Senate's SAFE Bill Targets China AI Chips as Military Use Raises Stakes in Export Controls Debate

The Senate's new SAFE bill is set to curb access to advanced chips to China, but that won't slow down the AI war -- training workloads still heavily rely on Nvidia, while alternatives remain inefficient

A new bipartisan bill in the Senate could pause shipments, but there are ways around it. A new bipartisan bill in the U.S. Senate threatens to put the brakes on Nvidia's efforts to sell its latest AI-training hardware to Chinese customers, even as the Trump administration mulls allowing lower-powered versions of the hardware. China is looking to restrict access to this kind of hardware too, to favor domestic chip firms, which would harden its supply chains and reduce trading turbulence. However, with no real alternatives to Nvidia's GPUs for training hardware and numerous ways to circumvent sanctions, tariffs, and trade barriers, it's hard to imagine Nvidia completely exiting the region. Nvidia CEO Jensen Huang spent much of last week meeting with U.S. legislators, including President Trump and Republican members of the Senate Banking Committee, which oversees U.S. export control programs. Huang clearly wasn't persuasive enough, though, as now the proposed Secure and Feasible Exports Act (SAFE) bill would force the Commerce Department to halt export licenses for the sale of the latest chips to U.S. adversaries, including China and Russia, for 30 months. This ban could cover all existing chips and anything more powerful than them developed by any of the major companies over that same period. Although it primarily targets Nvidia's Blackwell GPUs, it would also cover Nvidia's last-generation Hopper designs, AMD's graphics chips, and Google's latest TPU designs. This is devastating news for Nvidia and many of its chip-manufacturing contemporaries. China is a massive market for hardware and AI development, but it's certainly not proven to be the most willing of markets. Chinese authorities have spent months pushing back on the on-again, off-again availability of Nvidia hardware by encouraging its domestic companies to use domestic chip suppliers where possible. It mandates that Chinese companies use at least 50% domestically produced hardware and, more recently, has claimed that new packaging and assembly techniques can close the performance gap between Nvidia and its local producers. Chinese chip firms have responded with gusto, too, announcing enormous plans to manufacture several times the chips they managed in 2025, as soon as next year. It's not clear if those plans will be physically possible in such a short time frame, but they're shooting for the moon nonetheless. But even if the companies can fabricate these chips, there's no guarantee they'll be used, despite the double-ended carrot-and-stick approach of both the U.S. and Chinese authorities. China has made major leaps in its AI hardware development over the past few years, particularly in the past year, as it's sought to build more reliable access to powerful AI hardware, while the U.S. turned the tap on and off at the whim of its mercurial commander-in-chief. These conditions have led Huawei to make tremendous advances and to design high-power systems that scale well, at the expense of efficiency. But that's mainly in the realm of inference, which is the day-to-day running of an AI algorithm after it has been fully trained. Nvidia's GPU versatility is particularly well-suited for AI training, and it has no real rival. There have been some semi-hyperbolic claims of a new Chinese chip design that leverages 3D hybrid bonding techniques, and is claimed to deliver performance comparable to 4nm Nvidia silicon in training workloads. Given the restrictions in place for China's access to EUV machines from ASML, it's an interesting area of expansion. It's not proven yet, and questions remain over its efficiency, how manufacturers would handle thermal dissipation - memory and compute bonded directly raises serious overheating concerns - and such a complicated design could lead to yield issues when produced at scale. But even if all the claims about this hardware prove true and it's indeed a relative competitor to Nvidia, why wouldn't the companies that need this hardware at scale right now not just keep using Nvidia anyway? When Deepseek developers were forced to use locally produced chips for training, they ended up switching back to Nvidia hardware because the performance just wasn't there. Despite all the blocks and barriers from various governments and organizations, it hasn't been too difficult for companies to allegedly get their hands on. Singaporean companies have been used to allegedly circumvent trade blocks, and leasing computing power from international partners effectively allows Chinese national companies to use whatever hardware they like. There are always mules willing to help get the hardware across the border for a fee, too. So, even if new barriers are put in place to make it harder for Nvidia to ship hardware to China, it will probably still happen. It's better for training than anything Chinese producers can make, it's still readily available, albeit through ever-more convoluted routes, and the companies that want the hardware are trying to compete with markets that have better access to it. As Deepseek 3.2's latest whitepaper shows, the race for AGI is now entering the stage where those with the best pre-training compute might push ahead with breakthroughs. Now, the AI race is turning into a question of scale, regardless of who is making the chips.

China does not need Nvidia chips in the AI war -- export controls only pushed it to build its own AI machine. | Fortune

But they are wrong. These arguments assume that China cannot succeed in AI without access to these advanced AI chips, which is not the case. Advanced AI chips simply reduce the cost of AI. Today's state-of-the-art AI models require a large number of AI chips to build and run. An advanced chip has higher performance; therefore, fewer are needed to achieve the same AI performance. But AI costs can be reduced in other ways. As DeepSeek showed, clever software and algorithm design can dramatically reduce the number of AI chips needed. China's decision to open-source its AI models particularly allows it to leverage the best software and algorithms to reduce AI costs. Second, AI chips constitute only part of the overall costs. AI-based systems incur several other costs - engineering, data, software and licensing, regulations, energy, and infrastructure - where China has considerable cost advantages. Third, AI hardware performance depends greatly on packaging and interconnection - how AI chips are put together and connected. China can leverage its world-class strengths in both to achieve high performance. Recently announced Huawei SuperClusters are more powerful than any Nvidia system, despite not using the most advanced AI chips. Advanced chips also reduce the power cost of AI. These chips are manufactured using the latest technology from TSMC (and sometimes Samsung) - each new technology is more energy efficient than the last. High power consumption of an AI system worsens monetary cost and the speed of deployment since fast access to a large amount of power is challenging, especially in the U.S. However, China is growing its power supply much faster than the U.S. and is much more likely to successfully serve the power demands of its AI data centers, even if they consume more power due to lack of access to advanced AI chips. High power also leads to greater carbon footprint, but it should not limit Chinese ambitions in any technology it considers important. Besides, many AI applications do not need advanced chips. Several applications in network security, facial recognition, medical image analysis, advanced driver assistance systems (ADAS), logistics, and robotics can be handled using AI models much simpler than state-of-the-art models. These models can be built and run on chips that China can produce itself. China aims to dominate these applications. Even for more complex applications, recent work suggests that state-of-the-art models can be replaced by a collection of much simpler models. This collection does not need advanced AI chips to build and run. So, it is unclear if China will be left behind for these applications either. It is also not clear whether future development and use of state-of-the-art models will require advanced chips. There are signs that the benefits of state-of-the-art models are plateauing. Given the large investments these models require, future models may look different and use fewer resources, including chips. It will further level the playing field, even if access to advanced AI chips is controlled. There is also a possibility that China may learn how to produce advanced AI chips itself - it has certainly invested in several technologies with the potential to leapfrog past the state-of-the-art. Overall, China can significantly mitigate the disadvantages of not having access to advanced AI chips. Besides, China will be willing to absorb any higher upfront costs, especially for AI-based military and strategic technologies, since they know that they can reduce the downstream costs through scale and manufacturing strengths. Unsurprisingly, China continues to produce competitive state-of-the-art models and dominate AI-based applications such as robotics and autonomous vehicles despite the AI chip controls implemented over the last several years. The argument for AI chip controls may still make some sense - why not get the advantage of increasing AI development costs for China, however small, if there were no cost to it. But the costs are significant. China could have been one of the largest markets for U.S. advanced AI chip companies. The U.S. has lost the market. Second, AI chip controls have made this an issue of national pride and led to a wave of investments into a domestic AI chip ecosystem within China. It is unclear if the U.S. will ever regain market share even if chip controls are reversed. China has also retaliated in many ways - those measures have further hurt the U.S. economy and geopolitics. If the U.S. wants to lead in AI, chip controls are not the answer. Instead, it should focus on improving innovation, investment, energy, and regulatory ecosystems. It should make it easier for the best AI scientists in the world to live and work here. It should diversify, strengthen, and secure AI supply chains. It should work with allies to lead the development of international AI standards and practices. It should reduce the cost of AI (through selective open sourcing or public-private partnerships, for example) to ensure that American AI (alongside its values) is most pervasive. It should prioritize high-end and enterprise applications where the moat is wider against a talent and resource-rich fast follower that has cost and speed advantages. The value of AI chip controls is vastly exaggerated. These controls have barely slowed China down and caused significant economic and geopolitical damage to the U.S. It is time to abandon them and focus fully on maintaining and growing AI lead through innovation instead.

China wants an AI-powered military built with Nvidia chips, and that's a problem

Despite much speculation, President Trump did not allow China greater access to advanced AI semiconductors during his meeting with Chinese leader Xi Jinping at last month's Asia-Pacific Economic Cooperation summit. Although Trump balked in Seoul, he is reportedly considering greenlighting sales of advanced chips to Beijing. It would be a mistake, however, to expand the flow of one of the United States' most powerful strategic assets to China, especially with new evidence of the Chinese military's desire to harness advanced chips for battlefield advantage. Nvidia CEO Jensen Huang has played a central role in persuading Trump's team that the U.S. should promote its global leadership in AI by selling ever-more advanced semiconductors abroad. He has claimed that U.S. restrictions on the export of advanced semiconductors to China will only spur Beijing's capabilities across its domestic semiconductor supply chain. At the same time, he has argued that Chinese companies are only "nanoseconds" behind their U.S. competitors in designing and fabricating cutting-edge chips. He has repeatedly downplayed the risks of exporting advanced chips to China, claiming that that U.S. semiconductors will not enable China's military modernization. But our analysis suggests the opposite. We reviewed dozens of procurement documents published by the People's Liberation Army which reveal that the Chinese military is directly soliciting and using advanced U.S. chips, including those designed by Nvidia, to develop AI-enabled military capabilities. In addition, the PLA is deploying Chinese AI models trained using American hardware to advance its modernization and, eventually, gain a battlefield advantage over the U.S. These documents clearly state the PLA's intention to use Nvidia chips for a wide range of tasks. For example, one contract for an "intelligent optoelectronic target recognition system," which combines AI and sensors to automatically detect, identify and track militarily relevant objects, specifies the use of Nvidia computing resources. Another notice for the procurement of a server to help the Chinese military "perform AI algorithm calculations" and run large language models relies on Nvidia H100 GPUs -- a chip that was export controlled in 2022. A third document requests a cluster of Nvidia A800s, which are also controlled, for a "high-performance image algorithm training workstation," that would presumably be used to develop AI systems for image-processing tasks. Finally, one notice asks for "autonomous vehicles equipped with Nvidia's Jetson Orin chips," which provide the onboard computing power necessary to process visual information. To be sure, the PLA's procurement of Nvidia hardware is unlikely to make or break Beijing's military modernization ambitions. The Chinese military is still experimenting with AI applications, and there is little evidence that the export controls will dramatically shift the balance of power between the U.S. and China in the near term. But it is unrealistic to believe that the PLA will not use the world's most powerful AI chips to advance its military capabilities. China is requesting such technologies to develop emerging capabilities that would allow it to outcompete its adversaries. Indeed, authoritative Chinese documents indicate that Beijing believes the development and deployment of advanced AI-enabled military systems provides the best chance to catch up to or surpass the U.S. military. Beyond the contracts explicitly listing Nvidia products, we have seen hundreds of PLA contracts that reveal evidence of Beijing's investment in a wide range of AI-powered military capabilities. For example, the PLA is requesting systems that can generate, collect and analyze troves of battlefield data to quickly identify targets and accelerate decision-making cycles. Other documents feature requests for algorithms to power swarms of autonomous vehicles in the air, on the ground and at sea. Improved access to the world's preeminent computing hardware will only speed China's development of advanced systems that could be used against the U.S. military in a future conflict. But the risks of relaxing the controls are not solely limited to China's ability to acquire Nvidia chips. Just as important, Chinese frontier AI labs will be able to more easily acquire advanced computing hardware to train increasingly capable AI models, which can in turn power military activities. Various procurement documents indicate that China's military is adopting highly capable AI models, including those trained by DeepSeek. Some of the companies responsible for training these systems have noted that U.S. export controls are hampering their progress. While some Chinese companies are deploying their AI systems using domestically-produced chips, spurred in part by Beijing's directives, frontier model training still largely relies on Nvidia hardware. Chinese access to cutting-edge Nvidia chips will make it easier for companies like DeepSeek to develop more capable models that the PLA can then utilize. In ceding control over the chips that power the United States' AI lead, Washington would hand Beijing tools it wants to close the military gap. Export controls were meant to give the U.S. and its allies time to consolidate and build on their advantages in AI. Relaxing them makes it less likely that the U.S. will continue to lead in AI, which could have lasting national security implications. Sam Bresnick is a research fellow and an Andrew W. Marshall fellow at Georgetown University's Center for Security and Emerging Technology. Cole McFaul is a senior research analyst and an Andrew W. Marshall fellow at Georgetown University's Center for Security and Emerging Technology.