Deciphering the U.S.-China AI Showdown
Hello. Welcome to this webinar brought to you by the Asia Society Policy Institute's Center for China Analysis. It is the third episode in this series where we delve into the intricate intersections of technology, geopolitics and global affairs in the realm of A.I.. Today, we're navigating the complex landscape of military AI, US-China competition, Huawei's recent 5G breakthroughs, the effectiveness of U.S. export controls, etc., etc. And we'll be talking about how to build a balanced framework in the world of regulation. I'm your host, Lizzi. And joining me are two amazing expert guests. First up, we have the distinguished Paul Scharre. Paul Scharre is the executive vice president and director of studies at CNAS. Paul Scharre is award winning author of "Four Battlegrounds: Power in the Age of Artificial Intelligence". Paul, it's fantastic to have you here.
Thank you. Thanks for having me. And joining us alongside Paul Scharre is the equally distinguished Paul Triolo. Paul Triolo is a senior associate with the Trustee Chair in Chinese business and economics at the Center for Strategic and International Studies.
Paul's views and writings are frequently quoted in international media outlets. Paul, thank you so much for being here with me. Great to be here. Now, as you might have noticed, we have two Pauls with us. So to keep things friendly and clear, we will refer to them as Paul S and the Paul T throughout our discussion. So with that bit of housekeeping out of the way, let's jump right into our conversation. I should mention to our audience that you are free to use the webinar chat function to raise questions. We will leave 15 minutes at the very end of the webinar to address your questions. If your question is posed to a specific expert, please make that clear in the question.
So for the first part of our of our discussion, we're going to turn to military and diplomacy. And here I'm going to turn to Pau S. Many of our audience will remember the recent diplomatic discussions between President Biden and President Xi concerning the responsible use of military applications. So, Paul, can you just give us some insight into how technologies are currently utilized in military applications and why the integration is considered a game changer in modern warfare? Well, thanks so much and very excited for the discussion today.
There were certainly some some tantalizing hints prior to the Biden-Xi meeting about a potential agreement between the US and China on guardrails on military use of AI not coming to fruition as part of the dialog. But the US and China are talking, which is really important because this is an urgent area that needs more attention and certainly both the United States and China are working hard to adopt AI into their militaries, as are other countries and non-state groups around the world. AI technology is pretty widely available, and so we already see, for example, the use of image classifiers that are based on machine learning, that can use machine learning to identify images and objects in video feeds being used in drones in Ukraine. So you don't need to necessarily be one of the world's most advanced militaries or a leader in A.I. to use AI, technology and military. But there's some tremendous opportunities to do so, to process information faster, to make better and faster decisions.
And there's a lot of value in the US China looking for ways to manage some of that competition because there are risks in using AI systems that are unreliable and risking potential accidents or escalation and making sure that we don't see countries engage in a race to the bottom on safety, where they're deploying military AI systems before they're safe. Fantastic. So let's move on to China's military development. Can you give us a concise overview of the current state of development, the potential challenges, as well as implication for global security? Yeah. So China is very keen to use AI to improve its military. We see them doing experiments like large drone swarm demonstrations and doing experiments using AI fighter pilots. Oftentimes, if the US military does a major demonstration, we'll see China follow suit not very long after. And in Chinese writings, they talk about the importance of intelligent rising warfare, about using AI and other digital technologies, computer networks, big data, cloud computing to process information faster. I think China, like other militaries, are trying to figure out what do you do with these technologies to use them most effectively. And that's one of the things that stands out in the history of warfare, is that oftentimes new technologies proliferate pretty quickly. But what matters most is figuring out how do you use this most effectively. And that that pattern of innovation is probably going to be one of the decisive factors in which militaries are going to stay ahead.
Fantastic. And, you know, in the realm of military A.I., are there channels or what are the channels for the United States and China to establish international norms in agreements that both sides can commit to? We know our military to military communication have been on hold for quite a while before the discussion between President Biden and presidency. Well, certainly direct communication is really key. So it's encouraging to see progress in that domain. And hopefully AI is an important area of conversation. But there are also international forums.
There's obviously a tremendous amount of international discussions going on underway about A.I. broadly. The Bletchley Park Declaration at the UK summit very recently that both the US and China sign, you know, China had rolled out their own global governance initiative. But on the military use in particular, there was the responsible AI military summit, the Rim summit just last year. And there's another one coming up. So it's another opportunity for both nations to then shape global norms and expectations either independently or ideally together about how I should be using the military. Fantastic. Thank you so much, Paul for your insights. Shifting our focus a little bit, we're going to turn to the second part of our discussion where we talk about the relative strength between U.S. and China.
There are media reports suggesting that China and the US are actually nearing parity in development. Paul T, I'm going to turn to your expertise there here. While the U.S. leads in fundamental AI technology and research, China is actually rapidly implementing them in practical applications. So what factors drive this divergence in the style of U.S. development, and what are the potential implications?
Great question, Lizzi, and I'm not sure I would couch it quite like that. I think, you know, companies in both countries have been deploying A.I. for a long time in various types of applications. So I think it's important to just step back quickly and define what we mean by A.I. here. In 2017, I wrote an article with Kai-fu Lee, where we tried to look at China and the U.S. and where they were in AI development, and we broke it down into several different categories. Internet based A.I., things like recommendation algorithms, business and enterprise focused A.I., things like logistics and business operations,
perception A.I., which is things like facial recognition, object recognition and then autonomous A.I., which, which is things like A.I. being used as part of autonomous vehicle systems for example. So in all those areas we looked at who, which country was leading, which companies were leading. And China, of course, very early on companies like Alibaba and Tencent were using AI algorithms for things like e-commerce transactions and also in things like logistics. So Chinese companies were very, very early users of AI algorithms and very application related processes.
Perception AI too. China, Chinese companies were leaders. Chinese researchers had done a lot of research and and participated in national conferences on perception AI. Things like natural language processing. And then on autonomous AI, we gave the US a little bit of a lead here. So we tried to get under the hood and get away from just AI. I think if you look at sort of across the A.I. stack, you look at things like data, access to data and then the sort of hardware area. That's a that's a tricky one.
Kai-fu has said, you know, talk about China leading in data. That's a complicated issue. But yes, they have a lot of data sets that that may be unique. Each of the companies, for example, developing AI in China has access to big datasets. Baidu has, you know, 20 years of search history. Tencent has a lot of transactional data. Alibaba has a lot of e-commerce and logistics data. But I think the interesting thing is as we get this to the current situation, which is where the big focus is on general and large language models, we're sort of in a little bit of a different world.
And there, yes, arguably U.S. companies like open AI are leading in the sense of of of innovating and designing new approaches to generative AI. Transformers for example, came out of the US the US research community here. so in that sense the U.S. research is leading but Chinese companies are quickly you know coming up to speed on ideas in these areas and generative AI both in terms of deployment of applications and in terms of sort of improving on existing models and developing their own models.
There's something like 238 large language models now under development in China, which is a large number. And certainly companies like Baidu with Ernie bot that they just put out, it compares favorably, for example, in terms of a performance with GPT3.5 and even ChatGPT 4. So generative AI is sort of a new area where it comes out of the natural language processing and image processing, bu still the use cases and the applications are still under development. And so trying to assess who's leading in that is is a bit of a challenge. On the hardware side, though, I think obviously the U.S. and US companies in terms of developing advanced GPUs are far ahead of the game and that's what we'll talk more about that when we get to the export control piece of it and why that's such an obsession of the U.S. government in terms of controlling Chinese company access to GPUs. But clearly Chinese companies have been trying to catch up in the hardware development space, but that's really challenging.
And Nvidia and AMD and Intel, for example, have a big lead there, and particularly in things like software developer environments around AI, which which is part of the critical part of the AI stack. Chinese companies are lagging and most developers use PyTorch and TensorFlow. But you know it come out of Facebook and Google, for example. And so it depends on what you're talking about, what part of the AI stack and how quickly the Chinese companies have been able to sort of come up the learning curve. So when you talk about who's leading and I have to be very careful, I think, and quickly get under the hood and look at some of the individual applications and processes. But definitely the general consensus in 2017 when I wrote that first paper is that US companies tend to be leaders in sort of basic research and innovation in the space. But I think that's a general comment. But I think that's changing fast.
Chinese researchers are very good in these areas. And then, of course, in terms of applications using AI across the board. Chinese companies have been very outfront in all of these. And generative AI, finally, is is the area where it's hard to say who's leading.
And that's something that that will unfold over the next couple of years in terms of actual applications. Fantastic. And we'll be talking about China's strategy to catch up, so to speak, on hardware and other fronts. But before that, Paul T, can you also shed light on some of the key industrial or commercial applications of edge technology that China is currently prioritizing. I've heard a lot of talking about AI diagnosis, AI powered automonous driving, etc., etc.. So what are those key areas of commercial applications that you can see in China where investors are now focusing in terms of China's landscape? Yeah, great question. And you know, again, going back a little bit, this has been an evolving sort of situation in China. The first wave of applications did tend to be in things like e-commerce and in the facial recognition area.
Of course, public security in China was was interested in in robust facial recognition systems. And so a lot of investment in money went into those companies. But I think now we're sort of in a different wave here where we're looking at companies actually need to make money now and also to figure out applications that will drive revenue.
So, yes, one of the areas I would say is is medical diagnosis. And that's that's been, again, around for a long time. But Chinese companies have been particularly good in this regard. So just this week, for example, there was an interesting paper released, that talked about a AI application for detecting pancreatic cancer. And it's an interesting it's interesting sort of data point here. It was it was written by a whole range of authors from from including from Chinese medical researchers. And Alibaba's Damo Academy was one of the lead authors. But there were participants from Harvard, Johns Hopkins and a university in Prague. So that shows that particularly in that kind of application of medical diagnosis, it's a very global and cooperative field where A.I. researchers in the US and China and other countries are all collaborating. There's other really good companies like InfraVision, which is which is another company, Chinese company started by two Chinese grad students who went to the university in the US.
They then went back to China and trained their algorithm for detecting cancer in China, working closely with hospitals, and then they deploy that algorithm in the U.S., hiring engineers in the U.S. and they have a presence in Europe, in the UK, and then the rest of the Asia Pacific. So in that in that case, it's a good example of a Chinese A.I. company that that that is able to go global because there are maybe less sensitivities around things like medical diagnosis of tumors and other things. So there's some good examples of Chinese companies in that. The other area is ADA Systems, as I mentioned earlier. So a lot of Chinese companies, of course, are really advanced in deploying AI applications in autonomous vehicles. So if you look at a company like Pony A.I., for example, they have a level four capability. I've ridden in one of their vehicles in Beijing with no driver in the front seat navigating a complicated suburb of Beijing. And it was quite good. It was way better than the Tesla that I drive, which is only level two, two plus.
But it really felt like the car was in control the whole time. So Chinese companies, including Huawei and of course, Pony and weRide a bunch of other ones, are really doing quite advanced work in AI algorithms applied to autonomous driving. And then finally with generative AI, you see companies like Huawei doing things like port automation. So if you go to this demonstration port in Tianjin, there's no workers there. Everything is being done remotely by a couple of people and it's all automated and it uses a combination of AI and other techniques. 5G, is of course, the important part of that. And then there are applications in mining.
So Huawei is using its Pangu large language model, for example. And as part of a of automation of mining operations is a big mining operation in Shandong that's all automated, using a combination of generative AI and and 5G. So those are the kinds of applications that are that are a huge focus in China right now. Fantastic. Thank you so much. Since you already mentioned Huawei let's turn to Huawei next And we'll be talking more generally about China's quest for domestic hardware and chips.
So turning our attention to to Huawei, we know the company has face challenges and U.S. sanctions, etc., etc.. And this. Because they've managed to reenter the 5G phonemarket here. I really love to hear from both of you, starting with Paul S. So please, can you detail the strategies employed by Huawei and what are the lessons to be learned from Huawei's breakthrough? Well, they're certainly going to make do with the technology that they have on hand and see as far as they can take that in terms of building more advanced chips. I think what remains to be seen is how far they will be able to take that. And I think we'll see. They've they've certainly been able to develop some more capable chips. I think the biggest question is going to be, can they scale that in a way that's cost effective? And one of the big unknowns here is US export controls is then pushing Chinese companies to maybe use the technology in other ways or simply invest in new types of technology for leading edge nodes or to circumvent those means of production and potentially do different kinds of breakthroughs in a short term. They're going to have significant challenges going forward.
It's clear that the export controls are going to be able to slow down some development, put some roadblocks in their way. I think in the long term, there are some very open questions and it's going to be very challenging for the US control to actually hold back Chinese chip development over time. And so, Paul T, please also chime in here on how do you assess Huawei's current state of development in terms of its chip technology? What are the lessons we should learn from while we surprising breakthrough? Well, you have to put the break through some context. I got a hold of about Mate 60 in Shanghai a couple of weeks ago. It's a very impressive device. So, again, it's important to note that there are two real big strands coming together here. One is the export controls on Huawei itself and the extension of extraterritorial export controls on the power that prevents it from using TSMC, for example, to manufacture those advanced chips. And the other strand of export control, which is newer coming from last October 7th, is our controls on SMIC, the domestic foundry that actually manufactured this chip probably at the seven nanometer node, although, you know, both companies are being pretty mum about the origin of this chip.
But basically what happened was Hisilicon, which is the chip design arm of Huawei, was designing at a very advanced level before the US put controls on Huawei in 2020. And so Hisilicon was there was designing at five and three nanometers. And so what Huawei has been able to do is turn that hisilicon design around and use it with smic's capacity to do some layers at seven nanometers using their existing tools. Now, those tools that that Smic used, they've had since 2019, the actual techniques for manufacturing, advanced semiconductors at that level, using those tools that Smic has, are well known in the industry. So even before the October start of the controls, Smic, and Huawei, had all the everything they needed to actually manufacture that chip. Now the challenge is manufacturing that chip at commercial yields and scale. So there's still a lot of questions about, as Paul noted, about whether Smic can can ramp up and manufacture the tens of millions of systems on a chip that you need if you're going to be sort of competitive in the smartphone business. This is a very different business model, for example, than, say, you know, 5G base stations.
And so there's still some some question about whether, how quickly and rapidly smic can ramp up to produce the numbers of systems of chips Huawei will need and what's the roadmap going forward? There's still a lot of concerns about that currently. The thinking is you can take the equipment that Smic already has, the advanced DUV lithography equipment, for example, probably can go to some level at like five nanometers is capable of doing that. But it's just really challenging to do that and get the yields you need to support the tens of millions of chips. Before the export controls, for example, Huawei was producing 250 million chips, the Kirin 9000 series at TSMC. So that's the kind of scale you need to be able to sort of produce generation after generation of chips.
Now, one thing we should just mention, too, is that the U.S. export controls, one effect of them has been to force Chinese toolmakers to get better, to sort of be more innovative. And so what we've seen, for example, is the sort of vertical integration of Chinese chip toolmakers working very, very closely with the foundries in China like SMIC and Huahong, which is another key player. So it's when you try to determine how effective those the export controls are, of course, they're effective in cutting off some access to technology. But Chinese companies have been pretty effective in working around that, as we see from Mate 60. And also, Huawei uses a lot of very interesting engineering techniques to try to mitigate some of the deficiencies of having to use the less advanced node. So, for example, things like power consumption, they've managed to to do some novel things to improve performance there.
And they've included in the Mate 60 some apabilities like satellite communications, satellite messaging and satellite phone calls. So they're really pushing the envelope in terms of engineering, using their existing tools and using the existing capabilities that Chinese companies have. So it's a pretty impressive feat, you know, regardless of what you think of the export controls. Thank you so much. So we talk about Huawei and Smic, but what's the general status of China's domestic chip development, especially compared with global leaders?
I want to hear an evaluation from both of you. And besides Huawei and Smic, as we mentioned, who are the prominent players in this space that we should be on the watch out for. Please. Starting with Paul S. There are. So I think I mean, China's indigenous chipmaking capacity has lagged behind the industry for a very long time. It's an area that they've struggled to develop despite a tremendous amount of government investment in this space. Now, of course, the export controls change that dramatically in two key ways. One is that the US controls, along with Japan and the Netherlands, chipmaking equipment, make it even harder for foundries like SMIC to be making chips because they're cutting off their access to the most advanced technology things like extreme ultraviolet orthography and associated other tools and equipment. However, what the US controls also do is create enormous market incentives for Chinese chip makers and equipment providers to then find ways to innovate because the US controls the extra territory, controls and the chips themselves now create this huge market for advanced chips inside China.
The kind of chips that that Paul was talking about that Huawei was making. Now they're denied access to from TSMC, from their advanced nodes. And so there's now big advantages inside China for finding ways to maybe be more creative using deep ultraviolet lithography, seeing how far can you go there? I think very much an open question, but we've also seen that not just in China but globally, there's been a huge explosion of companies in the design space. So while there's been a concentration of companies in the foundry space, at each node, there's fewer and fewer companies making the most advanced chips, manufacturing them in the design space. There's lots of companies designing chips, and a lot of Chinese companies are doing innovative things in the design space, and that's a route for them to continue to grow, kind of to some extent bottlenecked within the nodes that they're going to have access to on manufacturing side, but continue to innovate in design. And so I think, you know, we're going to see how far China can push that, But there's big incentives for them to do so to try to meet that market. Thank you so much. And turning to Paul T, how would you assess China's current state of development in terms of.
Yeah, that's that's a great question. And I've actually got a paper in draft that that's looking at sort of where China is overall and the semiconductor industry. So just a quick highlight. I agree with Paul. On the design side, Chinese companies are very advanced. Hisilicon was designing chips that were comparable to Qualcomm and other other leading companies in the sort of, you know, smartphone space. And you've got other companies like Unisoc and Xiaomi, a lot of the phone companies have tried to get into design. That's a really tricky thing to get into though, because it's a big commitment financially if you're going to really get into advanced semiconductor design. But you also have companies in the GPU space like Biren that was also added to the entity list around October 17th.
But the big players in China, like Alibaba and Bytedance, are also designing their own chips. And so, you know, there's a lot of players in that space. Horizon Robotics, for example, is another key player designing Edge AI semiconductors and is doing quite well. But the problem, as Paul notes, is fabbing the stuff. And how do you fab this stuff when when in some cases you're cut off from doing advanced designs at TSMC. If you're a Chinese company that's on the entity list, for example, which will prevent you from using TSMC. So a lot of these companies are on those lists.
So in China you have Smic, and you have four really major fabs, you have Smic, Huahong and then for memory you have YMTC and CXMT. And each one of those has is under some sort of U.S. sanction. And it's trying to figure out a way around this by working with the tool makers. CXMT today just announced a breakthrough in in Dram for example.
But again, if you look at that, it's basically a 2020 technology or 2018 2019 technology. So a lot of the efforts in the fab space are Chinese companies making catching up to basically two or 3 or 4 years ago of where the Western leaders were. So the challenge going forward is how do they how do they keep moving up the the value chain? So as I noted, you know, the Chinese toolmakers and this is companies like NAURA, AMEC, SMEE, these are companies that are that are that are competing with Western toolmakers like Applied Materials, Lam Research and KLA . And so all those companies in China now have, as Paul noted, have this incentive to to get better move up the supply chain, but also work more, much more closely with the foundries because that's really how Western toolmakers work.
All the US toolmakers, for example, work very closely with TSMC and that's how you get better. You work with the design companies designing advanced designs. Then the toolmakers have to make better tools. And the companies that are fabbing those those chips have to figure out processes that can use those tools and be used for manufacturing advanced designs. So all that is happening within China. But it's a big challenge because as Paul noted, the export controls mean that those foundries can't get access to the EUV lithography, for example, which is what many of the other foundries are using, Intel, TSMC and Samsung.
And so they have to figure out a way around that. Now, there's lots of other things, but the Chinese government and the Chinese companies are trying to do to figure out alternative approaches to things like advanced lithography, because really in this space, these technologies, it's not like that US companies or Japanese or Netherlands companies have have technology that's locked in a box. And if you prevent China from getting it, you know, China can't do it. These are applied science. This is applied science. And so there's many different ways to do these things, for example, with the tool makers.
And so Chinese companies and with help from the Chinese government are trying to figure out ways to do things differently and do things better and trying to get to some point where they can get back in the game of manufacturing, advanced semiconductors, probably next year they'll have a 28 nanometer line. That's all domestic tools, probably still with ASML lithography as part of that. And then they'll move to 14 nanometers and have a fully qualified domestic production line largely free of Western tools, except for probably in lithography. So there's a lot of efforts underway in China to do that. And the Chinese government has revamped its whole approach to semiconductor industry policy and is going to be driving some really innovative efforts in that sector, for example, to share a government R&D with a select group of Chinese commercial companies to try to push and get around and advance despite some of the U.S. export controls. So this process is going to play out over the next 2 to 3 years as China, Chinese companies like Smic extend the capabilities of their existing tools and try to figure out where to go from from there.
And in the next section, we're going to turn to the policy. We're going to discuss the effectiveness of the United States policy of export controls and sanctions and the Chinese policy of semiconductor industrial policies, big funds, so to speak, as Paul just mentioned. So I'm going to turn to Paul S first. the Biden administration recently expanded its set of restrictions on exports of advanced computing trips to China. Do you see these new sets of rules as likely to be more effective in achieving the administration's tentacles? Well, to some extent we know that they are better because they capture now a wider set of chips.
And one of the things that we've seen is when the export controls first came out last year, Nvidia, of course, adjusted their chip designs to fall under the threshold, you was now moving that that that sort of yardstick in terms of what's permitted. And it's clear that the US government is going to continue to be adaptive to what we're seeing US companies do, as well as Chinese companies and others to try to cut off ways of China getting access to either chips or chip equipment. But it's going to be a moving target. And whether it's ultimately effective in restraining Chinese chip development and development is really challenging. Not the least of which, because when it comes to A.I., there's actually lots of ways to get access to the computing hardware that you need. Chips themselves are one way to do that, but another way is through cloud computing providers and the export controls, you know, go after that as well. Expanding the scope, looking at data centers of Chinese companies outside of China.
The executive order that the White House recently released begins to make moves on Know your customer requirements for US cloud providers. And so those are some steps the US is trying to take that I think are going to be needed. Ultimately, if the US is going to have to try to find ways to be restricting Chinese companies access to cloud compute.
Actually cloud is is a better approach for the US to take. That can be a little more fine grained in terms of its approach. The US has talked about concern about Chinese military use, human rights abuses inside China and certainly with the recent export controls, a lot of concern about what the executive order talks about as dual use foundation models for the most capable advanced AI models. And if you can control through the cloud, that actually can allow more fine grained controls, which could undercut some of these market incentives that exist that we've been talking about that are driving Chinese chip development internally. So there could be a better approach over time. And then the last component that really remains completely unaddressed in all of this is open source A.I. models, which is to say that right now in the current set up, you have large US companies training, cutting edge models and then releasing them open source, which entirely circumvents U.S. export controls. So even if the US is successful in restricting China's access to chips themselves, that doesn't matter.
If Chinese labs can go download an open source model and then fine tune it at very, very low cost, you don't necessarily need the most advanced chips to do that. And so I think we're going to continue to see the U.S. government be responsive to changes in the marketplace and in the technology. But it is a moving target and there's still a lot of gaps in U.S. export controls. Fantastic. And Paul T, your take on the effectiveness of the new set of export controls. Yeah, great, great, great question. And I agree with with with Paul's take on all this. I mean, I think the challenge here is these controls were not really designed for that.
I think we have to remember this. They were designed in an era of weapons of mass destruction where you had sort of small companies that were that were producing a key widget that could be used or key material that could be used in a nuclear weapon. They were not designed to maintain U.S. technology dominance or to prevent China from getting A.I. capability or high performance computing capability or deal with human rights issues. So I think that's that's sort of the backdrop of all this. And so we see this challenge in sort of drawing, trying to draw lines around the technology that we saw with October 7th on the GPUs and Nvidia right away came out with, you know, the CPU GPUs that were just below the threshold. We saw this again on October 17th. So that just highlights the challenge of using things like export controls for these these purposes that they weren't designed to do.
And also you're trying to in this age, you're trying to get on the tool making site, for example. It's a global issue. So you have companies in other countries that that would need to align with the US if you're going to have effective controls. And they're you know, we at one point had the agreement, which was a multilateral group. But Russia is part of that. So sort of dysfunctional. And so you've had to have these other sort of workarounds with the US and Japan and the Netherlands agreed to some collaboration on controls around targets for manufacture agreement, but not on all controls. And so there's still a lot of disagreement.
And the Dutch government, the Japanese government, you know, and both companies in those countries don't like the some of the controls because China remains a huge market for them. and they're reluctant to fully align so that the challenge of the export controls is, you know, it's a tool that arguably needs to be revamped. So it's not surprising that there are ways that companies find ways around this, both on the U.S. side and on the China side. And then the other piece of it I think is important is, you know, what's the ultimate goal here? Is the goal to really prevent China from getting advanced computing capability. This was called out last year, for example, in what I call the Sullivan Doctrine, where national security advisor Jake Sullivan talked about advanced compute and biotech and green tech is these force multipliers that are of key national security concern to the U.S. But advanced compute is a huge area, right? It includes semiconductor summit technology, factory equipment, AI, quantum computing and high performance computing.
And China already, for example, has a very robust high performance computing capacity. AI comes into play where where there's some convergence of AI and high performance computing with the use of advanced GPUs, for example, to do acceleration for certain kinds of workloads in the HPC space. And that's one of the arguments for controlling these these technologies is that China could use high performance computing in modern weapons systems like hypersonic glide vehicles and missiles. And so that's that's an argument that the administration continually makes that it's trying to control a small amount of technology that has military end use. The problem is that those GPUs are the vast majority of them are used for nonmilitary applications in China. They're used for drug discovery or, you know, as we talked about, cancer detection or materials development so that so the vast majority are not used for military and use.
And so the administration is controlling in a sense, this very, very dual use technology in ways that have a major impact on on China. So, for example, already after the October 17th controls, Alibaba and Tencent said this would have a big impact on their their cloud services. Right. These are companies that don't provide anything to the military arguably or very little put military. And they're they're commercial companies. They do they're e-commerce companies and social media and gaming companies. But the impact of those October 7th controls is hitting them because they can't have a reliable long term source of GPUs to do to offer services in the cloud or to design, you know, generative A.I. models.
Most companies in China have stockpiled but two years worth of those advanced GPUs and it would probably be okay for two years. But then after that, the question becomes, you know what? What happens? What are workarounds? Are they going to be able to find that to continue to have access to reliable, you know, advanced hardware? So again, the question of sort of how effective the controls are depends on, you know, what what one believes is the end goal of this and whether this is this is a realistic to believe that that the U.S. government can control China's ability to develop high performance computing and advanced computing and AI by controlling choke point, so-called choke point technologies like GPUs. And I think the jury is still out on that because, you know, the technology is very complicated and changes rapidly and these controls. There are ways to circumvent these controls.
But the longer term issue is I you know what with her I and how important that will be for national security and things like cybersecurity and, you know, enabling malicious actors to develop bioweapons using generative A.I. models and that's the fear, I think, with the U.S. government in the EO that Paul mentioned, I know there was a big focus on national security related issues around A.I. and then the sort of general fear that over the long term artificial general intelligence, whoever, whoever gets there first, will have a big advantage, both from a sort of national security and economic security point of view. But that's a that's a far, far flung sort of concern to be implementing controls now. And again, that again highlights my initial point, that that these controls were not designed to prevent, you know, the future development of artificial general agents by some malicious actor or by or by a country.
So it's a very challenging prospect. And in the short term, the big losers are arguably, in this case, U.S. companies that, you know, whether it's the toolmakers on the manufacturing side or the tech companies like Nvidia, Intel, Qualcomm on the on the sort of the chip sale side. So it's a it's a challenging, you know, issue for to determine, you know, are these effective and sort of who's winning. I see. So Paul, since you mentioned, I just wanted to talk a little bit about the misalignment of incentives, so to speak, here. Are other countries on board or to what extent or are there countries like the Netherlands and Japan and Korea on board with the United States set of export controls and sanctions? And how are industry leaders to whom China is still a vast market navigating this new regulatory landscape? How are they thinking about those export control measures? I'm going to start with Apple, please. Sure. So on the export controls, on the manufacturing equipment, I mean, the US has been dragging Japan and Netherlands along reluctantly. As you know, as Paul mentioned, the there are some disagreements between the allies on what specific kinds of, you know, where to draw the line on some of these controls.
And there's a lot of money at stake for these companies. And frankly, all three of the countries, including US companies. It's a little bit different with U.S. companies where U.S. companies, of course, depending on the company, many of them have huge incentives to be supplying technology, whether it's, you know, chip making tools that they can with export controls or chips themselves. As we see the case in Nvidia, to the Chinese market and U.S. companies are going to find ways to to, you know, continue to supply to the Chinese market to the extent that they can and it's permissible by the US government.
And I think, you know, the the latest development on the chip export controls, for example, where when they first came out was a certain threshold by the U.S. government, Nvidia redesigned their chips, released ones just under the threshold. Now we see the U. S government move that even further. That's not necessarily a bad thing if US companies are saying, okay, we're going to go ahead and ship whatever the technology is just below the threshold that the US government set. If you set the threshold the right place, that's fine. What that may highlight is if then the US looks at that and the government says, well, we're not happy about that, maybe we set the threshold of our place and we need to see some adjustments there. But ultimately, like the big picture here, and Paul was talking about this wide expanse of study, these export controls, is in the US government's interest for these controls to be as narrowly scoped as possible for whatever goal they're trying to accomplish.
And it's been a little bit uncertain exactly what is the goal the US government has here. You know, Jake Sullivan, given a number of speeches talking about and last fall when the export controls first of all now talking about the shifting the US goal to keep China as far behind as possible in these technologies. But specifically with regard to the chips, you just talked about these very narrow military specific applications, hypersonic missiles. Okay. Is that what they're going after? Is it to create some distance between Chinese labs and US labs and the most cutting edge models which are now we're seeing very much dual use and they have a lot of commercial applications, but also potentially applications in enabling the development of chemical and biological weapons in cyber tools and basic scientific discovery. There's a lot of unknowns there, but we can already start to see with GPG for some some kind of hints at some of those capabilities in potentially future models that may not be further down too far down the road. So was that the goal or is it just across the board to to try to hold China back? And right now, the export controls are so broad that they're affecting things like commercial cloud applications that are not only not. Militarily relevant.
It's just hard to see the strategic significance to the US. Even as sort of like a basic, basic foundational science and technology advantage here. And so the extent that you can find ways to more narrowly scope these controls, obviously we want to be multilateral with allies, but to make them as targeted as possible, it's going to be in US interests in the long run to make sure they do that. And that's it. Thank you so much. Paul T, please also weigh in here. Yeah. Just real quickly on on Japan, the Netherlands, you know, this is a complicated issue because neither of those countries has, for example, the type of end-use controls or domestic controls that were included in the U.S. package on October 7th. And so that has complicated the issue of getting some level of collaboration and companies in those countries like ASML, for example, in the Netherlands are really you know, there are huge company they're sort of the the Google and the Intel all rolled into one of the Netherlands.
And so the Netherlands, the Dutch government has not been eager, for example, to go along with all the U.S. controls. On October 17th, for example, the US added an added a technology parameter called dedicated chuck overlay, which is DCL, which is a really complicated term, but it refers to how you get accuracy, for example, in DUV lithography systems and that and that meant that the Dutch now have to control much older gear. So gear that's more than ten years old but could be used by some end user in China, for example, to try to get around some of the limitations on the end use controls that the US has imposed. And ASML of course is livid about this because you know, this is not cutting edge technology in their view. And you know, they have a lot of a lot of equipment in China that that falls under this this these categories. And so both the Japanese government and the Dutch government, for example, are really not eager to to align fully on these controls. Because their companies are going to be the ones impacted.
In the case of Japan, Intel and Cannon and Nikon, for example, are all producing the equipment. And China's a huge market. Those countries are also concerned about Chinese retaliation, which we haven't talked about. And there's there's growing concern that China will retaliate against against some of these controls. And the graphite controls they put into effect were a direct response to the October 17th controls.
So China has not yet pulled the trigger, for example, on on some of these restrictions they put in place on gallium and germanium last summer in response to U.S. export controls and in in October around graphite. But that's the concern that that that companies and in countries that are that are aligning with the U.S. for example on some of these export controls could be impacted by Chinese retaliation. So as Paul rightly notes, I think the argument in the Netherlands and Japan is that these controls should be very, very narrow.
And before October 7th, there was a there was some understanding about how narrow the controls should be among some of those allies. And then when the October 7th controls were released, they turned out to have been expanded considerably, for example, around some of the unused controls. And that surprised U.S. allies. And so that the last year has been in part, you know, very detailed negotiation around trying to arm twisting allies to align with those those more expanded controls. And so getting back to Paul's other great points, you know, Jake Sullivan has talked about a small yard high fence. But if you look at the US policy and other aspects of what Jake Sullivan has said, for example, around advanced compute, you know, the yard is is pretty big and and the fence, you know, is sort of, you know, growing and big. But because of these other these technologies are have global supply chains the yard is necessarily large because you do have other countries and companies involved in this.
And so that's been part of the challenge, that the discussion has been around that this is a narrow approach, but in reality it's been a much broader approach and that has impacted a lot of companies in the supply chain or the semiconductor supply chain. And arguably a lot of US technology leaders whose argument is that access to the China market is important for them to continue to do R&D and maintain their leadership position in these advanced technologies, that's a national security issue. That argument hasn't gotten a lot of traction in Washington, but that's what the how the industry views this. The controls sort of emerging away from a very narrow focus on on a smaller set of advanced technologies.
Fantastic. Let's turn to the other side of the coin, so to speak. How would you assess the likely success of the failure of China's industrial policy when it comes to a semiconductor big fund, the big fund version two, etc., starting with Paul S? Well, certainly we've seen that China's been willing to invest tremendous amounts of money in not just chips, but other areas in terms of industrial policy with mixed success. And some areas they've had significant progress. Chip making has been a long struggle. I think, you know, big picture, a lot of these efforts, they're very inefficient, they're very wasteful. But the government has been willing to spend tremendous amounts of money and they've been able to make progress in a number of significant areas.
Now, again, one of the biggest challenges here with chips is it's not just money in the past. Money has not been enough to get there. But what's different now is that there are these market incentives for Chinese toolmakers, chip makers, to be meeting this demand. And, you know, that creates these feedback loops where you've now got, you know, as Paul was talking about earlier in our session here, toolmakers working closely with the chip makers, getting that kind of feedback and getting products out into the marketplace, engaging with users, whether it's other companies or commercial users, and getting their feedback on performance is going to help to level up quickly. And certainly that's a very different kind of role that the government plays than we see here in the United States. We see some movement, of course, recently in the US in terms of the US government taking a bigger role in science, technology, innovation with the Chips Act and significant amount of funding towards US chip making here, R&D and more advanced foundries here in the US. But in general, still the US government's been just much more reluctant to be engaged. For better or worse.
Obviously, government spending can be much more inefficient than than the private market. But there are lots of ways where government investment in basic R&D might be doing things that private companies aren't willing to do on their own. Turning to Paul T. Yeah, I think, you know, and again, I'm writing a fairly detailed paper on this on China's sort of reaction to all this. But I think there's a couple of things just to highlight that are new. So there's going to be more of a top down leadership in China when it comes to semiconductors as a new leading small group at the very top that's going to be overseeing this. Basically, the Chinese will have gone from allowing sort of the scientific community to drive things. This is back in the 863 program days where there was a and there was a much heavier government role then with the set up of the National ICI Investment Fund in 2014, the idea was to give market forces more play.
And the view now in China is that failed. And so there's a there's a view now that China needs to figure out another approach to that. So potentially, for example, they'll have a they may have a large state owned enterprise that will play a more prominent role in helping to drive the overall sector, because that's worked, for example, in the aerospace and other other sectors where China has been has been successful in advancing the technology and particularly domestically. And so I think that's going that that's going to be one thing. Then it's also going to be, as I noted earlier, more sharing of R&D with key private sector companies. So those among those private sector companies are Huawei, Smic and now rather the toolmaker I mentioned, and that was announced earlier this year. And so there's going to be a lot more trying to leverage government R&D to these key choke point technologies, arguably things like lithography, to try to help advance the sector. And then I think the new round of the National ICI Fund, which just was just funded recently and is going to focus on on toolmakers, because they now belatedly realize that the initial a lot of the focus was on design and manufacturing.
But it turns out that the tool making area has been the big Achilles heel. Given, of course, that the US controls in place. And then I think there's also a lot of sort of tech of R&D going on under the radar. Huawei, for example, is has a tremendous amount of experience in all of this, manufacturing, all these manufacturing technologies. And I think they're working very closely with the toolmakers and with companies like Smic, obviously on a whole range of of of new approaches.
Huawei probably realizes that should have been more like Samsung. And probably pursuing that kind of a strategy where you know Samsung has its own foundry business so that they do all the commercial electronics but they have they have a massive foundry business that that manufactures most of their products. And I think while we probably regret not pursuing that that strategy earlier and then I think that one other goal is to keep access to Chinese companies for critical things like materials and things like substrates and processed gases, which haven't so far been the subject of export control. So, for example, Japan and Taiwan and other places are critical sources of substrates and some of the things you actually need to do, advanced manufacturing in China is going to continue to try to keep access to those. And then I think finally, the Chinese are looking to those companies looking to poach more engineers and knowledge from from companies in Japan and Korea and Taiwan, which is where a lot of that knowledge goes, because really, at the end of the day, this industry is about knowledge and knowledge transfer. And so you need the people who know what they're doing to to to actually do this. The guy, for example, at Smic who helped them move to the seven nanometer process was a previously an executive at TSMC and Samsung. And so those are the kind of people you need if China's going to going to make progress in these areas.
But I think you have the whole of government effort here and a whole of industry effort to try to figure out the way forward here. And and that's going to unfold over the next couple of years. With that brilliant insight from both of you, we conclude the first part of our discussion.