As U.S.-China tech competition heats up, Washington is slowly recognizing that gaining a first-mover advantage in critical technologies may be more vital than protecting its existing edges. At present, the U.S. national strategy aims to slow down its competitors and look to the effectiveness of stronger export controls, stricter enforcement, and measures to block strategic transfers to rivals. Yet as supply chains become more diverse and complex, the range of options to evade such sanctions grows—and the role of third-party intermediaries becomes more critical.
Since 2018, under both the Trump and Biden administrations, the United States has imposed sweeping restrictions on China, including Commerce Department “entity list” designations on certain companies (Huawei, SMIC, etc.); semiconductor export controls (announced in October 2022); and bans on some of the advanced chips needed for artificial intelligence technology, such as Nvidia’s A100 and H100 (imposed in October 2023).
On Jan. 15, just days before President Donald Trump took office, the U.S. Department of Commerce’s Bureau of Industry and Security released two rules: one that updates export controls on advanced computing semiconductors and another that places additional companies based in the People’s Republic of China and Singapore on the entity list.
Just days later, the “America First” trade policy—released on Jan. 20, the first day of the Trump administration—called for key officials, including the secretaries of state and commerce, to review the U.S. export control system in light of developments involving “strategic adversaries.”
The recent success of Chinese AI company DeepSeek has sparked calls for further measures. The U.S. House select committee on China has called for a review and strengthening of controls on chips, making specific reference to DeepSeek’s “extensive use” of Nvidia’s H800 chip, which lawmakers said was deliberately designed to fall outside of the scope of U.S. controls.
Nvidia’s H800 chips—which were modified to operate at around half the peak chip-to-chip bandwidth of Nvidia’s more powerful chips—were designed to comply with the U.S. export controls released in October 2022. Since then, Nvidia has announced plans to introduce new AI chips for Chinese market following U.S. bans on A800 and H800, with scaled-back computing power to navigate around the increased U.S. restrictions.
In February, U.S. officials launched an investigation into whether DeepSeek bypassed export restrictions by acquiring Nvidia semiconductors via Singaporean intermediaries. A Jan. 31 report published by leading semiconductor research and consultancy firm SemiAnalysis contained a comparative analysis of DeepSeek’s model vs. Anthropic’s Claude 3.5 Sonnet large language model—which, according to publicly disclosed data, the researchers found cost “$10s of millions to train.” Surprisingly, though, SemiAnalysis estimated that DeepSeek invested more than $500 million on Nvidia chips. In addition, SemiAnalysis reported that DeepSeek had access to 50,000 Hopper GPUs—graphic processing units, a type of chip—including the H800 and H100 chips, despite the company’s low-cost AI claims.
If DeepSeek has access to such a large number of Hopper GPUs, then the company has significant computational resources at its disposal. So far, DeepSeek has not released any public statement about SemiAnalysis’ claims, although it has released statements regarding its mission to promote open-source AI and its commitment to advancing the field of general artificial intelligence.
In response, the Singaporean Ministry of Trade and Industry denied that export-controlled Nvidia chips were being diverted to China and said that the country is in compliance with U.S. export laws and local regulations.
These worries are just part of broader concerns about the effectiveness of export controls in limiting China’s AI progress. The Chinese government has been supportive of the technology’s development, with national initiatives such as the Next Generation AI Development Plan, published in 2017, which aims to make China a global AI leader by 2030. Other than DeepSeek, Chinese firms such as Baidu, Tencent, Alibaba, SenseTime, and iFlytek are leading the charge by working on a range of AI applications, including facial recognition, natural language processing, and computer vision. These firms are independently advancing AI projects backed by state-led bank financing—whether by circumventing export restrictions, accelerating AI development, or finding alternative means to acquire critical technologies for their high-tech ambitions.
Third-party countries have played a significant role here. Reuters reported in early February that Chinese firms have reportedly obtained restricted chips via hubs such as Singapore, the United Arab Emirates, and Malaysia, which serve as reexport points. During my research, I found concerns about GPU restrictions in several countries, including Malaysia and Taiwan. Although concrete evidence is scarce, Taiwan’s geographical proximity to China has sparked worries about potential smuggling, particularly given the country’s relatively lax GPU restrictions.
Additionally, in July 2024, the Wall Street Journal reported on the existence of informal markets leveraging supply-chain blind spots to bypass US export controls. Another approach has been stockpiling chips before U.S. bans take effect. Tech giants such as Huawei and SMIC had acquired significant reserves of U.S. chips before tighter export controls were implemented in 2022 and 2023. Firms that produce AI products—such as ByteDance and Alibaba—also rushed to secure Nvidia’s A100 and H100 GPUs in anticipation of restrictions.
China is also advancing domestic alternatives, a strategy that has long been pushed by Chinese President Xi Jinping as part of the “Made in China 2025” policy program. Huawei, in collaboration with SMIC, developed the 7 nm Kirin 9000s chip. Compared to its predecessor, the Kirin 9000s falls behind in power efficiency and graphics workloads, with a 33 percent deficit in GPU performance. This means that it may not be as competitive as other flagship chips on the market. While SMIC still lags behind TSMC and Samsung, it is making strides in reducing Chinese reliance on foreign semiconductors.
To circumvent restrictions, Chinese firms have also turned to acquiring older-generation chips by ramping up investment in the production of these legacy chips. This could have significant implications for countries in the European Union, which are competing in parallel with China to revitalize their own tech industries. The need to compete with these heavily subsidized Chinese players could derail the EU’s plans and make European companies dependent on older-generation Chinese processors.
China has also launched tit-for-tat measures to defend itself from export controls and restrictions unleashed by the United States and its allies on its access to semiconductors. In September 2024, China warned of economic retaliation against Japan if it further restricted sales and servicing of chipmaking equipment to Chinese firms. The Chinese Foreign Ministry has spoken out against what it called the United States’ “overstretching of the concept of national security, abuse of export control measures, and malicious blockade and crackdown on China.”
China is not the only player in this game. Middle and smaller powers are responding to supply chain risks and vulnerabilities by diversifying. Singapore plays a strategic role in the U.S.-China tech rivalry due to its neutrality and position as a global trade hub. The city-state facilitates trade and collaboration between the two countries in both the AI and cloud computing industry. Major semiconductor companies, such as GlobalFoundries and Micron, operate in Singapore, which also serves as a crucial transit point for chip exports, including Nvidia’s hardware.
Chinese firms may further pursue backdoor channels to gain strategic access to chips. The Singapore-China Smart City Initiative could serve as a bridge between China’s tech ambitions and the global market, potentially rivaling Shenzhen, dubbed China’s Silicon Valley, and positioning Singapore as a new center for global innovation in AI, semiconductors, and digital economies in the region.
An often-overlooked middle power is India, which is emerging as a nascent semiconductor powerhouse. Unlike China, which has invested heavily in building its own domestic industry, India has focused on design and software development, becoming a hub for global tech companies such as Texas Instruments, Nvidia, and AMD.
India’s trade agreements with both the United States and China also make it an attractive location for countries looking to diversify their supply chains. The country is also taking bold steps to build its semiconductor market, which the India Electronics and Semiconductor Association estimates is expected to grow from $52 billion in 2024 to $103.4 billion by 2030.
India is also ramping up efforts to develop foundational models focusing on AI applications in critical sectors such as agriculture and climate change. According to Information Technology Minister Ashwini Vaishnaw, six major developers are expected to build AI models by the end of the year, aiming to position India’s AI capabilities among the world’s best.
To support this push, India plans to establish computing capacity exceeding 18,000 GPUs, with companies such as E2E Networks and businessman Mukesh Ambani’s Jio Platforms competing to develop this infrastructure using Nvidia’s H100 chips. According to Vaishnaw, the average cost per AI compute unit (CU) is $1.31 per hour.
The number of CUs required to power AI software is influenced by several factors, including the type of AI application, the complexity of the model, the volume and velocity of data, and the desired performance level. For small-scale AI applications, typically 1 to 10 CUs are sufficient. Medium-scale AI applications usually need between 10 and 100 CUs, while large-scale AI may require anywhere from 100 to 1,000 CUs or more. High-performance AI systems, designed for demanding tasks, often require 1,000 to 10,000 CUs or even more.
India’s reliance on Nvidia’s technology is similar to that of other countries. New Delhi anticipates $30 billion in private investment for data centers over the coming years, with the government planning to subsidize 40 percent of computing costs for selected AI projects. India’s reliance on Nvidia’s technology will likely provide the backbone for an AI-driven economy.
Rather than viewing third-party countries as undercutting its efforts, the United States can work with them for mutual benefit. To gain a competitive edge in the AI race to the top, Washington should strengthen cooperation with Singapore. This could include enhanced monitoring of AI chip exports, stricter compliance, and joint cybersecurity initiatives, as well as increased U.S. investment in Singapore’s tech sector, particularly in AI research and semiconductor manufacturing, and will deepen bilateral ties.
The United States could leverage Singapore’s role in the Indo-Pacific Economic Framework for Prosperity to promote transparent AI governance and digital trade standards. Balancing security concerns with support for Singapore’s neutrality will ensure a stable and resilient tech ecosystem in the region. The United States may also wish to work with the Quadrilateral Security Dialogue (which also comprises Japan, Australia, and India) to build resilient supply chains for semiconductors.
The United States may also find greater strategic success by prioritizing domestic innovation rather than solely focusing on restricting China’s technological advancements. While export controls can slow China’s access to cutting-edge semiconductor and AI technologies, they are not a long-term substitute for strengthening the United States’ own innovation ecosystem.
There are other reasons why containment does not work. First, export controls, especially on semiconductors and AI, have spurred innovation in China. Despite restrictions, Chinese companies have found ways to adapt and innovate—particularly since 2017-2018, when AI competition intensified. U.S. sanctions have encouraged companies in China to build a semiconductor ecosystem.
This includes capital investment in companies such as SMIC and other suppliers, which strengthens the broader semiconductor and smartphone industries. The latest AI diffusion rule, which limits GPU purchases for countries outside tier-one nations, could have negative consequences. It may strain the Washington’s relations with other countries and could be seen as overly restrictive, especially for countries not involved in smuggling.
By fostering conditions conducive to indigenous technological breakthroughs, Washington can maintain its competitive edge in AI, semiconductors, and other critical industries. This requires increased investment in research and development, robust public-private partnerships, and an industrial policy that supports emerging tech start-ups. Additionally, it will be essential to ensure a steady pipeline of skilled talent through STEM education and immigration policies that attract top global researchers.
The U.S. STEM industry is facing a significant overhaul, as the Trump administration’s budget proposals have consistently called for cuts to funding for STEM education programs and the National Science Foundation. But one silver lining might be Trump’s plans to invest in AI infrastructure in the country with the announcement of Stargate. Under the proposed $500 billion project, OpenAI will teaming up with SoftBank and Oracle to build multiple data centers for AI in the United States, with the goal of creating hundreds of jobs and securing U.S. leadership in the technology. The data centers could house chips designed by OpenAI as the tech firm aggressively builds out a team of chip designers and engineers.
The United States might need to take a long-term approach to not only slow down the competitor, but also to look beyond advancing its own domestic industrial policy to lead on the AI front. This might include the diversification of chip supply chains; seizing on the opportunity to coordinate with like-minded partners to preempt China from defining the rules and standards for regulating platforms and technological integration; securing funding for open-source projects; and driving secure research to maintain AI leadership. But all these must be done while promoting openness and mitigating misuse, fostering a competitive edge and aligning with national priorities.
A critical area for growth is investing in digital and technological infrastructure in the global south. The United States has made some progress through initiatives such as its Smart Cities Partnership with the Association of Southeast Asian Nations, an initiative that promotes ethical AI use, sustainable urban development, and collaboration on technological innovation across Southeast Asia. This partnership has been particularly welcomed by countries such as Singapore, reflecting its value in the region. Capitalizing on the slower rollout of 5G in Southeast Asia compared to other regions, the United States can invest in these areas to solidify its position as a reliable partner in the Indo-Pacific in the digital age.
As the AI race accelerates, the United States must pivot to a strategic long game, deftly balancing its Indo-Pacific engagement with domestic priorities and acknowledging that the quest for AI supremacy is a marathon, not a sprint—a decadeslong contest of innovation, perseverance, and ambition that will define the future of global leadership.
The post Containment Can’t Win the U.S-China Tech Race Alone appeared first on Foreign Policy.
