Building a Silicon Consensus between the United States and Its Allies

Artificial intelligence (AI) is among the most significant technological developments in decades. Consulting firms like Gartner and BCG call it transformative, and the technology has the potential to reorder the modern economy.

AI algorithms depend on advanced graphics processing units (GPUs) to achieve the computational power necessary to process the massive datasets fueling the AI revolution. The United States is currently the world leader in GPUs because it is home to the most advanced chip companies, such as Nvidia, AMD, and Intel. With emerging technology increasingly at the forefront of U.S.-China competition, some in Congress argue that the United States should suffocate China’s AI industry by completely cutting off the supply of these chips. But this approach could instead accelerate China’s self-reliance, further softening the market for Western chips in the country and weakening Taiwan’s “silicon shield”—a form of deterrence premised on the argument that an attack on Taiwan could disrupt the supply of these crucial components, causing worldwide economic damage, including to China.

Alternatively, a comprehensive plan of building up a platform of managed competition with China, via managed risk and engagement by allowing the monitored yet unrestricted export of high-end American AI chips, all the while creating a technology security alliance to strengthen partners in the region, could provide a more effective path forward. The equipment to make the chips would stay within Taiwan and its allies, keeping China dependent on them for access—and Taiwan’s silicon shield intact.

The Unintended Consequences of an Overly Broad Ban

As China’s advanced technology and AI companies—which are reliant on U.S.-designed chips and U.S. semiconductor manufacturing equipment—become increasingly competitive, concerns over these technologies’ military applications and related national security concerns led to actions from the White House banning their export.

One unintended consequence is that China’s tech companies, having been denied access to Nvidia’s latest powerful GPUs, are now hard at work developing their own. Jensen Huang, the CEO of Nvidia, said that his company has two dozen new competitors in China due to the restrictions.[1] It has also been reported that thousands of gaming-grade GPUs are being dismantled by factories in China and repurposed for AI server farms.[2] China is home to hundreds of thousands of these GPUs—some left over from the crypto-mining boom. This could provide servers with satisfactory capacity until other chips can be procured, allowing companies to evade restrictions. Additionally, to get around these controls, Nvidia has been developing export-compliant versions of their AI GPUs. Each individual GPU is less powerful than the versions sold in the United States, but this simply means users will need more of them. It is less efficient to scale up, but the end result is the same: Chinese AI server farms powered by American chips.

U.S. regulators are not happy with these workarounds. Commerce Secretary Gina Raimondo warned, “If you redesign a chip around a particular cut line that enables them to do AI, I’m going to control it the very next day.” She has promised to enforce the spirit and the letter of the law.[3]

An Asia-Pacific Technology Security Alliance

If China can successfully develop a domestically designed and fabricated AI chip that is competitive with Nvidia, it would no longer be dependent on Taiwan Semiconductor Manufacturing Company (TSMC), which currently produces 90% of the world’s advanced chips.[4] To counter this possible reordering of the global technology order, Taiwan and the United States should consider developing a “technology security alliance,” modeled on a hybrid of NATO and NAFTA.

TSMC’s critical role in the global economy acts as a deterrent against China’s potential invasion of Taiwan. China’s reliance on TSMC chips must be balanced against its foreign policy goals of limiting U.S. influence and integrating Taiwan. Meanwhile, Taiwan’s chip fabrication capabilities also present a security challenge for Taipei and Washington, given U.S. reliance on TSMC chips. The United States should therefore engage Taiwan as a strategic partner rather than impose unilateral policies.

A technology security alliance is where such a partnership of equals could begin. Stakeholders such as Taiwan, Japan, and South Korea could share ideas and strategies in domains like 5G, quantum computing, and AI. Such a working group should include national security, commerce, trade, and economic departments as well as semiconductor ecosystem players, including foundries, integrated circuit design, and packaging firms.

Critical issues this group must address are coordination of supply chain management, technology standards development, policy consultation, and workforce development. Policy discussions around export controls and outbound investments should incorporate the views and considerations of allies and their companies.

Another tangible step is creating a visa program to facilitate the fast-track exchange of skilled tech professionals. With companies like TSMC growing in U.S. states such as Arizona, access to a skilled workforce is vital. This would not pose a threat to U.S. jobs because these types of jobs have never existed stateside as the semiconductor fabrication business matured in Asia. Arizona and other states currently lack the specific vocational training required to create a workforce pipelines; there simply is not the labor available to meet the sector’s demands. By working with Taiwan and other U.S. partners to build a future-proof talent stream, these new fabrication plants could be staffed with well-paying jobs from which the United States stands to benefit.

Another vital area for a technology security alliance is communication between regulators, producers, and clients about policy changes that might affect cross-border business. Allies and partners need to be informed of these developments with appropriate lead time so that firms can make necessary adjustments to reallocate resources and reorganize supply chains.

The success of TSMC in the United States should be a strategic priority for Washington, as the company is integral for building a domestic semiconductor industry. It is important to remember that TSMC’s new fab in Kumamoto, Japan, was completed before the facility in Arizona—which has been delayed because of surging costs and labor disputes. The fact that TSMC’s fab is delayed in Arizona sends a signal to Beijing that U.S. efforts to onshore strategic chips are not going well. If the Biden administration wants to count the CHIPS and Science Act as a success, it needs to ensure that the world’s largest semiconductor factory succeeds in its endeavor to do business in the country. This can only happen as a collaborative effort between the United States and its partners in Asia like Taiwan.

Coordination and Stronger Trust

The Wassenaar Arrangement, enacted in 1996, currently regulates the export of arms and technologies. The agreement certainly was a feat for its time, as it brought together former Soviet-aligned countries, the West, and neutral powers like India. But it is inadequate for addressing modern challenges, particularly given the blurred lines between civilian and military technology in China’s military-civil fusion and the proliferation of digital and dual-use items, such as advanced semiconductor chips. The treaty was not written for an era of complex, globalized supply chains.

These limitations are where a technology security alliance led by the United States and joined by its partners and allies, including Taiwan, could begin. The alliance would encompass a globalized and complex supply chain while providing an accelerated immigration stream for tech talent. It would also create a forum to ensure that material changes to regulation are published in an expedient manner and stakeholders are given a chance to provide feedback.

Unilateral actions such as banning the export of lithography machines to China or changing regulations regarding chip exports to China with minimal notice only work to degrade trust. Instead, a technology security alliance would bring all stakeholders to the table in Taiwan to build bridges through discussion rather than unilateral action.

The world is a global village only because all nations have agreed on universal standards for interconnectivity. Packets can be sent from Beijing to Washington, D.C., and from New York to Moscow because all computers understand the same fundamental electronic language. For now, the United States leads in developing these standards and languages. But this could change. When pushed to the extreme, China will invest heavily, rallying its brightest minds to develop independent technologies. When they succeed, Chinese tech standards might dominate emerging markets, projecting power to countries that rely on China for economic or technology support and undermining U.S. interests. China could also develop technologies unknown to the United States due to a lack of access to its ecosystem, which could put the United States in a detrimental position.

Beijing has done this once before, when Washington restricted shipments of Intel Xeon processors to China’s supercomputers in 2015. Only one year later, China debuted the world’s fastest supercomputer made without U.S. chips. Instead of going down this path again, the United States and its allies need to engage with China to manage risk through monitored yet unrestricted chip exports, which can be turned off entirely if Beijing crosses red lines. Moreover, this policy needs to be developed in conjunction with Taiwan and other allies and partners in Asia.

Taiwan’s measured approach toward China, for instance, means balancing trade with caution against potential aggression. Despite unresolved differences between Taipei and Beijing over the definition of China and how Taiwan fits into it, the two are able to come to a compromise and maintain unofficial relations. China is Taiwan’s largest trading partner, with special laws dictating the relationship that differ from other countries. It is not a free-for-all; inbound and outbound investments are closely monitored, with the latest chip-manufacturing technology staying at home. TSMC has facilities within mainland China, but the process nodes are comparatively ancient to what is being built within Taiwan. Nonetheless, the $205 billion trade relationship supplies the world with electronics, given the two nations’ complementary economies.[5]

A measured approach that similarly balances economic opportunity with political risk would allow the United States to gain strategic insights and influence China’s peaceful use of technology, acknowledging the risk but recognizing the benefits of such engagement. China’s access to the tools it needs to make advanced chips and processors would still be controlled, which keeps it dependent on a U.S. supply that can be turned off. Such an approach would ensure that China and Taiwan can coexist in the global tech landscape while promoting stability.

This theme is echoed in U.S. national security advisor Jake Sullivan’s phrase “a small yard and high fence,” which describes the White House’s intent to allow most trade and economic relations with China to continue, albeit while exerting control and maintaining restrictions for certain parts. Chips would be no different. The tens of billions of dollars of trade between the two nations could continue, but controls would remain in place for certain sectors, like military applications. AI, however, would not be on the list of restricted technologies.

If U.S.-China competition is managed in the wrong way—with the United States continuing to overprioritize national security and implement unilateral policies rather than consult with partners and allies—this could be the start of the United States’ silicon sunset, where the servers that power the developing world’s fast-growing markets are run by Huawei or HiSilicon. U.S. leaders must not let that happen.

---

Jason Hsu is a former legislator at-large from Taiwan and currently a Mason Fellow at Harvard Kennedy School. Sam Reynolds is a Technology Researcher in Taipei.

Endnotes

[1] Olivia Poh, “Nvidia Sees Huawei as Formidable AI Chipmaking Rival, CEO Says,” Bloomberg, December 5, 2023, https://www.bloomberg.com/news/articles/2023-12-06/nvidia-sees-huawei-as-formidable-ai-chipmaking-rival-ceo-says.

[2] Hassan Mujtaba, “Chinese Factories Dismantling Thousands of NVIDIA GeForce RTX 4090 ‘Gaming’ GPUs & Turning Them into ‘AI’ Solutions with Blower-Style Coolers,” Wccftech, November 23, 2023, https://wccftech.com/chinese-factories-dismantling-thousands-of-nvidia-geforce-rtx-4090-gaming-gpus-turning-ai-solutions.

[3] Michael Crider, “U.S. Regulator Threatens Nvidia’s Chinese Chips,” PC World, December 4, 2023, https://www.pcworld.com/article/2161579/us-regulator-threatens-nvidias-chinese-chips.html.

[4] “Taiwan’s Dominance of the Chip Industry Makes It More Important,” Economist, March 6, 2023, https://www.economist.com/special-report/2023/03/06/taiwans-dominance-of-the-chip-industry-makes-it-more-important.

[5] “Cross-Strait Relations,” Government Portal of the Republic of China (Taiwan) https://www.taiwan.gov.tw/content_6.php#:~:text=Today%2C%20Taiwan%20is%20one%20of,trade%20was%20US%24205.11%20billionI.