Back to Insights
Managing Rivalry: The New Logic of US-China Strategic Stability
The latest Trump-Xi summit emphasized managing US-China rivalry rather than resolving it. Despite competition in trade, technology, and security, both nations seek to prevent escalating tensions into crises. However, AI’s autonomous nature and potential for ambiguous incidents complicate this approach, challenging traditional crisis management mechanisms based on clear attribution and intent.
The latest meeting between Donald Trump and Xi Jinping produced little that would normally qualify as a diplomatic breakthrough. There was no comprehensive political agreement, nor any attempt to redefine the bilateral relationship. Progress was also limited on some of the issues likely to shape future competition, not least artificial intelligence. Yet the summit’s lack of ambition may be its most revealing feature. The most immediate outcome is that, following their meetings in May and September, the leaders of the world’s two leading powers are expected to meet again before the end of the year — possibly twice, at the APEC summit and the G20.
Washington and Beijing appear increasingly interested in managing their rivalry rather than resolving it. This is a narrower objective, but not necessarily a less consequential one. Competition remains entrenched across trade, technology and security, while the structural sources of tension have not disappeared. What seems to be emerging instead is a shared interest in preventing friction in any one of these areas from automatically triggering a broader crisis.
The distinction matters. Strategic stability between the United States and China does not require political convergence, nor does it imply a return to the era of engagement. It requires both sides to recognise the costs of uncontrolled escalation and, as a result, to preserve mechanisms through which competition can be kept within manageable bounds.
The White House summit offered some indications of this approach. The most contentious issues did not dominate the meeting. Trade negotiations continued despite the deterioration in political trust, and Trump and Xi announced tariff reductions worth around $60 billion. The relatively routine coverage of the meeting in Chinese media also suggested an encounter intended more to maintain the relationship than to present the beginning of a new phase.
This is not détente. It is a more limited form of stability, built not on trust but on the need to make competition sufficiently predictable. The economic relationship illustrates both why such management is necessary and why it remains difficult.
In recent years, the debate over US-China economic security has focused on decoupling, de-risking and the reduction of strategic dependencies. In practice, however, interdependence has proved more resilient than the political rhetoric surrounding it. More importantly, it has not simply survived strategic competition. It has become part of it.
Tariffs, export controls, access to technology, critical minerals and supply-chain vulnerabilities are no longer simply instruments of economic policy. They are sources of leverage. Both governments are seeking to reduce dependencies that the other side can exploit while preserving those that provide them with coercive or bargaining power.
China’s restrictions on critical-mineral exports have demonstrated this dynamic particularly clearly. They exposed the difficulty of rapidly reorganising sophisticated industrial supply chains and showed how Beijing’s position in key sectors can translate into strategic leverage. But that leverage is not one-directional. Washington retains significant instruments of its own through technology controls, access to the US market, American financial power and the United States’ position within advanced technology ecosystems.
The relationship is therefore characterised less by progressive separation than by a form of competitive interdependence. The United States and China are simultaneously seeking to limit their vulnerabilities and preserve instruments capable of imposing costs on the other. Economic security is becoming part of strategic bargaining rather than simply a policy aimed at reducing economic ties.
This also helps explain why trade negotiations and systemic competition can proceed simultaneously. Negotiation does not necessarily signal an easing of geoeconomic rivalry. It is one of the mechanisms through which that rivalry is managed.
There is, however, a condition for this mechanism to contribute to stability: each side must be able to read the other’s actions. An export restriction can be calibrated. A tariff can be negotiated. A technology control can be tightened or relaxed. More importantly, it is generally possible to identify who made the decision, which instrument was used and, at least to some extent, what signal was intended. The problem becomes harder when intent, attribution and responsibility are less clear. This is where artificial intelligence takes on significance beyond technological competition itself.
The lack of meaningful progress on AI safety at the summit deserves attention for precisely this reason. The United States and China do not simply have different approaches to AI governance. They are developing increasingly capable technologies without the parallel emergence of an international architecture — let alone a bilateral one — capable of managing the incidents those technologies may produce.
Consider a plausible incident. An AI agent developed by an American company penetrates the systems of a Chinese company or government institution. The intrusion could result from malicious use, inadequate safeguards, or behaviour that was neither anticipated nor authorised by the developer. Technically, these are very different circumstances. For the government on the receiving end, however, distinguishing between them may initially be impossible.
At that point, the problem becomes political before it is technological. Beijing would have to determine whether it was dealing with criminal activity, negligence by a private company, unexpected behaviour by an autonomous system, or an operation connected to the US government. Washington, meanwhile, might not immediately have the information required to reconstruct the incident and provide a credible explanation.
This is where AI places pressure on some of the traditional assumptions of crisis management. Managing crises depends on the ability to attribute an action, interpret its signal and understand, at least approximately, the adversary’s chain of decision-making. Systems with greater degrees of autonomy can make all three more uncertain. The actor involved may be private; the behaviour of the system may not fully correspond to the intention of its operator; and the speed of an incident may outpace the diplomatic processes needed to clarify responsibility and intent.
The comparison with geoeconomic competition is instructive. When Beijing restricts exports of a critical mineral, Washington knows who has acted and can interpret the measure within the context of a broader political relationship. When the United States restricts access to particular technologies, China can identify the decision, the responsible authority and the strategic signal. Both sides may regard these measures as hostile, but the structure of the interaction remains legible. An AI-related incident may provide none of these certainties.
This may be the most difficult feature of the model of competition now emerging. Washington and Beijing are developing mechanisms to manage areas in which escalation is predominantly intentional and therefore, at least to some degree, negotiable. They are considerably less prepared for crises generated by ambiguity, technological autonomy or actors that do not fall entirely under state control.
The latest Trump-Xi summit therefore matters less for the agreements it produced than for what it suggests about the relationship the two powers are trying to sustain. Neither appears prepared to abandon strategic competition, just as neither is likely to give up the economic and technological instruments that provide leverage over the other. The more realistic objective is to build sufficient predictability around that competition to prevent a single dispute from destabilising the broader relationship.
This approach may work in areas where pressure, signalling and retaliation remain largely under state control. Artificial intelligence exposes a potential limit. If strategic stability is becoming one of the central functions of US-China dialogue, the question will not only be how Washington and Beijing manage the confrontations they choose to have. It will be how they manage crises that neither intended to create.
