In recent discussions among technology executives, national security analysts, and U.S. officials, a consensus has emerged: the United States must secure its position in the AI competition against China. In October 2024, National Security Adviser Jake Sullivan cautioned that the U.S. risks ‘squandering [its] hard-earned lead’ without rapid and comprehensive AI deployment to bolster national security.
To address this challenge, the current Biden administration and the reluctant second Trump administration have enacted a dual strategy aimed at achieving AI supremacy. This strategy involves both constraining China’s technological growth through export restrictions and fostering domestic innovation to advance foundational AI models.
Through minimal regulatory oversight, targeted investments in semiconductors, and the promotion of AI adoption in federal agencies, particularly within defense and intelligence sectors, the U.S. has managed to maintain its lead in both market share and technological performance against its Chinese counterparts.
However, experts caution that this lead may not be permanent. Breakthroughs by notable Chinese AI firms like DeepSeek, Alibaba Cloud, Baidu, and Tencent indicate that the disparity in advanced AI capabilities is decreasing, and that U.S. dominance is increasingly at risk. With the race for AI supremacy intensifying, Washington must acknowledge the possibility of a future where Chinese AI technology holds global popularity.
The prospect of falling behind does not mean that the U.S. is destined for failure, as seen in the case of 5G technology. Instead, the U.S. can focus on developing comprehensive frameworks that recognize the appeal of AI in emerging markets. By facilitating easier migration between different AI models, creating systems for comparing outputs, and sharing U.S. data securely with developers and allies, Washington can position itself to benefit from the AI revolution, even if it does not lead it.
As of mid-2024, U.S. companies appeared to have effective mechanisms for maintaining their AI dominance. The collaborative environment of academia, private investment, and light regulation fostered rapid model innovation. Foundational models such as OpenAI’s GPT and Google’s Gemini made significant advancements since the advent of ChatGPT in 2022.
By the end of 2023, U.S. models outperformed their Chinese counterparts by notable percentages in response accuracy. Nonetheless, China has swiftly gained ground through initiatives such as the Next Generation AI Development Plan, extensive public investment in research, and strong incentives for workforce development.
These advancements have allowed China to narrow performance gaps to single-digit percentages by late 2024. Recent achievements by companies like DeepSeek and Qwen have further raised concerns that the United States’ previously robust lead is eroding.
Additionally, China has made significant strides in the practical integration of AI into high-tech manufacturing. For instance, Xiaomi employs over 700 AI-assisted robots at its Beijing facility, producing a new electric vehicle approximately every 76 seconds.
Chinese cities have widely adopted AI technologies for traffic management, surveillance, and law enforcement, while local governments actively explore AI-innovation zones to enhance governance, healthcare, and education.
Moreover, the effectiveness of U.S. export controls in limiting China’s access to advanced chips has not lived up to expectations. Reports suggest that China has adeptly circumvented these measures, relying on shell companies and accelerating its domestic semiconductor development to build a more resilient tech industry.
Pioneering software techniques that maximize existing hardware, Chinese firms have optimized training and inference times, suggesting that while the contest for supremacy remains tight, the days of unquestioned American AI dominance may be numbered.
Despite these developments, U.S. technology firms still have opportunities to maintain leadership in foundational AI model development. Notable investments, such as OpenAI’s collaboration with SoftBank and Oracle on the $500 billion Stargate AI infrastructure project, demonstrate the ongoing commitment of the American tech industry to drive innovation.
Cloud computing giants like Amazon, Microsoft, and Google collectively command more than 60% of the global cloud market, a fundamental resource for developing and deploying AI models. However, the rapid pace of advancements in AI technology has led to concerns that this momentum may not be sustainable, especially as the challenges facing innovation become more complex.
As many experts anticipate a shift toward coexisting AI ecosystems, Washington should proactively strategize to ensure that U.S. firms can thrive in a landscape where both American and Chinese AI models compete.
One essential strategy could involve showcasing the advantages of American models in international markets. The U.S. National Institute of Standards and Technology can spearhead new evaluation frameworks for foundational AI models, transcending traditional benchmarks that focus solely on raw capabilities like language understanding and reasoning.
Incorporating metrics for transparency, accountability, and cost-efficiency would enable U.S. models to appeal more broadly to emerging markets. Additionally, as model diversification increases, consumers will lean towards versatile offerings that minimize the transition costs associated with switching between different AI systems.
To this end, U.S. companies must focus on reducing migration expenses and risks associated with adopting new models. One potential avenue for enhancing consumer confidence in the transition process is to standardize application programming interfaces across foundational models, facilitating seamless data exchange between systems.
This requirement ensures that the global AI market remains competitive, allowing users to tap into the advantages of multiple foundational models without becoming beholden to one specific provider.
The increasing prominence of Chinese foundational models also compels U.S. technology firms to create applications that act independently from any singular model while addressing inherent risks, such as the possibility of unreliable outputs or compromises of sensitive data.
Implementing intermediate abstraction layers can shield downstream applications from the direct dependencies on foundational models. This independence allows for quick shifts in response to model improvements or failures, thus preserving application performance and continuity.
Nevertheless, the widespread adoption of Chinese models does carry significant risks, including vulnerabilities related to incorrect outputs, potential data breaches, and interruptions in services. In light of these concerns, U.S. firms must develop robust adjudication systems to verify outputs from both trusted and untrusted models, ensuring that responses align with user expectations and standards.
In identifying the appropriate balance between innovation and security, U.S. regulators will need to set guidelines for data sharing with foreign model developers. While the risks associated with sharing U.S. data with China remain substantial, specific situations may warrant such exchanges, particularly if a Chinese model proves vastly superior in areas like medical diagnostics.
In these cases, the potential benefits of utilizing advanced foreign AI models could justify an assessment of the associated risks, especially if safeguards are in place to protect sensitive information.
It is imperative that Washington standardizes new evaluation metrics and formulates comprehensive guidelines for data sharing with allies and partners. Additionally, technical assistance should be made available to suppliers struggling to migrate between models and to establish the necessary adjudication systems for competitive models.
In navigating the evolving AI landscape, U.S. policymakers must recognize that absolute dominance is no longer a given. Striving to maintain their lead remains essential; however, embracing a strategic approach that accommodates the realities of global competition will ultimately benefit American AI advancements.
A failure to adapt could impose a dire scenario where the U.S. faces a formidable rival empowered by AI while its domestic industry struggles and remains hamstrung in its ability to leverage innovative technologies from abroad.
In conclusion, acknowledging the possibility of a second-place finish in the AI race does not signify a defeat for American innovation. Rather, it invites a call for adaptation, creativity, and collaboration within the AI ecosystem.
As the stakes continue to rise in the competition for AI dominance, the path forward for U.S. companies will be shaped not only by maintaining current leadership but also by preparing for a complex and competitive future in the global AI landscape.
image source from：https://www.foreignaffairs.com/united-states/what-if-china-wins-ai-race
