Control and Competition: Export Controls in the Age of AI
By: Sahil Shah
Artificial intelligence sits squarely at the core of strategic competition between the United States and China, not solely as a frontier technology but also as a general-purpose capability that is of immense importance. Economically, it will drive cross-sectoral innovation and productivity gains, and the country leading in AI will capture a disproportionate share of future economic growth and global exports. From a security lens, AI leadership translates to major operational advantages that can shift the military balance in flashpoints in Taiwan, be it in intelligence analysis, targeting precision, or autonomous systems. Washington also stands to gain geopolitically from AI leadership as it gives itself more attractive technology to offer to partners, strengthening coalitions and reducing their need to adopt Chinese systems.
Semiconductors provide the computing power, data storage, and networking capabilities that drive AI infrastructure, models, and applications. For example, they power the training and inference clusters — specialized groups of graphics processing units (GPUs) — that support large language models (LLMs) like ChatGPT. Semiconductors are also critical to agentic AI — widely termed as the next frontier of artificial intelligence for its ability to independently complete different tasks — by providing the backbone that executes and links sensing, reasoning, and action in a single, continuous loop. As a result, successive U.S. administrations that have made it a priority to preserve American technological moats in AI have placed an emphasis on export controls that restrict the sale of advanced semiconductors to China.
During President Trump’s first term, his administration intensified scrutiny of China’s access to advanced semiconductors and semiconductor manufacturing equipment (SME). Beginning in 2018, senior officials persuaded the Dutch government to curb the sale of extreme ultraviolet (EUV) lithography machines to China. These machines — used to print extremely intricate patterns onto microchips — are integral to the chip-making process and are extremely complex, with over 100,000 individual components. They are only produced by the Dutch manufacturer ASML, so this restriction greatly dented Chinese aspirations of onshoring and controlling the semiconductor supply chain. In 2019, the administration began placing major Chinese technology firms like Huawei on the Entity List, which is a blacklist of firms maintained by the U.S. Department of Commerce upon whom extremely stringent export licensing requirements are applied.
The Biden administration further expanded these restrictions, most notably through its sweeping October 2022 export controls package. These regulations targeted entire classes of advanced semiconductors, SME, and AI-related technologies. They introduced technical thresholds like floating-point operations per second (measuring computational speed) and memory bandwidth (measuring data transfer speeds between a computer’s main memory and processing units) that governed which technologies would be cleared for export to China. They also applied the Foreign Direct Product Rule to restrict the China-bound sale of products produced by other countries if made using U.S.-origin software and equipment, further constricting Chinese access to advanced SME and technologies.
The second Trump administration has been much more ambivalent and unconventional in its export control policies. Early in his second term, President Trump suspended the export of additional advanced AI semiconductors to China over national security concerns. However, in mid-2025, his administration negotiated an agreement with manufacturers like Nvidia and AMD to re-open some exports to China in exchange for remitting 15% of revenues from such sales (later increased to 25%) to the U.S. government. Policy analysts critiqued the agreement as neglecting long-held security concerns, while legal experts claimed it violated the constitutional ban on export taxes. Though the merit of these arguments is certainly open to discussion, it is indisputable that the administration views export controls less as a hard stop for security and more as leverage for extracting economic concessions.
U.S. export controls have clearly bitten into China’s ability to access advanced chips. Washington has drawn a performance line at the level of Nvidia’s A100/H100-class chips and is attempting to ensure that Chinese firms cannot produce or acquire chips above that level, such as by preventing those firms from using top-level foundries (i.e., chip manufacturers) like TSMC to manufacture powerful chips for them (e.g., 7 nm or below). Additionally, allied controls on ASML’s EUV tools prevent Chinese fabrication plants (“fabs”) from producing state‑of‑the‑art 5 nm and 3 nm chips at scale.
That said, in some cases, these regulations have driven China in more efficient and creative directions. Chinese firms like Huawei and Biren are designing advanced AI chips that are fabricated using deep-UV (DUV) because of allied export bans on EUV machines while High-Flyer introduced the DeepSeek LLM that matched, and even exceeded in certain areas, the performance of U.S. LLMs across several benchmarks and use cases despite being trained on lower-performance chips. Additionally, recent reports published by Reuters indicate that China has built, or is close to building, an indigenous EUV lithography prototype machine. If confirmed, this is a large leap forward for China’s technology sector and AI leadership ambitions that will unlock the near-complete control of its semiconductor supply chain and enable chip production at unmatched scale, considering China’s expansive advanced manufacturing base.
The Chinese government has also retaliated against American regulations, using security reviews, investigations, and ad hoc regulatory actions against selected U.S. firms, most notably declaring chip manufacturer Micron as a “major security risk” in 2023. It aims to signal that American companies could lose access to the Chinese market if Washington tightens restrictions. Beijing has also begun to mandate that publicly owned data centers source at least 50% of their chips from domestic manufacturers, a policy intended to reduce reliance on American vendors and accelerate the scale-up of Chinese chipmakers. Furthermore, Beijing’s industrial policy includes investing tens of billions of dollars into state-backed semiconductor investment funds and expanding export controls on critical resource inputs — such as gallium, germanium, and several rare‑earth metals — used in semiconductors and high‑end electronics, choking American supply chains.
Many factions in Washington seek to respond to these developments in-kind, with even stricter regulations. Members of Congress have floated several such proposals, like the GAIN AI Act, which would require U.S. chipmakers to prioritize domestic customers before exporting AI chips to “countries of concern,” and the SAFE Chips Act, which seek to lock in stricter export controls and curb the administration’s ability to approve licenses over the next few years. Additionally, the recently-passed 2026 NDAA includes a Comprehensive Outbound Investment National Security regime that allows the administration to block or sanction U.S. investments in sensitive Chinese sectors, including AI and semiconductors.
It is unclear whether these policies will effectively curb Chinese technological progress or end up catalyzing further resourcefulness and innovation that advances Beijing’s aspirations. Each regulation makes a short-term incision but risks fueling the very progress it seeks to limit. As it stands, control and competition are inseparable.
