Key Points:
- A growing consensus of security experts and policymakers argues that the United States must actively deny geopolitical adversaries access to high-end AI compute.
- While software weights and algorithms are easy to replicate, physical semiconductor hardware represents a highly tangible, controllable bottleneck.
- Reports indicate that China leverages Western-designed chips to accelerate military AI, including battlefield-oriented simulations and autonomous drone systems.
- Advocates call for tighter multi-lateral export controls, stricter cloud provider screening, and stronger alliances with chipmakers in Japan and Europe.
The geopolitical race for artificial intelligence supremacy has entered a highly aggressive and hardware-centric phase. As advanced model architectures become more accessible, national security experts are shifting their focus from protecting software code to controlling the physical infrastructure that enables AI. In a prominent opinion piece recently published in the Wall Street Journal, foreign policy and security analysts argued that the United States must implement a far stricter policy to starve its strategic adversaries of advanced computing power, or “compute.” This emerging perspective highlights a critical reality in 2026: in the global technology cold war, controlling high-end silicon is the most effective way to protect democratic interests.
Unlike software algorithms, which developers can easily copy, leak, or distribute across the internet in seconds, physical AI compute is a highly tangible, easily localized bottleneck. Training state-of-the-art frontier models requires thousands of specialized, advanced GPUs operating in massive, power-hungry data centers. This specialized hardware depends on an incredibly complex and highly concentrated global supply chain. This supply chain relies on a small handful of key players, such as ASML in the Netherlands for lithography machines and TSMC in Taiwan for fabrication. Because the physical production of advanced microchips accounts for less than 1% of the global semiconductor market, democratic allies have an extraordinary opportunity to choke off the supply of advanced computing power to hostile nations.
The urgency behind restricting AI compute stems from the immediate military advantages that advanced artificial intelligence provides on the modern battlefield. Intelligence reports suggest that China actively leverages Western-designed semiconductor technology to accelerate its own military AI capabilities. The Chinese military uses these high-end chips to train autonomous drone swarms, automate cyber warfare operations, and run highly complex battlefield-oriented simulations. For instance, Wuhan-based developers recently unveiled the AU100, a vehicle-mounted drone countermeasure system that uses advanced AI to track and neutralize high-speed drones. By allowing advanced chips to flow to strategic competitors, the West is indirectly funding the development of weapon systems that could eventually target its own forces.
The sheer scale of China’s domestic computational buildup illustrates why the United States must act decisively. While the United States’ total AI compute capacity remains approximately 20 times China’s today, Beijing is closing the gap rapidly. Current estimates place China’s total computing capacity at approximately 246 EFLOPS (quintillion floating-point operations per second). Analysts project this figure to reach a staggering 1,117 EFLOPS by 2027, driven by massive state-backed investments and a concerted effort to circumvent international restrictions. Allowing this exponential growth to continue unhindered could eventually erode the technological superiority that the U.S. military has relied on for decades.
Despite existing trade restrictions, significant loopholes continue to undermine the effectiveness of current export controls. Although the U.S. Department of Commerce has banned the direct sale of cutting-edge chips, such as Nvidia’s Hopper and Blackwell architectures, to Chinese entities, a parallel grey market continues to thrive. China regularly circumvents these rules by utilizing shell companies, third-party intermediaries, and illicit smuggling networks in Southeast Asia and the Middle East. Furthermore, many hostile actors simply bypass physical chip restrictions by renting computing power from Western cloud providers, effectively training their advanced military models on U.S.-soil data centers without ever needing to import a single piece of hardware.
This unrestricted export of technology also places an unsustainable strain on the domestic tech sector. A recent report by the Center for a New American Security (CNAS) titled “American AI Companies Can’t Get Enough Chips” points out that the global supply of specialized AI chips remains highly constrained. Building additional fabrication facilities takes years and requires billions of dollars in upfront capital. The CNAS report argues that because domestic chip supply is a hard bottleneck, every single high-end processor that companies export to a strategic competitor is one fewer chip available to a domestic firm or a democratic ally. Aggressively restricting exports to adversaries is therefore not just a defensive measure, but a way to preserve scarce resources for domestic innovators.
To effectively close these loopholes, policymakers must adopt a far more holistic and aggressive strategy. First, the federal government must enforce strict “Know Your Customer” (KYC) rules for cloud computing providers, making it a crime to lease high-end computing to foreign adversaries. Second, export controls must expand to cover enabling technologies, including high-bandwidth memory (HBM) and advanced chip-packaging services (OSAT). Without these critical inputs, even smuggled GPUs become virtually useless for training large-scale, frontier-level military models. By targeting the entire compute ecosystem rather than just individual processors, the U.S. can establish a much more resilient regulatory barrier.
Furthermore, the United States cannot succeed in this effort alone; it must coordinate closely with its democratic allies to build a unified technological front. Initiatives like “Pax Silica” must play a central role in aligning export policies across key semiconductor-producing nations, including Japan, South Korea, and the Netherlands. By establishing common security standards and coordinating enforcement, these nations can prevent adversaries from playing allies against one another to acquire restricted technology. This collaborative approach ensures that the global chip supply chain remains highly secure and aligned with democratic values.
Ultimately, the battle for AI supremacy is an infrastructure war that the West cannot afford to lose. The physical limitations of chip manufacturing, power grid capacity, and global supply chains mean that compute is a finite, precious resource. By implementing tighter export controls, enforcing cloud KYC rules, and strengthening international alliances, the United States can effectively deprive its adversaries of the computational power needed to build dangerous military systems. Protecting the democratic world’s technological lead requires a hard, unsentimental approach to hardware security, proving that the most effective way to govern AI is to control the very silicon it runs on.
