Trump’s Balancing Act with China on Frontier AI Policy

Trump’s Balancing Act with China on Frontier AI Policy

During a Dec. 18 press conference in Mar-a-Lago, President-elect Donald Trump took an unexpected tack, suggesting the United States and China could “work together to solve all of the world’s problems.” With China hawks poised to fill key posts in his administration, Trump’s conciliatory tone contrasts sharply with his team’s overarching tough-on-Beijing stance.

Yet Trump’s history with China suggests a willingness to pair tough public posturing with pragmatic dealmaking, a strategy that could define his artificial intelligence (AI) policy. Such an approach echoes Trump’s handling of the ZTE crisis during his first term in 2018, when a seven-year ban on U.S. firms selling parts to the Chinese telecom giant threatened to cripple the company. Trump reversed the decision in exchange for costly concessions, including a $1.4 billion fine, showcasing his readiness to break from hawkish pressures when a favorable bargain aligned with his goals.

In AI policy, the next administration will likely embrace a transaction-based approach to promote U.S. AI technology abroad and win global market share. Data centers, wide-ranging AI applications, and even advanced chips could all be for sale across the Gulf, Southeast Asia, and Africa as part of a concerted attempt to win what top administration officials often refer to as the “AI race against China.” Yet as Trump and his team are expected to pursue their global AI ambitions to strengthen American national competitiveness, the U.S.-China bilateral dynamic looms largest. With rapidly improving frontier AI capabilities, headlined by substantial capabilities increases in the new o3 model OpenAI released Dec. 20, the relationship between the great powers remains arguably both the greatest obstacle and the greatest opportunity for Trump to shape AI’s future.

Bringing American Innovation to the Global Stage

The United States currently leads the world in cutting-edge frontier AI models and outpaces China in other key areas such as AI R&D. This lead grew first and foremost from the United States’ early investment and accumulation of talent in AI. The lead was extended through export controls first imposed during Trump’s first administration aimed at stifling Chinese access to advanced semiconductors. Earlier this month, the Biden administration expanded its export controls with new restrictions on semiconductor equipment and high-bandwidth memory. These controls are expected to significantly increase the costs associated with the production of China’s most advanced chips. The current lead gives the United States power and leverage, as it has better products to sell than its competitors.

But leading tech policy figures – including some of Trump’s key backers – are concerned that current advantages in frontier models alone will not suffice. Some fear U.S. AI progress could slow, or that embedding AI into critical infrastructures or applications, which China excels in, will ultimately be as or more important for national competitiveness. China’s open source models have become as good – or better – than U.S. open source alternatives. Tencent’s Hunyuan model outperformed Meta’s LLaMa 3.1-405B across a range of benchmarks. Alibaba’s Qwen2.5 model did better across various capability evaluations than OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet models. And most impressively, DeepSeek has released a “reasoning model” that legitimately challenges OpenAI’s o1 model capabilities across a range of benchmarks.

Given the United States’ comparative advantages in compute access and cutting-edge models, the incoming administration could find the time to be right to cash in and put AI export globally at the heart of Trump’s tech policy. While the Biden administration sought to strategically protect U.S. AI advantages, Trump may seek to promote the country’s AI technology.

So far, the Biden administration has put off the challenging decision of whether to send advanced semiconductors to countries stuck in the middle of U.S.-China competition, such as Saudi Arabia and the UAE. But the Trump administration will ultimately need to set a course for its international compute policy. It will need to decide whether to control U.S. compute resources, diffuse them globally or leverage them to secure diplomatic concessions.

While Trump will certainly try to use the United States’ advantage in frontier model capabilities for concessions, he may ultimately be more supportive of an international market-focused approach that unleashes U.S. overseas exports of graphic processing units (GPUs), computer chips especially critical for training state-of-the-art models. His administration may be more supportive of partnerships to build data centers abroad, such as the deal Microsoft struck with G42, a UAE-backed company critical to the country’s efforts to expand its investments in AI. And it might more actively support deals such as the one Nvidia recently made to partner with Vietnam’s government to open an AI research and development center.

These companies have pursued global expansion independently, but the Trump administration could provide incentives for these firms to build an international presence and entrench U.S. technology in key critical infrastructures, just as Chinese’s leading telecom company Huawei did in global 5G markets over the past decade. AI models are easy to replace; critical infrastructures, in contrast, are not. Moreover, Trump’s team could seek to specifically empower smaller firms and start-ups, which might otherwise struggle to compete on the international market without government backing. And it may begin to explore new ways to empower the open source ecosystem domestically with an eye toward international competitiveness, creating financial incentives to develop open source solutions.

Trump could also leverage the United States’ AI advantages in the development sector, where the country faces continued challenges from China. China’s Global AI Governance Initiative offers a platform for embedding Chinese AI systems globally, such as through implementing smart city technology like networked cameras and sensors. Through its AI Capacity-Building Action Plan for Good and for All, China has explicitly stated its goal of sharing its best practices with the developing world, carrying out AI education and exchange programs, and building data infrastructure to promote fair and inclusive access to global data.

China’s efforts build on a strong tradition of exporting both technology and talent in regions like Latin America, where the United States has failed to compete. Trump’s team will likely want to compete in the development sector, but hesitate to hand over development aid resources in AI to the United Nations, reflecting his wariness of international institutions with large membership and rigid bureaucratic structures. Specifically, the United Nations’s ambition to establish a global fund for AI might struggle to gain substantial U.S. backing.

But Trump’s reluctance to cooperate with the U.N. will not necessarily stop the next administration from pushing to increase access to talent, compute, and data in lower- and middle-income countries. Instead, Trump and his allies could empower development-focused agencies like USAID, which has already begun to leverage AI in its aid plans. It could also empower export promotion agencies, such as the Export-Import Bank, to engage in development-based dealmaking with the rest of the world. Such deals would allow the United States to set global standards through embedding technology in critical infrastructures as opposed to negotiating them in international fora. In a September report, now Secretary of State nominee Marco Rubio explicitly stated the need for the United States to provide compelling technological alternatives in third countries to combat Chinese efforts abroad. His office wrote, “the United States will need to offer a substantive alternative, not just strong condemnations and humanitarian aid.”

Trump has long preferred one-on-one trade deals over working through international institutions.  Trump administration AI development deals could similarly be conducted bilaterally. But his first administration showed willingness to engage in carefully scoped multilateral efforts when they served U.S. interests. Trump and Michael Kratsios, who was recently nominated as Director of the White House’s Office of Science and Technology Policy, brought the United States into the G7’s Global Partnership on AI, framed largely as a multilateral effort to counter China’s AI ambitions. Similar deals could plausibly be made for targeted development projects within the G7 or other carefully scoped multilateral efforts, so long as any deal is ultimately seen to boost U.S. national competitiveness.

Let’s Make a Deal, China AI Edition?

As Trump pursues this global AI strategy, the bilateral relationship with China looms as both the greatest challenge and, potentially, the most intriguing opportunity. He inherits a third round of export controls that, while heavily criticized, follows a core logic that places U.S. AI frontier model supremacy at the core of AI policy. Key nominees, such as Undersecretary of State for Economic Growth Jacob Helberg, a strong supporter of efforts to ban TikTok, signal continued pressure to decouple critical technology supply chains from China.

The broader context of U.S.-China relations presents additional hurdles. Trump’s threat to impose 100 percent tariffs on BRICS countries and ongoing cross-Strait tensions create an environment where substantive AI dialogue seems unlikely. Indeed, the first official U.S.-China AI dialogue, held in May in Geneva, yielded little progress toward consensus on frontier risks. Given the Trump administration’s general hawkishness, it is unlikely that Trump and Chinese President Xi Jinping will prioritize a U.S.-China agreement on frontier AI when models in both countries are becoming increasingly powerful.

Yet history suggests opportunity in unlikely places. Just as Richard Nixon’s hawkish credentials enabled him to open relations with China in 1972, Trump’s position could create space for targeted cooperation. Such a deal is certainly unlikely. And it is a near impossible exercise to predict what types of deals might emerge in a rapidly changing geopolitical environment and an unforeseeable AI technological trajectory.

Trump may find compelling business or strategic reasons to engage China on AI. In adjacent parts of the emerging tech ecosystem, Trump is already toying with the idea of intervening in TikTok’s impending ban in the United States, saying, “I have a warm spot in my heart for TikTok,” and that he “won youth by 34 points, and there are those that say that TikTok had something to do with it.” The seeds for Trump wheeling and dealing with China in the emerging tech sphere have been planted.

Beyond economic motives, security concerns surrounding increasingly powerful frontier AI systems in both the United States and China could create a sufficiently large zone of possible agreement for a deal to be struck. There is already precedent for high-level U.S.-China coordination to tackle shared AI security concerns: last month, Biden and Xi agreed humans should make all decisions regarding the use of nuclear weapons.

There are already signs that the Trump administration will need to take model security systems concerns even more seriously. When led to believe it would be monitored and shut down for scheming to pursue a specific goal, OpenAI’s o1 model attempted to deactivate its oversight mechanism in five percent of cases, and Anthropic’s Claude 3 Opus Model engaged in strategic deception to avoid its preferences from being modified in 12 percent of cases. DeepSeek’s R1 model, meanwhile, has proven easy to jailbreak, with one X user reportedly inducing the model to provide a detailed recipe for methamphetamine. These concerns have long been held by some of the most important figures in Trump’s orbit. Elon Musk, arguably Trump’s most influential advisor on AI, has said he believes there is a 10 to 20 percent chance artificial intelligence “goes bad.”

The convergence of rising AI capabilities and security concerns could create unexpected opportunities for U.S.-China coordination, even as competition between the great powers intensifies globally. Trump’s combination of dealmaking instincts and hawkish credibility positions him uniquely to pursue both aggressive global expansion of U.S. AI technology and targeted cooperation where interests align.

For now, the specific contours of any potential AI agreement remain speculative. But Trump’s track record suggests that deals once thought impossible can emerge when security imperatives and business opportunities align. In the high-stakes domain of frontier AI, Trump’s transactional approach to foreign policy could prove conducive to breakthrough agreements – even, or especially, with China.

No Comments Yet

Leave a Reply

Your email address will not be published.

©2024 Global Cyber Security Report. Use Our Intel. All Rights Reserved. Washington, D.C.