The most consequential question in the global artificial intelligence race is no longer who can build the most powerful systems. It is whether the governments and companies building them can create meaningful limits before commercial and strategic competition makes restraint politically impossible.
That question has become particularly urgent in Washington.
The Trump administration’s latest approach to artificial intelligence places technological leadership at the centre of national strategy. The White House has sought to accelerate AI development, expand computing infrastructure and strengthen the United States’ position in global technology competition, particularly against China. At the same time, the administration has embraced a voluntary framework with leading technology companies aimed at addressing some of the risks associated with increasingly capable AI systems.
There is an obvious contradiction here.
The same policy environment that encourages companies to move faster is asking those companies to police the risks created by moving faster.
That is where the debate over AI governance becomes more complicated than the familiar argument between regulation and innovation.
No serious government wants to surrender technological leadership. AI is already connected to economic competitiveness, military capabilities, cybersecurity, intelligence and critical infrastructure. Washington therefore has strong incentives to ensure that American companies remain at the technological frontier.
But the assumption that technological leadership requires minimal constraints deserves greater scrutiny.
A voluntary safety agreement can establish useful standards. It can encourage companies to develop internal safeguards, independent oversight and stronger security practices. Such measures are preferable to having no safeguards at all.
The harder question is enforcement.
What happens when a company determines that slowing a model would threaten its competitive position? Who decides whether a dangerous capability has crossed an acceptable threshold? Who verifies the company's own assessment? And what happens when commercial incentives collide with public safety?
These are not theoretical questions.
Frontier AI companies have powerful incentives to release increasingly capable systems quickly. Investors expect growth. Competitors are advancing. Governments want technological advantages. Researchers are rewarded for breakthroughs.
The resulting system can create a structural problem: the organisations best positioned to recognise emerging risks are often the same organisations under the greatest pressure to keep accelerating.
That does not mean technology companies are incapable of acting responsibly. It means responsibility cannot depend entirely on voluntary restraint.
The deeper problem is the idea that one country can ultimately dominate the entire AI ecosystem.
Artificial intelligence is not a conventional weapons platform that can be controlled from a single industrial base. Its development depends on semiconductors, cloud computing, energy, data, algorithms, research institutions, open-source communities, capital and highly skilled workers spread across multiple countries.
The United States can dominate important parts of that ecosystem. China can build strength in others. Taiwan remains central to advanced semiconductor manufacturing. Gulf states are investing heavily in computing infrastructure and capital. India and other countries provide research talent, engineering capacity and enormous markets.
The ecosystem is global even when national strategies are not.
That makes international coordination unavoidable.
Yet the geopolitical framing of AI increasingly pushes in the opposite direction. If Washington sees AI primarily through the lens of strategic competition with Beijing, cooperation can begin to look like a concession. If Beijing sees American technological restrictions as an attempt to contain China's development, it has little incentive to accept rules written primarily in Washington.
The result is a familiar great-power dilemma.
Both sides have reasons to accelerate.
Both sides have reasons to distrust the other.
And both sides have reasons to fear falling behind.
That combination is precisely what makes technological races dangerous.
The history of strategic competition offers an important lesson. The most successful mechanisms for reducing catastrophic risk have rarely depended on complete trust between rivals. They have depended on establishing limited rules, verification mechanisms and channels of communication even when broader disagreements remained unresolved.
AI needs the same pragmatism.
Washington and Beijing do not need to agree on the future of technology, their economic systems or their geopolitical ambitions. But they have a shared interest in preventing an AI-related incident from escalating into a wider crisis.
That could include basic agreements on the testing of highly capable systems, incident reporting, cybersecurity, military applications and emergency communication.
The objective should not be to stop competition.
It should be to prevent competition from eliminating every incentive for restraint.
There is also a broader international question. If AI governance becomes exclusively a US-China contest, much of the world will have little influence over rules that will nevertheless affect its economies and societies.
Europe, India, Japan, South Korea, the Gulf states and other major technology powers have interests that do not fit neatly into either Washington’s or Beijing’s strategic framework.
A global technology ecosystem requires a broader conversation about global technology governance.
The argument for cooperation is therefore not an argument against American technological leadership. Nor is it an argument for allowing competitors unrestricted access to sensitive technologies.
It is an argument against confusing dominance with control.
No single government is likely to control every layer of artificial intelligence. The more realistic objective is influence over critical nodes, combined with safeguards that prevent failures in one part of the ecosystem from becoming crises everywhere else.
The Trump administration has an opportunity to pursue both technological leadership and meaningful AI governance. The two objectives do not have to be mutually exclusive.
But voluntary commitments will remain vulnerable if they are treated as substitutes for credible oversight.
The real test of America's AI strategy will therefore not be how quickly its companies can build more powerful systems.
It will be whether Washington can maintain technological ambition without allowing the race itself to become the governing principle.
AI competition is inevitable.
An AI race without guardrails does not have to be.
Make the argument more concreteStrengthen the opening with Washington’s policy tension
