U.S.-China High-Level AI and Trade Negotiations and the Open-Weight Model Competition: Conditions for Managed Competition and South Korea’s Choices

Executive Summary

The Besant-He Lifeng negotiations in New York package AI, trade, and critical minerals together, suggesting that discussions on AI norms may function as a bargaining chip in tariff and mineral negotiations rather than as a separate track for security cooperation. U.S. technology containment measures against China are shifting from a focus on chip export controls to a new phase of countering the rapid proliferation of Chinese open-weight models. The low-cost, high-speed diffusion of models like Qwen and DeepSeek exposes the structural limitations of the U.S. closed-weight model strategy. The dual structure of a cooperation track led by the White House and a pressure track from regulatory agencies and Congress is likely to persist after the summit. Consequently, even if the tariff truce is extended and issues over rare earth implementation are resolved, substantive progress on closing the technology gap or reaching normative agreements will likely be limited. South Korea should adopt a selective dual-track approach that avoids premature alignment with either side, using the principle of prioritizing stability for national core infrastructure while allowing for cost-efficiency in the private sector as its policy framework. To this end, it is urgent to secure early details on the review programs that will emerge from the U.S.-China AI dialogue and to establish a system for continuously monitoring the adoption of Chinese-origin open-weight models in the domestic public, financial, and telecommunications sectors.

I. Situational Analysis

The Start of U.S.-China High-Level AI and Trade Negotiations and the Open-Weight Model Competition

1. Background and Developments

On Sunday, September 20, 2026, U.S. Treasury Secretary Scott Besant and Chinese Vice Premier He Lifeng met at the JPMorgan Chase headquarters in New York [14]. The choice of JPMorgan's headquarters as the venue was unusual. While JPMorgan was not directly involved in the meeting, it agreed to provide the building [14]. The meeting was largely seen as a working-level coordination session ahead of the Trump-Xi summit scheduled for September 24 in Washington [4][20].

The agenda for these negotiations is structured differently than in the past. AI, trade, and critical minerals were all on the table simultaneously [1][10][15]. The fact that the tariff truce was set to expire on November 10 created significant time pressure [15]. The Brazilian newspaper Valor Econômico identified this deadline as a key factor behind the negotiations [15]. The issue of Iran was also mentioned as a potential source of additional friction [17].

The inclusion of AI on the agenda can be understood in the context of China's strategic shift to open-weight models. Following the "DeepSeek shock," China elevated the distribution of open-weight models to a national strategy [3]. Alibaba's Qwen has been downloaded 3 billion times, and purchasing delegations from 177 countries attended the WAIC 2026 [3]. This is the result of China shifting the competitive front to expanding its developer ecosystem under conditions where access to advanced semiconductors is blocked [3]. U.S. technology containment measures against China had previously focused on chip export controls. Now, however, they face a new variable: the rapid proliferation of Chinese open-weight models.

2. Current Situation

Secretary Besant publicly stated on September 18 that the negotiations would cover "both open and closed models" [1][4][10][12]. This statement was quoted identically by multiple media outlets, including Channel NewsAsia, The Business Times, Daily Sabah, and The Globe and Mail [1][4][10][12]. This can be read as an indication that the U.S. Treasury Department coordinated the framing of the negotiations in advance and leaked it to the press.

In a communication with Al Jazeera, Secretary Besant emphasized that "the United States is still the leader in AI," but added, "We are open to discussions to avoid shared risks and to avoid a decoupling of our two systems" [17]. This statement illustrates the U.S. attempt to strike a balance: keeping the door open for AI safety cooperation while not relinquishing its narrative of technological superiority.

A statement from a senior U.S. official, quoted by The Hankyoreh, reveals the actual temperature of the negotiations more specifically. The official mentioned that proposals under consideration included allowing mutual pre-review of new AI models weeks before their public release and incorporating Chinese open-source models into the U.S. government's existing voluntary review program [19]. However, the official did not state a specific position [19]. At the same time, the official expressed dissatisfaction with the implementation of the rare earths agreement, noting, "We're not going to trade away export controls to get something we've already negotiated for," and pointed out that China's implementation "has not met expectations" [19]. This suggests that the AI dialogue is de facto linked to the negotiations on rare earths and tariffs.

The Peterson Institute for International Economics (PIIE) holds low expectations for the summit. It does not anticipate fundamental changes in either country's policies, diagnosing that "inertia is a powerful force" [13]. This reflects the view of local experts that the establishment of this new dialogue channel may remain largely symbolic.

3. Key Actors and Positions

U.S. Department of the Treasury (Besant): The Treasury Department has taken the lead for the U.S. side in these negotiations. This can be interpreted as an intention to handle AI from a trade and macroeconomic perspective, moving away from the export control framework centered on the Department of Commerce. Besant's remarks appear to acknowledge the reality that the proliferation of open-weight models is affecting the cost structures of U.S. companies [1][17].

Chinese Vice Premier He Lifeng: The Chinese side appears poised to maximize its negotiating leverage by linking AI, trade, and minerals. It is in a position to use the commercial success of its open-weight models as leverage, turning the demand from U.S. companies into a bargaining chip.

U.S. Big Tech: A large number of U.S. Big Tech CEOs are reportedly expected to attend the state dinner for the Trump-Xi summit [18]. This shows a trend of the industry pursuing the practical benefits of using Chinese models, separate from the government-to-government negotiations. The Council on Foreign Relations (CFR) has pointed out that the U.S. government's concern is not with the open-weight approach itself, but with the Chinese origin of the models [2]. This highlights the divergence between the industry's pursuit of cost efficiency and the government's security concerns about origin.

Third-Country Developers and Markets: There is a growing trend of Indian startups rebuilding their products using Qwen, DeepSeek, and Kimi [16]. A bifurcated approach, as seen in the case of Nigeria's Curacel, is becoming a market standard: using Chinese models for high-frequency, repetitive tasks and Western models for high-difficulty reasoning [6]. This demonstrates that in the gap where neither the U.S. nor China has yet established a coherent response system, pragmatic choices by third countries are proliferating [3][6].

4. Key Issues

The first issue is whether the AI dialogue is an independent security cooperation effort or a bargaining chip in a trade deal. The fact that the November 10 tariff truce deadline and dissatisfaction over rare earth implementation are on the same table increases the likelihood that the AI agenda will be used as trade leverage [15][19].

The second issue is the effectiveness of the U.S. stance to include both open-weight and closed-weight models in the discussion. This is also an admission that the existing strategy centered on chip export controls has reached its limits in the face of rapidly proliferating open-weight models.

The third issue is whether the talks will remain at the level of confidence-building information exchange. PIIE's low expectations and the vague responses from government officials suggest that this dialogue channel is more likely to result in a managed dual-track system rather than the formation of substantive norms [13][19].

II. In-Depth Analysis

The Start of U.S.-China High-Level AI and Trade Negotiations and the Open-Weight Model Competition: In-Depth Analysis

1. Analysis of Root Causes

The fundamental reason these negotiations came about is that the U.S. strategy of technology containment against China has reached its structural limits. Since 2022, the U.S. has used advanced semiconductor export controls as its primary tool of containment. It restricted the sale of Nvidia's top-tier chips to China and expanded controls to include equipment and software. However, this strategy failed to prevent the proliferation of the models themselves [2]. Even with restricted access to chips, China opened a workaround by training and releasing models like DeepSeek and Qwen at low cost [3]. CFR expert McGuire pointed out that the U.S. administration is not taking issue with Chinese-made open-weight models per se. The problem, he argues, is the very fact that the models "came from China" [2]. This implies that U.S. concerns stem more from political distrust of the models' origin than from a technical risk assessment.

The second root cause is the asymmetry in cost structures. The reason U.S. companies are adopting Chinese open-weight models over domestic closed-weight ones is not ideology but a cost-benefit calculation [1][4][10][12]. The case of Indian startups rebuilding their products with Qwen, DeepSeek, and Kimi is a prime example. These models have a performance gap of only about six months compared to the U.S. frontier models, but their cost is a fraction of the price, as low as one-tenth or less [16]. This asymmetry has created a natural market selection, leading to a structural conflict between the government's security logic and the corporate world's cost logic.

The third cause is the political pressure of the tariff truce deadline. The fixed deadline of November 10 created an incentive to bring AI, trade, and rare earths to the negotiating table simultaneously [15]. The reason a senior U.S. official continues to negotiate despite expressing dissatisfaction with delays in rare earth implementation is that progress on the AI agenda can be used as leverage in the tariff and mineral negotiations [19].

2. Structural Context

Economic Structure: The U.S.-China AI competition is now shifting its center of gravity from semiconductor hardware to software and ecosystem competition. China has demonstrated the expansion of its developer ecosystem with achievements like 3 billion downloads of Alibaba's Qwen and the participation of purchasing delegations from 177 countries at WAIC 2026 [3]. This pattern is also being replicated in the African market. The Nigerian insurtech company Curacel has adopted a dual-track strategy, using Chinese models for high-frequency, repetitive tasks and Western models for high-difficulty reasoning [6]. If such pragmatic choices spread widely at the individual company level, it could entrench a structure where the U.S. falls behind in the standards competition. The CFR noted that while the U.S. employs a strategy of protecting corporate profits and national security primarily through closed-weight models, China induces easy localization by developers worldwide through free and open distribution [11].

Political Structure: Within the U.S. administration, two different tracks are operating simultaneously. A cooperation track led by the White House and a pressure track led by regulatory agencies like the FCC and Congress are moving separately, preventing any real narrowing of the gap [9]. Secretary Besant's statement that he is "open to discussions to avoid shared risks and to avoid a decoupling of our two systems" [17], contrasted with another senior official's expression of dissatisfaction with the level of rare earth implementation [19], shows that there are differing temperatures even within the same administration. This suggests that even if a U.S.-China AI dialogue is agreed upon, its implementation could be limited by this dual-track structure within the U.S. government.

Security Structure: The focus of security concerns has shifted. In the past, the primary concern was the military diversion of advanced chips. Now, the data and governance exposure arising from the routine adoption of Chinese-made models by U.S. companies and individuals has emerged as a new concern [2][5]. The case of the Nvidia-Hugging Face deal, discussed in a CFR podcast, illustrates that when the proliferation of open-weight models combines with the commercial incentives of infrastructure companies, control becomes even more difficult [5].

3. Historical Precedents and Comparative Cases

The structure of these negotiations follows a recurring pattern in past U.S.-China technology talks. At the Beijing summit in May 2025, the two countries had already agreed to launch an intergovernmental dialogue on AI [9]. However, that agreement remained at the level of "confidence-building information exchange," and the pressure track from regulatory agencies continued to operate separately from the White House's cooperation track [9]. This precedent suggests that the September New York negotiations and the Trump-Xi summit are likely to follow a similar path.

The practice of linking the critical minerals card to trade negotiations is also a recurring pattern. The senior U.S. official's statement that "We're not going to trade away export controls to get something we've already negotiated for" [19] shows that rare earth export controls have become a permanent lever, brought up again in each new round of negotiations. This is similar to the structure seen during the 2019-2020 U.S.-China Phase One trade deal, where implementation verification issues were repeatedly raised.

In terms of the open-weight proliferation competition, the precedent in the African market offers important insights. While Huawei gradually captured the financial infrastructure over a long period and Alibaba and DeepSeek expanded their market base through free distribution, U.S. Big Tech retreated to a discourse of trust and partnership [6]. If this pattern is repeated in third-country markets like India and Southeast Asia, the U.S. could find itself in a paradoxical situation of maintaining its frontier model superiority while lagging in actual adoption rates [16].

4. Key Variables Shaping Future Developments

The first variable is whether the proposal for pre-release model review is concretized. A plan to allow mutual review of new models weeks before their public release and a proposal to include Chinese open-source models in the existing U.S. voluntary review program are on the table [19]. If this plan is actually institutionalized, it would signal that the AI dialogue is becoming an independent security cooperation track. Conversely, if discussions are merely repeated without a concrete framework emerging, the AI agenda will remain a subsidiary card in the tariff and mineral negotiations.

The second variable is the level of implementation on rare earths. With the U.S. side having already assessed it as "not meeting expectations" [19], if no improvement in implementation is confirmed by the November 10 tariff truce deadline, the AI dialogue channel itself could be expended as a tool for negotiation pressure.

The third variable is whether the dual tracks within the U.S. administration converge. If the White House's cooperative stance and the pressure from regulatory agencies and Congress continue to operate separately, any summit agreement will remain largely symbolic [9]. PIIE's low expectations for this summit are also rooted in this inertial structure [13].

The fourth variable is the speed of adoption in third-country markets. If the shift by Indian startups to Chinese models [16] and the adoption of task-specific dual-sourcing by African companies [6] continue to spread, the U.S. will find itself at a structural disadvantage that cannot be countered by regulating its own companies alone. This is also the substantive reason why Washington has brought the open-weight issue to the bilateral negotiating table. With the leadership in the standards competition already being largely decided in the market, it is highly likely that government-to-government negotiations will continue to follow these developments after the fact.