for-tat escalation spiral with export controls that is corrupting the original export control system and making it function like any other geopolitical tool rather than a security-focused mechanism.
“we're now in a much different place what than we were 10 years ago. Both sides are using tools. They're potentially open for negotiation...it's basically a corruption of the export control system that we used to have and it makes it just like any other tool tool and and that's that's really dangerous cuz there are genuine national security concerns”
The goals articulated in the October 2022 export controls (to stop China from developing high-performance computing or AI) were vague and incompletely vetted, unlike the clear universal agreement on preventing weapons of mass destruction proliferation.
“the October 2022 controls, for example, were were implemented and conceived by a very small group of people within the Biden administration that that didn't really, I think, vet the the purpose and the goal and and fully, um, with a broader group of people or the industry...the goals in, uh, as articulated in October 2022, were like were were pretty vague. It's like, you know, stop China from developing high-performance computing or AI...With WMD, everybody can agree, uh, that, you know, a certain country shouldn't get a a weapon of mass destruction.”
The October 2022 US export controls defined advanced semiconductors at 16 and 14 nanometers for logic, but this definition was outdated even as the rules were implemented, and by 2026 Chinese companies are operating far beyond these thresholds, rendering the controls ineffective.
“the the definition, for example, of advanced semiconductors, whether it was memory or logic, was outdated. And now here we are 4 years later and there has been no change in that...industry did not believe that that the the line drawn in 2022 at 16 and 14 nanometers for logic was advanced at the time.”
Many Chinese auto companies are not making profits despite technological competitiveness, due to excess capacity and inventory issues hurting company bottom lines.
“there's a significant challenge of other being too much capacity and and too much inventory which is which is hurting companies bottom lines and there's not a lot of these Chinese companies that are making a profit but but technologically speaking they they look extremely impressive”
China's AI and robotics sector is very dynamic with tremendous capital lining up to support the system, and the state is providing backing, subsidies, and favorable policies while companies seek to tap capital markets more systematically through Hong Kong, Shanghai, and foreign investors in unrestricted sectors.
“the takeaway here is that there's a tremendous now amount of capital that's that's lining up to support the system in China...the Chinese system now is in these key areas like AI and robotics semiconductors is is a is again a sort of evolving system where the state is providing backing and some subsidies and some some some favorable policies but there's a great a great desire to tap capital markets in a more systematic way whether it's Hong Kong whether it's Shanghai”
Recent Chinese company listings in Hong Kong and Shanghai (ByteDance, Threads, Jurpu, Moonshot, and soon CXMT and YMTC memory leaders) reflect the Chinese government's preference to avoid too much foreign investment in companies critical to its technology self-reliance goals.
“over the last 6 months we've seen Beer and More Threads, Jurpu, you know, Moonshot is going to go go soon. We've seen a tremendous amount of listing of Chinese companies...because the Chinese government is eager to not not have too much foreign investment in companies that are so critical to its its its goals of whatever self do you want to call it self-reliance”
AI is fundamentally different from nuclear weapons because AI is needed by every sector for economic growth, making it inappropriate to apply the same export control logic used to prevent WMD proliferation, as Jensen (CEO of Nvidia) noted: 'Nobody needs nuclear weapons, but everybody needs AI.'
“AI is not like nuclear weapons. Uh, I think Jensen said, you know, 'Nobody needs nuclear weapons, but everybody needs AI.' So, AI is inherently different, um, and it's going to undergird all future economic growth, uh, in many sectors of the economy.”
Both US and Chinese export control systems have both produced multiple innovations (extraterritorial jurisdiction, informal letters, trading of controls, rare earth restrictions), but the fundamental record of effectiveness of these tools on chips and AI has not yet been discussed and should inform the need for reform.
“I think there's a a variety of different things that that we should do. Uh I also think it uh hopefully in the conversation we'll talk a little bit about what the record has been um of the specific use of these tools with regard to, you know, chips and and and and and AI cuz I think the need for reform is should also be predicated partly on what what has been how effective or ineffective has have these innovations been.”
Any time a Chinese company wants to move up the technology curve, which all companies do through R&D, it triggers a reaction from the US export control system, creating a dynamic where there is no limitation or attempt to rethink how export controls should evolve with technology.
“anytime a Chinese company wants to move up the technology curve and every company, by the way, is always has an R&D capacity and is always looking to move up the the technology curve to be more more competitive, this this this triggers some reaction from from the US export control system. And so that dynamic in itself is really dangerous in the sense that there's no there's no limitation or there's no no attempt to re re rethink um how these the export controls and around this very dynamic technology should evolve”
The October 2022 export control rules on semiconductors were designed by a very small group within the Biden administration without broad vetting with industry or other government agencies, limiting input on effectiveness and implementation.
“the October 2022 controls, for example, were were implemented and conceived by a very small group of people within the Biden administration that that didn't really, I think, vet the the purpose and the goal and and fully, um, with a broader group of people or the industry.”
At the Beijing Auto Show, Chinese car makers demonstrated improving battery and range technologies, impressive car designs, and significant presence with large crowds, while foreign players (VW, Honda, Ford, Buick) showed they had gotten competitive models, indicating significant technological competition across all sides.
“At the same time they're also, you know, foreign players there who were showing that they had actually gotten the signal or gotten the message and that they now had competitive models as well. And so VW, Honda, Ford, Buick, there were they all had had new cars there to show and lots of of interest from from people there. So I think what we're seeing now is, you know, technological competition engaged on all sides.”
China continues to be a two-track economy with macro slowdown and weak consumption alongside continued technological progress, new Chinese AI models, and impressive AI diffusion across industries.
“China still appears to be having sort of a two-track economy, right? There's the the macro slowdown still seems quite obvious level of consumption etc. Is it still weak there's prices of stabilized but they've only stabilized because of higher price higher oil prices because of the conflict in the Middle East so there's been no turnaround in the macroeconomy yet that's still quite challenging and heard more about that this time than even even before it still seems really quite problematic at the same time continued progress in technology you know there's new Chinese models there's new impressive ways in which AI is being applied to various Industries”
The choke-point technology approach of targeting critical supply chain vulnerabilities has reached the limit of its utility because it imposes massive costs on both US and Chinese companies with no clear sense of where escalation stops, and should be moved away from entirely.
“I think that that model is has not proven to be really, really damaging and and counterproductive...if you're going to start targeting all choke points uh on both sides, the the result is this huge disruption in supply chains globally uh and damage for US companies, damage for Chinese companies. Again, with no clear uh sense of where this stops...the choke point technology uh uh approach approach, you know, has sort of reached, I think, it's it's sort of it's it's it's useful life, if it if it ever was useful”
Paul recommends removing memory controls entirely and updating end-use controls to 5 nanometers (considered the industry threshold for 'advanced'), eliminating the current 16-14 nanometer standard, to better align static rules with dynamic technology reality.
“I would remove the memory controls. I think the end use controls, of course, as I mentioned earlier...should be eliminated or at a minimum updated to something like 5 nanometers, right? Because that's, you know, below 5 nanometers, that's what's generally considered advanced in the industry.”
The Meta-Manus deal killed by China raises questions about whether China has decided to pull up the drawbridge and prevent its technology from leaving its control, or whether it is a short-term reaction to US-China relations and a summit bargaining chip.
“uh the Meta Manus decision uh raises a whole variety of questions about uh the extent to which China has decided to basically uh pull up the drawbridge in terms of now that it has technology, not letting it out of its control. Um and so, that's a question is is this a signal uh about that? Um or is this a short-term reaction to other uh developments in US-China relations and ahead of the summit China needing a bargaining chip”
A return to something like the Wassenaar Agreement, which allowed multilateral consensus-based updates to export control rules each year, would be preferable to the current unilateral US approach with arm-twisting of allies, as it was a slower but consensual process that allies participated in.
“what what should happen is a return to something like the Wassenaar Agreement, which allowed, in fact, every year tried to make changes to the to the to the rules around technology as technology advanced. And it was a yes, it was a bit of a messy and slow process, but it also was a consensual a consensus process that our allies participated in and preferred to participate in.”
Engine AI, a Chinese robotics company, reported to Paul that they are under significant pressure to go public within 3-4 years to support R&D and provide investor returns, in contrast to Western companies that have the 'luxury' of years of R&D before profitability.
“we talked to Chinese companies like Engine AI that and they say they're under a lot of pressure, for example, to go public to to they said in the West companies have the luxury of three or four years of R&D and raising money and they said in China, you know, the the the the pressure now is to is to is to raise money and then go public”
There is still inertia and desire within the US system for continued export control escalation and use of these tools despite stated preferences for detente, suggesting institutional rather than purely political drivers of escalation.
“there's still sort of an inertia, I would argue, um in this desire for uh you know, export controls and and escalation dominance and and other things that have been the hallmark of the relationship for the last, you know, six, seven, eight years.”
Scott recommends better coordination with allies on dual-use technologies, noting that allies have their own regulatory systems and views, and that without sustainable collaboration, export controls will not be implementable.
“we also need to better coordinate with allies and coordinate not just simply arm twist. They have their own regulatory systems, their own views about their technologies. And of course if we don't collaborate in in a sustainable way, these won't be implementable.”
China's innovation progress has historically not been done in isolation; globalization has been critical for talent, funding, and market opportunities, and when China talks about self-reliance, they really mean wanting to be at the top of the food chain where others depend on them more than they depend on others, not autarky.
“the record is that Chinese innovation progress has not been done in isolation. That actually in every industry where the Chinese have made progress, globalization has been critical for for talent, for for funding, for market opportunities, uh being part of the global innovation ecosystem has been really critical for for the Chinese. Of course, and and when China talks about self-reliance, really what they're talking about is they want to be at the top of the food chain. They don't want to be autarkic.”
Manus was not a developer of core AI technology but was packaging tools around Western models, initially not drawing much government attention until an investor disclosed finding a way to take Chinese AI companies offshore, which triggered angst in propaganda departments and brought in the NDRC.
“Mana is not a a developer of core technology. They don't do models. They're sort of packaging tools around Western models...one of the key investors uh in Mana spoke spoke uh in a in a some kind of a podcast or on social media that suggested that that they had found a way to to you know, to take Chinese AI companies offshore. Um and this generated a considerable amount of angst within the Chinese system. Uh particularly, you know, places like the propaganda department.”
Both the US government and Biden administration officials refuse to acknowledge that US export controls on semiconductors directly caused China's retaliatory controls on rare earths, instead attributing them to Trump-era tariffs, preventing honest discussion of mutual causality.
“there's still in my view and and and recent meetings, no acknowledgment of this really from the the the the Biden administration officials who implemented these controls. They they seem to want to blame the tariffs last year from the Trump administration”
Meta/Manus deal was actually completed with money and technology already transferred, so attempting to 'unwind' it would require forgetting technologies already learned—making the Chinese termination announcement essentially symbolic rather than operationally reversible.
“in this arrangement uh the deal's long since been completed, money has changed hands, technology has changed hands. You can't really unwind the actual consequences of financial and technological of of the deal. Um so, figuring out how to how to actually do that is is going to require a lot of magic, I think. It's going to require people to like forget the technologies that that they've learned.”
Memory chip export controls have had huge impacts on the competitiveness of US companies that provide leading tools for the memory sector by preventing Samsung and SK Hynix from rapidly expanding capacity in China, which in turn prevents US companies from acquiring enough memory for PCs, laptops, and smartphones.
“the control goals that were put in place in October 22 on memory are having a huge impact, for example, on the ability of play key players in that sector, like Samsung and SK Hynix, which have major facilities in China, to expand capacity in response to market demand...that's impacting US companies, for example, who can't acquire enough memory in the PC space, laptops, smartphones because of the the inability to rapidly ramp up memory capacity that's caused by export the export control situation.”
The Manus case signals a broader shift: China is moving away from allowing foreign ownership of Chinese AI companies, and the days of VIE structures (where Chinese companies set up offshore entities with foreign investment) like ByteDance's Cayman Islands structure are ending.
“we're sort of now in an era where foreign ownership of Chinese AI companies for example is going to be a bigger issue...the days of ByteDance going offshore to the Cayman Islands and setting up a VIE structure and and inviting a large foreign investors to own parts of the company are gone. ByteDance is probably going to have to unwind in fact some of that foreign investment.”
Scott Kennedy recommends eliminating the bargaining of export controls for market access or other concessions and ensuring that policy is based on solid evidence about items, end users, or other factors.
“eliminating the bargaining uh of of being able to horse trade an export control for market access or or some other kind of thing. And being really careful so that when we take actions, they are really based on solid evidence. Whether it's about an item or an end user”
The Huawei Hong informality letter was an example of the 'system coughing up' new export control challenges before the Trump-Xi summit despite hopes for detente, with Chinese response still being formulated in Beijing.
“the system keeps coughing up um you know, different different challenges...the system keeps coughing up um you know, different different challenges because there there's still sort of an inertia...the is informed letters were about uh a particular Chinese foundry Hua Hong...that could provide yet another potential sort of flash point before the summit.”
Singapore investors in Paul's discussion group cited the US entity list (companies sanctioned by the US government) as a positive list of companies to invest in because they have been vetted by the US government.
“We had a lot of Singapore investors in our group and they were all very excited about investing in China. And one of the key things they said was the entity list as a for them is a is a list of companies to invest in because they've been vetted by the US government.”
Linkerbot is a Chinese robotics startup that manufactures robotic hands for various purposes, discussed a valuation of $20 billion at the Shenzhen AI and robotics fair, exemplifying the startup landscape.
“Linkerbot, for example, which was at the the Shenzhen AI and robotics fair. They they just talked about a valuation of $20...guess what they make hands, robotic hands. We saw all of their many robotic hands for many different purposes”