LIVE: House CCP Committee Holds Hearing on China’s AI Competition

Select committee will come to order and good morning everyone. A busy day today on the hill. I know members are having a lot of hearings and people are going to be coming and going but I want to welcome everyone and it's a very important topic. China's smuggling of advanced AI chips is a pervasive threat facing law enforcement and in several cases over the past year the Trump administration disrupted multiple smuggling rings. This sought to arm China with advanced American AI chips. Just last month the Department of Justice announced a 2.5 billion dollar chip smuggling case which would be the largest export control violation in US history. The alleged smugglers went to great lengths to cover their tracks. One of them was Supermicro co-founder Wally Liu. He and his alleged conspirator used a hair dryer to remove Supermicro serial number labels and then put those labels on fake server boxes. They even created fake AI servers to fool US enforcement officers into thinking the chips were in Southeast Asia when they'd already been smuggled to China. Why is China so desperate to acquire US designated chips? The reason is obvious. AI is truly transformative transformative technology and it's already changing how we fight wars, run our government, and operate companies. It's essential for the United States to maintain a decisive lead in the AI race. We cannot afford a future where Beijing dominates this technology. In a new report released this morning, the select committee documents how China currently depends on the United States and our allies across parts of the AI technology stack. China's AI ecosystem depends on the western semiconductor manufacturing equipment required to make most advanced AI chips. They rely on western AI chips to develop their own most advanced AI models. They rely on western AI models to develop their own models and AI services. To try to keep up in the AI race, Chinese companies are buying what they legally can under existing export control regimes and stealing what they cannot. This pattern of legal purchases and theft is happening at every layer of the AI technology stack. Let's start with the bottom of the stack. Beijing has funneled billions of dollars into their domestic chip makers so they can control their supply chain. Yet at the same time, Chinese chip makers continue to rely on US and allied semiconductor manufacturing tools to make their chips. As of now, Chinese chip makers remain unable to produce five nanometer chips at scale, putting them at least seven years behind the leading edge. That's why Chinese chip makers are the world's biggest customers for chip making tools, buying $38 billion in foreign tools in 2024 alone. The country buying the most tools today is China, and that means that one day we might depend on China for our own chips. We cannot allow that to happen. Let's move on to the next layer of the AI tech stack, the chips themselves. Despite billions in government subsidies, China's best chips are far inferior to our cutting edge semiconductors, nor are they keeping up in quantity. Huawei proudly announced recently that it will produce 750,000 of its 950 PR chips, but that figure amounts to less than one week of quality adjusted US AI chip production. Given the sheer quality and quantity gap, Chinese AI companies continue to rely on American chips. As the founder of Chinese AI champion DeepSeek has said, our problem has never been funding. It's the embargo on high-end chips. So it should come as no surprise that Chinese AI firms are doing everything they can to legally and illegally acquire US chips. To work around export controls, Chinese companies are building billion-dollar AI supercomputers in countries like Malaysia, and then accessing that compute power remotely through the cloud. They've also resorted to outright chip smuggling. Estimates of China's illegally obtained chips range anywhere from tens of thousands of chips to several hundred thousand chips. Finally, the last layer of the AI tech stack are the models themselves. American AI labs like OpenAI continue to produce the best proprietary models in the world, using tens of billions of dollars worth of AI chips. That's why Chinese labs are resorting to unauthorized distillation attacks to extract information from our best AI models. Since they don't have enough AI chips to develop the models on their own, they prefer to simply steal them from their American competitors. Anthropic, OpenAI, and Google have all verified that this is happening. The pattern is clear. China is dependent on our tech stack to continue their AI development. China is willing to buy what they can and steal what they cannot to advance their AI ambitions. The task for Congress is to pass legislation that will stop China's multi-pronged effort to legally and illegally acquire America's tech stack to use it against us. I look forward to hearing from our witnesses today, but I now want to recognize Representative Andre Carson for his opening statement. Thank you, Chairman. You're appreciated very dearly. Thank you. And thank you to the witnesses for appearing before the committee today. Last year, the Select Committee held a very important hearing on why American AI must lead. It was very clear from that hearing that the stakes of the AI competition with China could not be higher. If Chinese AI models win, we face a future filled with censorship and surveillance, not to mention in the key industries of the future, including sectors like manufacturing and biotechnology that drive Indiana's economy, we will fall behind. But if American AI models keep its lead, the American people will get to decide on the rules of the road. We can enact protections for workers against displacement, ensure communities do not face hikes to their utility bills, and ensure civil rights and human rights are designed into the algorithms. China understands these stakes very well. That's why the CCP will stop at no lengths to win this competition. Just last week, OpenAI, Anthropic, and Google all came together to form the Frontier Model Forum, a joint effort to stop China's campaign to reverse engineer their AI models. And just last month, the Department of Justice indicted individuals for trying to smuggle restricted ships to China. We should all be deeply concerned by how far Chinese companies are willing to go to gain an edge, even if it means breaking the law. But we should also remember why China wants to get our technologies in the first place. It's because American technologies, built by talented engineers around the world, often with support of federal funding, are simply better. Unfortunately, the Trump administration's immigration policies, budgetary cuts, and attacks on leading AI labs are putting us on a path where we can't take advantage of anything, nor can we take anything for granted. Nearly two-thirds of the top AI companies in America were founded or co-founded by immigrants. That's what's great about our country. Thirty-eight percent of the top AI researchers in America are originally from China. Anyone who hears those statistics and uses common sense knows we need those immigrants. We should welcome them and reject wholeheartedly any form of xenophobia. But instead, unfortunately, this administration is banning immigration from dozens of countries, putting a $100,000 fee on the H-1B visas that AI researchers frequently use, and scaring away students with ICE raids. The scorecard on federal research funding, the foundation on which so much American innovation depends, is a similar story. We are slashing federal grants to universities, including colleges like IU-Indy, Purdue, as China increases overall scientific research spending by nearly 10 percent annually. Meanwhile, the Defense Department is trying to put one of our leading AI labs out of business because of a contract dispute. Killing Claw doesn't help us beat China. That sounds to me like benching one of our top scorers in a championship game. If we keep down this path, I fear that 10 years from now, we may not have as many technologies that China wants to take. That means we need a two-pronged approach. Firstly, I agree wholeheartedly we need to enforce our laws and protect our advantages. Chinese companies like DeepSea cannot and should not get away with stealing our most innovative tech. The Commerce Department should go after those who break laws and break the rules with everything we've got. And it shouldn't just be us holding China accountable. We need a global coalition. We should work with our friends in the Netherlands and Japan to protect sensitive IP in areas like diplomacy. Now, we're all in this competition together. But secondly, we need to make sure our technology remains the best. We need to attract top talent, train the next generation of workers, and keep America's innovation engines running. The easiest way for China to win is if we shoot ourselves in the foot. The future of AI must be built on American values and not the CCP's dystopian vision of censorship or surveillance. I believe we can keep our edge and win the future if we get to work. Thank you. I yield back, Chairman. Thank you very much, Representative Carson. And if any other member wishes to submit a statement for the record, without objection, those statements will be added to the record. Now I'd like to get to our witnesses. And we're very fortunate to have you with us today. Mr. Dmitri Alperovitch is the co-founder and chairman of Silverado Policy Accelerator and co-founder and former CTO of CrowdStrike, one of the world's largest cybersecurity firms. Mr. Alperovitch has served as a special advisor to the Department of Defense and served on the Department of Homeland Security Advisory Council and the Cybersecurity and Infrastructure Security Agency's Cyber Safety Review Board. Mr. Yusuf Mahmoud is currently serving as Director of AI and Emerging Technology at the America First Policy Institute. Prior to joining AFPI, Mr. Mahmoud worked part-time during law school at RAND, where he focused on energy deregulation for US AI leadership. At AFPI, Mr. Mahmoud's research has covered US CCP AI competition, including accelerating the US AI data center buildout, building federal government capacity to counter distillation, and forthcoming research on securing US AI companies from CCP espionage. Dr. Kyle Chan is a fellow in the John Thornton China Center at Brookings Institution. His research focuses on China's technology development and industrial policy across a wide range of sectors, including AI, robotics, semiconductors, electric vehicles, and batteries. Dr. Chan has testified as an expert on China's industrial policy before the US-China Economic and Security Review Commission, and he has a Ph.D. in sociology from Princeton University. And with that, I want to welcome all of our witnesses. And beginning here this morning, let's start with you, Mr. Alperovitch. You're now recognized for your opening remarks. Thank you, Chairman Moulinar, Congressman Carson, and distinguished members of the committee. Thank you for the opportunity to testify today on how China is acquiring, through both legal and illicit channels, advanced US artificial intelligence technology. We are in a race, and the stakes could not be higher. Artificial intelligence will transform every industry, every battlefield, and every government. And whoever fields the best models, running on the best infrastructure, will likely win not just the AI race, but the 21st century. The single most important input to winning is compute, the processing power used to train and run AI models. Let me say that plainly, because it is a finding that should anchor everything that Congress and this committee do on AI policy. The binding constraint in this competition is not talent, it is not data, and it is not cash. Both the United States and China have talented researchers and financial resources. And we have learned that frontier models trained on generic data routinely outperform smaller models trained on more specialized inputs. It all comes down to compute. At its core, compute is based on chips, how many a country has, how powerful they are, and how well they work together. The exponential scaling of that chip power is what has driven virtually every meaningful improvement in AI over the last decade. This is not a secret, and China understands it as clearly as we do. DeepSeek founder said it succinctly, our problem has never been funding, it is the embargo on high-end chips, quote. Because they cannot access the best technology DeepSeek needs, he says, twice the computing power to achieve the same results as international competitors. But if any doubt remains about how seriously China takes this, the lengths to which it is willing to go to acquire compute should settle that question. In the last several months, the Department of Justice has unsealed indictments involving billions of dollars in illicitly exported Nvidia chips routed through Southeast Asia. The stakes of that gap could not be higher. There is a moment coming in AI development, a threshold of what is called recursive self-improvement, when AI systems begin to accelerate their own advancements. The country that crosses that threshold first will enjoy a compounding advantage that may never be closed. The good news is that we have a choke point, and export controls have prevented China from closing the compute gap. But we are not fully exploiting it, and in some cases we are actively undermining it. To ensure the United States remains ahead and China cannot catch up, my recommendations condense into two overarching imperatives. Don't arm the enemy, and don't create another Strait of Hormuz. To ensure that we do not arm the enemy, we need to first hold the line on export controls and close the loopholes. Providing China with cutting-edge AI chips is the modern equivalent of selling rockets to the Soviets during the moon race. No one in 1962 would have seriously entertained licensing our best rocket technology, or even our second best, to Moscow just for business returns or for the notion that it would get Soviets addicted to our space tech stack. The strategic stakes today are just as clear, and arguably even higher. We should be no more willing to hand China the tools it needs to beat us in this race than we were to hand them to the Soviets in the last one. There should be no licenses granted for advanced AI chips to China, period. To that end, the AI Overwatch Act offers a sound framework, mandatory congressional notification before any such export is approved, verifiable safeguards preventing military end use, and a guarantee that chip exports do not come at the expense of domestic availability. Second, stop supplying China with equipment to build its own chips. Export controls and finished chips mean little if the U.S., Netherlands, and Japan continue selling China the semiconductor manufacturing equipment to eventually produce them domestically. The MATCH Act is a strong step forward, and I'm glad to support it. Third, strengthen protection against AI model distillation attacks that enable theft from leading U.S. AI frontier models. Fourth, establish Know Your Customer KYC rules for cloud companies to ensure that Chinese models do not get trained using chips outside of China. Finally, we need to ensure we do not create a semiconductor strait of Hormuz, a choke point where a strategic adversary can disrupt global supply and constrain U.S. policy choices. The United States must address a longer-term strategic risk, China's drive to dominate foundational or so-called legacy chip markets. While the United States and allies lead on advanced chips, China is pursuing foundational chip dominance through the same state-backed non-market policies it has deployed in steel, solar, and electric vehicles. Following these recommendations can ensure the United States keeps its lead in the most consequential technological race of our time. Thank you, and I look forward to your questions. Thank you very much. And now, Mr. Mahmoud, you may proceed. Thank you, Chairman Molnar, Ranking Member, and members of the committee. I'm Yusuf Mahmoud, the Director of AI and Emerging Technology Policy at the America First Policy Institute. It's an honor to speak to this committee about the CCP's campaign to steal and sabotage American AI technology. Today I want to discuss a stark and urgent reality. We are not prepared to secure our AI systems. If we decided tomorrow that it was a top national security priority to prevent the CCP from stealing our most capable AI software, we would face extraordinary challenges. We're starting from a vulnerable position. The CCP seeks full AI domination by 2030, and it lacks the capital and talent to win fairly so it steals. These aren't theoretical harms. Right now, Chinese AI developed from stolen American technology is helping Iran target American war fighters. Let me walk through four ways that we are vulnerable to CCP theft, especially on the software side. First, distillation attacks. Chinese companies run industrial-scale attacks against American AI platforms using stolen outputs to train their own systems. Every major American AI lab has been hit, and distillation of American IP helps to fuel China's AI ecosystem. Although there are legitimate uses of distillation, distillation by our adversaries to steal American AI outputs is a threat to American AI dominance. Second, trade secret theft. The CCP actively supports a campaign to steal American AI trade secrets. In the Lin Wei Ding case, a Google engineer stole over 2,000 pages of AI trade secrets for Chinese intelligence. Around 38% of top AI researchers at American labs received their undergraduate education in China, and the vast majority are likely Chinese nationals. I think this should make us clear-eyed about continued vulnerability. Third, model weight security. Model weights can be thought of as the brains of AI systems. If you steal them and have the compute to run them, you can weaponize and replicate the system at low cost as you wish. Insiders at labs believe we are not prepared to resist nation-state level efforts to steal model weights. Fourth, and a threat I believe is underrated, is the poisoning of American AI models. Frontier AI models train on trillions of data points scraped from the open internet. If an attacker manipulates even a tiny fraction of that data, they can insert hidden backdoors, triggers that cause the model to malfunction under specific conditions while appearing perfectly safe otherwise. Last year, an independent researcher proved this was possible. Working alone with essentially no budget, he planted manipulated text on public websites, DeepSeek scraped it, and the result was a hidden backdoor baked into China's best AI model. The CCP can do the same to us. And a poison model might pass every evaluation and then be activated by an adversary at the worst possible moment, for example, during a conflict when American lives are on the line. Deploying that system in a secure setting like a DOW compute cluster may not help if the model was compromised before it ever arrived. To address these threats, I have five recommendations for Congress. First, to combat distillation, direct the Center for AI Standards and Innovation in coordination with the NSA, other IC partners, and industry to develop voluntary guidance on counter-distillation best practices. Second, authorize and fund the Center for AI Standards and Innovation in the Department of Commerce at $50 to $100 million annually, and authorize the Bureau of Emerging Threats in the State Department. These offices are critical for the situational awareness needed to monitor and counter the CCP. Third, direct the Department of War to publish frontier AI lab security standards and extend them to all frontier labs, not just Pentagon contractors. Fourth, direct the NSA to run red team exercises against the full AI development pipeline to find security vulnerabilities and work with industry to address them. And fifth, pass whistleblower protections so AI researchers can inform policymakers of security vulnerabilities without fear of retaliation. The U.S. is in an AI race where the other side is cheating. Today's AI is the least capable it will ever be, and as AI becomes more powerful, the need to secure it will become ever more urgent. I urge this committee to act with the speed this moment demands. Thank you, and I look forward to your questions. Thank you very much, and now let's turn to Dr. Chan. Chairman, congressmen, members of the committee, thank you for inviting me to speak. I'd like to my opening remarks around one big question. What does it mean for America to win in AI? We can all agree that AI is likely to be the most consequential technology of our lifetimes, and that the United States must win the AI race with China. But what does that actually mean? Let me offer three versions of what it means to win in AI and describe how the U.S. stacks up against China. One, does winning mean developing the world's most powerful AI models? If yes, then the U.S. has the decisive lead. American AI models are widely recognized as the best in the world on virtually every metric you can think of, from math and reasoning to coding and AI agents. American private companies continue to set the pace at the technological frontier, supported by deep capital markets, a dynamic innovation ecosystem, and the best talent from around the world. China's AI models are improving quickly, but continue to lag behind. Chinese AI companies face real constraints on compute due to our export controls and their own limited capital resources. So instead, they're prioritizing other goals. They're focused on efficiency, building models that are cheaper to train and run. They're focused on adoption, using an open source strategy to win users around the world. And they're focused on integration with other platforms and services. On sheer performance, I would expect the gap between our models and theirs to persist or even widen over time. But model performance is not everything. Two, does winning mean having the strongest AI stack? America leads in several critical layers, frontier models, advanced semiconductors, and compute infrastructure. But on energy, China is ahead. In the past four years alone, China built the equivalent of the entire US power grid. Energy is one of our biggest bottlenecks, and we must find ways to add capacity while keeping a lid on energy bills for our local communities. In general, China is pursuing a full stack approach to accelerating their AI development from chips and compute to models and applications. They're focused on not just the best models, but on deployment and diffusion. They want to use AI to turbocharge a wide range of sectors from manufacturing and research to education and healthcare, as well as, of course, their military. They're placing a special emphasis on physical applications of AI, including humanoid robots, autonomous vehicles, and industrial automation. In particular, a large part of their industrial policy efforts are aimed at semiconductors, which they've kicked into high gear in the face of our export controls and other measures. China is now racing to build a resilient domestic semiconductor industry across the entire supply chain. Photo resists, etching and deposition tools, chip design software, memory, packaging, and even advanced lithography machines. Export controls are not a panacea, but they have bought us some time. The question is how we use this time effectively. Three, does winning mean making sure that AI benefits the American economy and the American people? This may be the most important definition of winning and the one we focus on the least. Technological leadership does not automatically translate into broad-based social and economic gains. Having the biggest and baddest models alone won't cut it. The real test will be whether we can leverage AI to make our economy more productive, our workers better off, and our communities more vibrant. This also means managing the disruptions caused by such a transformative and new technology. When many Americans hear about AI, they feel worried. Worried about jobs, worried about energy costs, worried about their children's futures. We cannot just focus on the upsides while leaving these real concerns unaddressed. So when I think about what it means for America to win in AI, I think about all three definitions. We need the best AI models. We need the strongest AI stack, which includes energy. And we need to make sure AI benefits ordinary Americans while addressing the disruptions it may cause. This is where Congress plays a crucial role and where American democracy has the greatest edge over China. Thank you. Thank you very much all three of you and now we'll move to questions and I'd like to start with a few. Mr. Alperovitch, some people are saying that we can keep China addicted to our tech stack by selling them our top AI capabilities. Is that even an achievable goal? Can we actually foster China's reliance on the American AI stack by selling them more capable chips? Thank you, Mr. Chairman. This is a great question. And first of all, it's important to understand what the American AI tech stack is. It is not chips. It's like saying the car is tires. The American AI stack is chips running American cloud providers and operating American models. That's the American AI stack. Selling China chips that they run in their own data centers that are used to train and run Chinese AI models is not helping us one bit. And chips themselves are not cocaine. They're not addictive. In fact, if you look today at all of the major AI models, whether it's Google's, Anthropix, or OpenAI's, they're all trained on a variety of different chips. Google's TPUs, Amazon's Traniums, and NVIDIA chips. And in fact, I was told by one of the leading labs that it took them just less than a dozen engineers in a couple of months to move off NVIDIA's architecture and add other chips to the mix of the training infrastructure that they run. So the idea that by selling advanced chips to China we would keep them addicted is contradicted by historical evidence. If it takes this little effort today for our own labs to move off this infrastructure, why would it take China any more effort? Thank you. And Dr. Chen, you have mentioned that you believe that export controls have meaningfully slowed China's AI development in the near term, while also noting that they've spurred China's domestic semiconductor efforts in ways that would ultimately reduce our leverage in the long term. How do you reconcile those two findings, and what recommendations do you have in that area? So I think this is really key when we talk about the effects of export controls, that they are not a be-all, end-all solution to winning the AI race. They can slow China down in the near term, and we have talked about, and you guys have talked about, how Chinese AI developers need better chips and their compute constraint. So this is slowing them down today. The issue is, as Mr. Alperovitch has noted, there's a broader AI stack, and China's trying to develop a more robust semiconductor industry in response to some of our actions. So I do think that we have to consider the net effects, and we have to think about this as an important, but temporary and partial measure that we need to complement with a whole host of other policies. Thank you. And Mr. Mahmoud, right now Chinese companies' distillation of models may violate the terms of service of American AI labs without clearly violating U.S. law. Should the Bureau of Industry and Security make certain types of adversarial distillation illegal? And what can the federal government do to deter adversarial distillation? And you mentioned some of that in your opening remarks, but if you could continue to elaborate. Thank you for the question, Chairman Molnar. I think the answer is yes. The federal government has a lot that it can do. In part, the intelligence community has a massive ability to collect threat intelligence and determine where the actors are that are engaging in distillation attacks. I think it's worth noting that over the past few years, distillation attacks have changed in their strategies in response to counter-distillation efforts by the top AI labs. So whereas before, when there were fewer counter-distillation efforts, there would be a small number of accounts prompting the models many hundreds of thousands of times, now there are thousands of small accounts that are prompting the models very small number of times, and that's because they are actively trying to avoid getting caught doing distillation attacks. So I think that intelligence community monitoring and collection can help significantly. I also think that Congress can direct the Center for AI Standards and Innovation to create voluntary guidance for AI companies in best practices in counter-distillation, and they can do so working with the IC and working with industry to create those best practices. I think even the ability for companies to share intelligence with one another is very valuable, and I think having the federal government take that initial position would be very useful. Thank you very much. Representative Carson. Thank you, Chairman. We all know that AI competition is either make or break, and the stakes couldn't be higher. Last week, Anthropic released a new model known as Mythos. It's so powerful that the company isn't even releasing it to the public. That's because hackers would use it to find bugs in the system that essentially supports our financial system and critical infrastructure. If China manages to create a model as powerful as Mythos, do you all think they'll take the same precautions as Anthropic, or might they just release it, allowing hackers, or even China itself, to use it against our infrastructure? What are your thoughts? Congressman, this is not even a hypothetical, because we know that China has built a vulnerability disclosure database where they mandate for researchers within China to disclose vulnerabilities first to the government before they release them publicly. We also know that they have used that database to attack American and other Western targets. If they had capabilities like Mythos to automatically discover bugs, absolutely they would use it to compromise our government networks, the commercial sector, on an unprecedented scale. This Mythos situation actually highlights the chips point even more so, because the estimates out there from AI researcher Ryan Fedesuk are that Mythos is running on at least 100,000, maybe 200,000 NVIDIA chips just to work with companies to identify the bugs in their systems today under Project Glasswing. According to Secretary Lutnick, the entire production line for China estimated this year is 200,000 chips. If China had this model, they would literally use the entire year's worth of production just to run this type of project, aside from, you know, being able to run any other models or even train new ones. So this is why compute is so essential, not just to help stop China from creating these models, but also to prevent them from running them effectively. Thank you. Mr. Mahmoud. Thank you for the question, Representative. I think it's also worth asking how they would come to acquire that capability to begin with, and one key way they could acquire that capability is by stealing it directly from American AI companies. Now, as I mentioned in my oral testimony, today's AI systems are the least capable they will ever be, and as impressive as the most recent models are, I think if one extrapolates the trends, one can imagine even more impressive cyber offensive capabilities, much more impressive than we have today. If such impressive cyber capabilities were to be unlocked, or I suppose the more powerful they are, the stronger the incentive is for the CCP to steal the model weights directly. All of the top labs believe that they are not robust against CCP theft of the model weights themselves. If you're able to steal the model weights, you can fine-tune away any safeguards that exist in those model weights trivially. It's very, very easy to do so, and you could immediately weaponize it against the American people. Lastly, Chairman, the second is energy data centers, which are being placed in communities, many of whom don't want them, for different reasons. One, raising electricity prices for ordinary folks. Your thoughts on data centers, and I think that they should pay their way and have in-line systems like water systems that will not pass cost on to users. Do you all agree that data centers should cover their own energy needs and not place an undue burden on citizens and rate users? I think that if we want American AI to be competitive in the long run, to be sustainable in the long run, and to build out this kind of compute infrastructure, we need to address these local community issues. They're very real. They're very concerned. This will become a bigger and bigger issue over time. I completely agree. We need a robust framework to make sure that local communities don't end up bearing the costs for an American capability. Thank you. I would just like to second my fellow witness. I agree completely, and I think that Congress has a role to play in codifying the ratepayer protection pledge. I'll echo those comments. Thank you. Representative Bilirakis. And members, both have been called, so I want to make sure we get a question in if we'd like one. Thank you. Thank you, Mr. Chairman. Earlier this week, I chaired the Commerce, Manufacturing, and Trade Subcommittee hearing on energy and commerce on the state of semiconductors ecosystem. The scale of the vulnerabilities of the global semiconductor supply chain was staggering. What particularly struck me was a report by the Semiconductors Industry Association that estimated that the impact of a disruption to semiconductor supply chains coming from Taiwan could decrease U.S. economic output by 11 percent, double the 2008 recession. Mr. Alperovitch, what do you view are the most important steps the United States needs to take individually and with allied countries to isolate China's potential impact to the semiconductor supply chain? I know you mentioned it. Maybe you can elaborate briefly. Absolutely, Congressman. This is a really important issue. We need to focus on the and the CHIPS Act that this Congress has passed, really vital to this effort. If TSMC follows through on their commitment to build six fabs in Arizona, six advanced fabs, that will go a long way to achieving independence of production for U.S. needs, at least in advanced chips. But we also have to focus on foundational chips because in every single AI system, in every single device, if you look at this iPhone, there's only three advanced chips in there. There's 150 foundational chips that drive this phone. China is trying to dominate that industry. We need to make sure that allied countries, and the United States in particular, invest in fabs across the spectrum of both advanced foundational logic memory. This is really, really vital, and we need to work with allies like Korea, Japan, and others to help build additional capacity there as well. Thank you. Agreed. To win the race against China, American AI companies need to succeed at home and abroad. However, we've seen large foreign trading partners, especially the EU, enact sweeping new AI data regulations that could be used to target U.S. companies. Mr. Mahmoud, how can we address new and emerging digital trade barriers to ensure American AI companies can out-compete their Chinese competitors on an even playing field? Thank you for the question, Representative. I think we should have a willingness to use the full toolkit of our own trade practices to deter efforts that may undermine the competitiveness of our own companies. The details of what that looks like would have to be determined by the executive branch, but I think that we should broadcast a willingness to, for example, potentially apply tariffs. And I think the example goes beyond the EU. There are also ways that other countries, in particular China, might do the same. Very good. Thank you. I yield back, Mr. Chairman. Appreciate it. We're in the midst of figuring out if we're going to come back after there's a multiple vote series that they have and trying to figure that out, but why don't we take a recess right now and is Representative Takuda coming back? No? Okay. And I don't, because of the multiple vote series, I don't know that we're going to be able to come back. And I just want to thank you. What a great hearing this has been and for the information you've shared with us. And there's multiple appropriations hearings and different things happening where members have three and four hearings at the same time. So my apologies for the abrupt, but I just want to know, you know, we are, anyone who wants to add a document is going to be allowed to add it to the record. And with that, I just want to thank you again. And with that, we are adjourned.