Hearing: China’s Campaign to Steal America’s AI Edge

The select committee will come to order, backed by popular demand. I'd like to uh call on Representative Brown for some questions.

Thank you so much. Thank you, Mr. Chairman. For a district like mine in Cleveland in Northeast Ohio, strategic competition with China is about whether working people will have a fair shot at a good paying union job, and a future they can count on, or whether China's unfair labor and trade practices will undercut the industries we depend on. My district knows manufacturing, innovation and competition. And we know what happens when Washington talks tough on competition, but fails to protect workers here at home. In Ohio and across the West Belt, we know how to build, there is a deep history and connection to manufacturing. Yes, we need to stop China from stealing American innovation, smuggling advanced chips and exploiting loopholes in our export control system. But we will not out-compete China by hollowing out our own industrial base or asking American workers to pay the price. We must invest at home. And my district magnet is helping manufacturers grow through workforce and advanced techn- technology support and Cuyahoga Community College is preparing students and workers for the future of work including AI related skills. That is what an American first competitiveness strategy looks like. So, when we talk about America's AI edge, I am thinking about the people who are part of that edge, the shop floor worker, the student learning, AI skills, the technician adapting to new tools, and the union member wondering whether this technology will strengthen their job or ship it overseas. I've said before that if we are serious about competing with China, It starts with standing up for communities like those in Northeast Ohio and keep keeping good paying union jobs here at home. We cannot claim to defend American competitiveness while jobs are offshore, workforce programs are undermined, or manufacturing support is frozen. So, Doctor Chan, your testimony suggests China is competing not only at the frontier, but through efficiency, adoption and integration into the physical world. What should Congress do to ensure the United States leads not only in the top-end models, but also in deploying AI across real industries in a way that supports American manufacturing workers rather than displacing them?

It's an excellent question. I think it's absolutely important that we focus on not just having the best models, but on diffusing this technology through our society, preparing our workers for a new AI-powered economy. I think that we want to make ourselves more competitive, including at the local level, down to our industries, down to our manufacturing, down to our services. We wanna make sure that AI can make the country better off, but also make our workers better off. I absolutely agree that that's what it really means to win in AI.

Thank you so much for that. And Mister Alperovitch, did I say that right? Alp- Alperberit, Alpelberich. Okay. You argue that c- that um You argue that compute is the key choke point and that we should not arm the enemy, quote unquote. How do we make sure a tougher export control strategy is paired with a domest with domestic jobs and workforce strategy so that communities like mine are not just told what we must stop China from getting, but also what we are building for American workers here at home.

Great question, Congresswoman. Thank you for asking it. Every single chip that actually goes to China it's a zero sum game, does not go to American companies. Right now if you're a small business that's trying to build your own AI models, maybe b- based off of uh some of the open source models that are out there, you cannot buy the compute. Uh it's not, simply not available, so when I hear that uh American AI companies like Jensen's uh wants NVIDIA, wanna sell chips to China, when American companies are saying we can't go get enough of the compute themselves, that's hurting American business, that's hurting American consumers directly. because there is such a shortage of chips right now. But you're absolutely right, we have to invest more in production, you know, we need fabs in Ohio, we need fabs in Arizona, we ha- need fabs to manufacture these chips uh in as many quantities as we possibly can.

Thank you so very much.

Thank you.

As I close, I just wanna say, um, as we examine China's actions, we need to protect American innovation, close the loopholes, invest in domestic manufacturing, and make sure workers are not left behind or forgotten. And with that, I I I wanna say I look forward to working with my colleagues on addressing these issues and I yield back, Mister Chairman.

Thank you. Representative Stan.

Thank you very much, uh, Mister Chairman. Thank you for continuing the hearing after votes. I know you have a busy schedule, but you were kind enough to allow it to continue, so appreciate that very much. Thank you also the witnesses for being here today on such an important and timely issue. The competition to continuously create more powerful chips will decide who wins the AI race. The area that I represent, the Phoenix, Arizona area, is quickly becoming a major world player for semiconductor manufacturing, its term semiconductor ground zero here in the United States. But we still have a critical gap, advanced packaging. For years, most advanced packaging has been done overseas, while semiconductor companies like Intel and TSMC, which I represent in my community, are growing their advanced packaging capabilities in Arizona, the US is still risking falling further behind on advanced packaging capacity and prototyping. Advanced packaging requires continual research and development to produce the most advanced AI chips. Despite how important advanced packaging is, the United States still relies on offshore facilities for critical prototyping of next-generation semiconductor designs. Mister Alperovitch, how does this advanced packaging, research and development gap create vulnerabilities for the United States?

Well Congressman, because of the physical limitations we're now facing with jamming more transistors onto a chip, the evolution of the technology is such that we're now trying to go three D in the, in the chip design, and for that to work, packaging is essential. So packaging really is fundamental to the advancements in, in uh uh chip efficiencies and chip power. And luckily, as you mentioned, both Intel and TSMC have some of the leading packaging firms, uh technologies in the world. We need to make sure it it comes here. I was very heartened to hear that as part of the TSMC's investment in Arizona to build out six fabs, they're they're also gonna bring in advanced packaging, they're gonna bring in the entire supply chain. That's what we need, to make sure that the supply chain comes here to the United States. Lot of it is gonna come to to your district and and uh to Arizona.

Thank you very much. Um, when we do another chips and science act, chips two point O, um, workforce development will certainly be the highest priority, advanced packaging needs to be a part of it. Uh, I am proud to represent Arizona State University, which leads the country in semiconductor R and D, including advanced packaging, prototyping. ASU's Macro Technology Works is a semiconductor facility in my district with nearly fifty thousand square feet of industrial grade clean room, great for the university, its students, faculty. I noted that the U I noted that the United States risks falling behind on advanced technology capacity in prototyping. The Macro Technology Works facility should be expanded to meet U S advanced packaging R and D demand. Our state previously received a multi-billion dollar chips and science award to create national prototyping and packaging center. We were selected, it was announced, the community was very excited, the university was very excited. That national center would have allowed domestic semiconductor manufacturers to develop and prototype new advanced packaging technologies in the United States. However, uh, the Department of Commerce rescinded this funding and has not articulated a plan to reallocate this funding or to fill the critical gap to have a national prototyping and packaging center. Mister Alperovitch, what will the effect of of funding rescission, such as I've mentioned, be on domestic semiconductor industry, and how can the federal government work with industry and r- and research institutions to fill the advanced packaging R and D gap?

Congressman, I think it's really vital for us to invest in R and D and not just stay resting our laurels here the reason why the most advanced uh semiconductor manufacturing equipment in the world the the sml machine uh was built in the netherlands is because we abandoned research in that field the dutch continued it and now they are the only ones with a machine that can do uh extreme ultraviolet orthography there is something next after extreme ultraviolet orthography there is there is more advanced packaging to get there first before china before frankly anyone else in the world we have to invest in r and d we have to invest in our universities they're at the leading edge here.

Appreciate that. Final question. Uh, even when US firms successfully guard their proprietary code and hardware, the CCP is utilizing model distillation to train their own AI systems using the output outputs of our most advanced models. This essentially allows them to leapfrog years of R and D and billions in investment. Last year, the select committee showed how deep-seek was built using model distillation, basically plagiarizing American innovation through systemically extracting and replicating a model's capabilities. This is unacceptable, violates every US model's term of use. Mister Chan, given that traditional export controls are designed for physical goods and software code how should we be thinking about policy interventions for distillation?

So this is a major issue, it's an ongoing issue. I think the first step is to work with the private sector, to work with our top AI companies. They are the closest to the data, they are the closest to the issue. And they're already taking the first big step, which is information sharing. That's really critical because it's very difficult for a single company or a single lab to identify some of these patterns. Um, the next step could possibly be support from the federal government in terms of facilitating this information sharing, and spotting some of these bad actors.

Thank you for that. This is a important hearing. I yield back.

Thank you very much, uh, Representative Kim.

Thank you, Chairman Molinar, and I want to thank our witnesses for being with us today and even through our little break. Thank you. You know, there are some who argue that despite efforts to slow China down, Chinese chip and AI compan companies appear to be accelerating progress. The argument is that the fact that Huawei is expending and that China has new and bigger models new start-ups and IPOs and rising revenues, sizeable improvements to manufacturing capacity, and even plans to begin exporting competitive systems around the world later this year. All of this is evidence that expert controls are not working. So what's your assessment of these arguments, and uh how do you view the process, I mean progress that China has been making, and how should we be thinking about our strategy to beating our global competitors? And I would open this up for all of you.

Well Congresswoman, all we have to do is listen to the founders of Chinese AI companies to know that export controls are working, because they tell us that it is working. They're telling us that the embargo on high-end chips, uh the DeepStick founder said, is his biggest issue. We hear that the impact of NVIDIA chip supply to China is huge, very huge, according to one of the leaders of AliCloud. Uh another uh researcher from Alibaba says we're stretched

Thank you for the question, Representative. I think uh the the point that you're making about the extent to which China is getting uh uh is is continuing to develop its capabilities,

Mm.

is an important one, but I think if we look at the extent to which they're behind, we see a regular pattern. On average, they're about seven months behind. There's a question of whether that is emerg whether it's a natural outcome of their innovation that they're perpetually seven months behind or if it's partly a r a result of large scale theft of our IP and I think uh from the evidence that we've seen it suggests that the large scale theft of our IP is a huge factor for why they're perpetually seven months behind. And uh you you need only look to what the labs have said about how much data they're noticing is getting uh essentially stolen in the form of distillation.

Mm-hmm. Thank you. Mr. Chan.

So to build on um my fellow witnesses' comments, export controls are important, but they're not the whole story. And while we are trying to slow China down, we need to make sure that we're also trying to speed up our own AI development and bolster our own AI industry. And that includes not only our chips and compute and improving our models, but as I've said earlier, on trying to make AI work for the American economy, for the American people, to diffuse, to create new applications, and to make sure that this is an engine of growth, um rather than a sort of narrowly focused uh technology.

Thank you. You know last year, the Biden administration approved over one hundred billion dollars in tech licenses for Huawei and SMIC. So, do you think that decision played a role in why Huawei is still growing as a semiconductor producer?

I do. I I think that anything that we've done to help Huawei enhance its business to purchase equipment uh to uh be able to harvest expertise both from United States and Taiwan, uh has been detrimental to the United States' security. And in fact, we know that uh with all the um concerns around the wow we ascend ai chips uh we know that vast majority of them are actually not manufactured by wow we through shell companies they've been able to procure those chips from tsmc uh pretending to be someone else we need to cut off not just wow we but all the chip companies in china that are trying to use illicit methods to acquire equipment to to acquire expertise and to acquire the chips themselves for the ai use in china

thank you i wanna put in one more uh point over here you know on april fourth The Washington Post reported that Chinese AI firms with direct links to the PLA are using AI to analyze and publish intelligence on US military positions in the Middle East. Even more alarming, Chinese companies have supplied their Houthis with satellite imagery and precision targeting data to strike US and allied shipping in the Red Sea. So let me ask you, Mister Mahmoud, how should Congress think about the fact that China's AI ambitions are not just a bilateral technology competition, and that they are already actively fueling conflicts involving China's partners like Iran.

Thank you very much for the question, representative. This is exactly why I highlighted that very fact in in my oral testimony. I think what what this should cause us to appreciate is that the the stakes of securing our AI stack are is only going to increase, and that the harms are not theoretical, it's actually American war fighters that be that are being put at risk. This heightens the urgency of going after distillation and also safeguarding other aspects of our software stack.

Thank you for that. I yield back.

Thank you. Uh, represent ranking member Khanna.

Thank you, Mr. Chairman. Uh, I, uh, appreciated all of the, uh, testimony. Do all of the witnesses think that the distillation is the biggest security threat to the United States right now in terms of China getting AI models? Mr. Chen?

Um, it is uh it is certainly a major issue. Um, it would be hard to rank it relative to a whole bunch of other issues that we

What would you say are the top two?

Um, I do think that uh access to compute is is right up there. Um, there uh yeah, I would defer to to my fellow witnesses to to lay out some of the other examples.

Yes sir, my mind.

Thank you for the question, representative. Um, I think that beyond distillation, so I I would agree that it is difficult to evaluate but i also think other layers of the software stack are extremely important here for example model weights which i talked about in my testimony we are not in a position to safeguard our model weights if the ccp decides that it wants to steal them all of the labs acknowledge and admit this and i think that as ai gets better you know today's ai again is the worst it will ever be as ai becomes more capable we can expect the ccp and other adversaries to place more and more efforts into stealing secrets as as uh safeguarded as model weights.

And they would steal that from OpenAI, Google, I mean from the peop- people developing the models?

Precisely, exactly. So all of the companies are um uh h are are not really in a position to reliably ensure that the CCP cannot steal their model weights.

Mr. Rahn, aprovech.

Uh, Mr. Rahn can remember, I I've talked extensively to leading AI researchers at our frontier labs. They all uniformly agree that that there are two reasons why the Chinese companies have gotten as good as they have in their model development. One is distillation of uh US intellectual property by querying our models and getting our outputs. Two is access to US compute, particularly leading chips that they've been able to use in this post-training of distillation. Cut off one,

Been able to use

ideally both of those,

Uh,

and they would be years behind.

sorry to interrupt, I don't wanna I wanted to get your answer, but just for clarification. Been able to use because they've illegally copied them or because been able to use

Uh, I I think both. They've been able to both procure them legally, use them in the cloud elsewhere, and then uh because of the export control violations that we've already seen that DOJ has just uh issued indictment against.

But but you're saying it's because not that the policy hasn't been strict enough, that that that people have been violating the policy, or have they have we had too permissive a policy?

I I think actually both, you know, even if you look at when we implemented um controls on chips like A one hundreds going to China. Uh, they've been able to procure a lot of them before those chips got banned, and and have been able to train models on them.

And uh, would all of you recommend that this uh AI institute at commerce be beefed up uh in terms of the safety regulations and security regulations? Mister.

One of the highlights of uh the highlighted recommendations in my testimony has been to significantly fund the Center for AI Standards and Innovation. So we believe that it should be funded at around fifty to a hundred million dollars a year because it has such an important mandate.

You agree with that?

I do.

I absolutely agree with that. It's important for AI safety, AI standards, for helping the industry develop in a beneficial way.

So in this case, I mean, I've articulated a need for AI safety regulations to protect Americans, to protect uh, consumers to protect jobs. But would it be fair to say that we need smart regulations for our own national security? That having AI regulation and a strong, robust agency is needed for national security? And I'd let the witnesses answer, and then my time is up.

I I I do think that it is very important to take national security very seriously. I was heartened to see that in the federal framework, uh, that the White House put out, there were suggestions that national security related functions in the government be beefed up that better t ai talent go into those agencies and that better plans be made to respond to certain contingencies i i will flag that i do think that there are trade-offs between uh having an institution like kc the center for ai centers and innovation do regulations versus just having it uh issue voluntary guidance and the main trade-off is uh that i wanna flag is that if it is a regulatory entity, it may make it significantly harder for private industry to work well with with Casey. So my recommendation at the time would be for it to be not regulatory.

I think it's very important that we uh have smart regulation, that we can have both innovation uh and progress, and make sure that that progress gets funneled back into our workers and our communities. So I think we need to make sure that we play an active role, and by we I mean uh Congress and uh the American people.

Great. All right, thank you, uh Representative Dekula.

Thank you so much to the witnesses for staying here. Let me just follow up briefly on Ranking Member Khanna's questions. Um, absolutely agree, national security guidelines, smart regulations domestically, we absolutely do need that. Um, my concern is it's not just about us. Because again, they steal our technology, they morph it, they steal the weights, they steal, you know, they distill. And here we have China with no guardrails, no guidelines, or any other country or adversary for that fact. I'm concerned that we are also currently abdicating our global role on a number of entities that play oversight role over AI. So whether it's the World Health Organ organization that does AI oversight for their jurisdiction, UNESCO, the UN, OECD, We have really moved away from there, and as we know China has quickly moved in to provide leadership and resources in these particular entities. China's also proposing their world's uh their Waco, their world's AI cooperative organization, as well as some kind of global framework. Should we not just be looking at ourselves, but be concerned about the global instability? And maybe perhaps trying to go back to providing both that leadership and that direction in terms of the smart regulation and guidelines that we should have as the world develops. um, towards AGI and quite frankly ASI right behind it.

I completely agree. We need to not only lead the world in innovation on AI, but lead the world on trying to understand how we can use AI and and shape regulation in an adequate way, not just at home but around the world. So I completely agree, we need to be working with multilateral institutions, with these international fora, with other countries, to figure out a framework where um, some of these issues are not just domestic issues, these are global issues now.

Thank you.

I agree that the the concept of American AI dominance requires that we have American standards dominance, and I think that requires that we play a role on the global stage ensuring that American standards are leading the way and that we aren't, uh, you know, unwittingly being subject to to other standards. And I fully agree with your point that there are other major players in the game. And I think this does require uh some level of i international understanding and and uh cooperation especially in in certain uh in certain use cases. I think the president referred to this um uh a couple of days ago when asked about the latest AI technology.

Uh I'll just add I don't think we should be ceding any ground to China, whether it's an AI models, whether it's in national standard bodies.

Thank you. I I would agree with you that there are consequences to other actors' actions or inactions as well, and I really do hope that we start to move more towards uh that pivotal um America first like role in many of these organizations to to help lead provide that guidance and that structure um I did wanna also elaborate Mister Chan in your testimony, I really appreciated your analysis that China is essentially looking beyond AGI. Um I often say in this committee that sometimes I feel like we're playing checkers and they're on three D chess many times because it's not just about getting to AGI it's the mass integration. of the technology in order to create, as you said, economic and social gains as well. And as we know, the model for China often is to take something, make it cheaper, make it faster, proliferate, and ultimately dominate the global market. We have seen this time and time again in so many markets. They're doing the same playbook when it comes to AI, and I'm just concerned that we're still focused, as many of our allies are, are singularity on that getting to that AGI versus understanding how you actually use it um within your c- the community, you know, integrate it essentially safely, of course, while China is doing their typical model where we're singularly focused, and they literally will come around and just absorb everything once it's done. What do we need to do from a policy and an investment perspective to counter China on this type of strategy, export control, absolutely that's a given, research, you know, increase our compute, investments, all of these things we have to do that we talked about today to be more of a tip of the spear, but what really understanding their strategy and how is ours is more linear and theirs is more multi-dimensional what do we as Congress need to do uh within the next literally year two years in order to be able to counter uh China's strategy.

So I can offer uh two suggestions. One is um thinking about the diffusion problem, um China has been doing a lot on the open source front. And I think this is one area that we uh risk ceding to China as an important channel, not just for domestic diffusion but for global diffusion of AI models. Um, so this is an area where I think Congress can step in and try to support an open ecosystem an open model ecosystem that's that's American. Um, the second thing is making sure that we have support for uh public research, um, for AI and for AI standards and for AI safety more broadly. Um, this can ensure that not only are our companies competitive, but that those benefits are going to broader society, that, uh, they're going to, um, uh, you know, they're not just geographically concentrated in Silicon Valley, but they're spreading across, uh, the whole country.

Thank you. I know my time is up, but I absolutely agree. We need to invest more, support more, encourage more open sor- open source. Um, on that front, if we look at many of the AI that's out there, you've got China DNA,

Thank you and thanks for your patience, and I wanna thank our witnesses again for spending pretty much the day with us today, and uh it's great having you with us. Um questions for the record are due one week from today and without objection the committee hearing is adjourned, again.