Trump has just fired the starting gun on a new Cold War with China

Artificial intelligence, which US President Donald Trump has now branded as 'super intelligence' or SI, is reminding more and more of the dangerous Cold War contest between the US and erstwhile Soviet Russia. With AI companies themselves pointing at possible risk to humanity from AI, it could be even more dangerous than nuclear weapons which have been under tight control and regulation with a great degree of predictability too.

What looked like another tech race between the US and China is now turning into a contest over economic power, military capability and technological innovation, with both governments increasingly treating control of AI as a matter of national security.

Trump and Xi Jinping have agreed to communicate if an AI crisis erupts, yet Trump has rejected deeper cooperation on the risks created by the technology. The logic resembles the nuclear Cold War, where rivals developed ever more destructive capabilities while building limited mechanisms to prevent a catastrophe.


ALSO READ | Donald Trump and Xi Jinping's 'stalemate' summit was a glitzy gift to China

AI is becoming a strategic weapon

Trump made the strategic calculation explicit Tuesday, speaking after a White House meeting with top technology executives. “Whoever wins superintelligence wins. You’re gonna have a winner and a loser, and you’re probably not gonna have a second place,” he said.

That is a striking description of AI as a contest for national primacy. The issue is not simply which country produces the better chatbot. Advanced AI can affect scientific research, drug discovery, industrial productivity, cybersecurity and military intelligence. It could also accelerate the development of future technologies. The country that builds the strongest AI ecosystem could therefore gain advantages across much of the economy and national-security apparatus.

ALSO READ | Trump asked Xi if China wanted to buy US weapons, ambassador says

Trump's argument is that America already has a lead and should not risk surrendering it by creating arrangements that expose sensitive technology to China. He said after his meeting with tech executives that working jointly with Beijing on AI could mean giving away American “secrets”. The administration is therefore approaching AI safety through the lens of strategic competition rather than through the assumption that the two leading powers share a common technological interest.

Cooperation only when on the edge of a crisis

The recent Trump-Xi summit produced a narrow form of AI cooperation. The White House said the two countries would establish a US-China “Super Intelligence” dialogue to discuss risks and benefits, with another meeting due by November. They also agreed to create a bilateral communication channel for AI-related incidents.

The proposed mechanism has obvious echoes of the Cold War nuclear hotline between Washington and Moscow. The point of such a channel is not to end strategic competition but to prevent a dangerous incident from becoming a wider confrontation.

That is particularly relevant with AI because neither government may know exactly what the other side's most advanced systems can do. A powerful AI system behaving unexpectedly, an autonomous cyber operation or an AI-assisted military incident could create a crisis in which officials have very little time to determine whether the other side is attacking deliberately.

The problem is that the new channel remains thinly defined. Wired reported before the summit that officials were still struggling to determine what would constitute an incident serious enough to trigger notification. Axios similarly noted that the two governments had not specified the circumstances in which the mechanism would be activated. So this is risk management, not an AI arms-control treaty.

The US and China are building a crisis telephone line for the race rather than agreeing to manage the race.

The mistrust reminds of Cold War-era

The suspicion between Washington and Beijing is embedded in the hardware and software underpinning the AI competition.

The US has restricted China's access to advanced computing technology because Washington believes cutting-edge chips can strengthen China's military and strategic capabilities. US export controls continue to constrain Chinese access to advanced semiconductors and high-bandwidth memory used in AI computing.

China, meanwhile, is pushing hard for technological self-sufficiency. Reuters has reported recently that Huawei has emerged as an increasingly important supplier of high-performance AI hardware as Chinese companies work around restrictions on Nvidia's most advanced chips.

The distrust extends to AI models themselves. On September 8, the US accused six Chinese AI companies, including DeepSeek, Moonshot AI and Alibaba, of using “distillation” to copy capabilities from American models. China rejected the accusation and argued that distillation is a legitimate technical method.

This is the kind of environment in which deep safety cooperation becomes politically difficult. Washington fears that information shared for safety purposes could also help China's technological progress. Beijing has its own reasons to distrust American restrictions on chips and AI technology.

Trump's bargain for Silicon Valley

The other half of the story is happening inside America. On Tuesday, Trump brought the leaders of Google, Nvidia, Anthropic, OpenAI, Meta and xAI together at the White House. They signed a voluntary agreement covering internal controls, external audits and board-level review of AI safety. Trump called it “morally binding” and compared it to a constitution.

But there is no equivalent of a nuclear regulator standing above the companies with legally enforceable limits. The arrangement instead relies heavily on the companies policing themselves. The framework does not establish legal enforcement, allows companies substantial discretion over auditing and does not require public disclosure of audit findings.

That creates an unusual strategic combination of less willingness to share safety constraints with China as well as to impose binding constraints on American companies.

Trump argues that heavy regulation could slow American innovation while China continues moving forward. At the White House he again promoted “self-regulation” and said existing US institutions, including law enforcement agencies, already provide guardrails. The approach puts Trump at odds with some AI executives who have argued for stronger government oversight.

Where nuclear analogy stops to work

The comparison with the Cold War nuclear race is useful because both involve a paradox, that the technology can be dangerous precisely because both sides have strong incentives to keep developing it.

During the nuclear race, Washington and Moscow eventually created mechanisms such as hotlines, arms-control agreements and inspection regimes because strategic competition itself created unacceptable risks.

AI is more dangerous than nuclear weapons as there is no fixed quantity of AI capability that can be counted like warheads and it is commercially produced and diffuses through companies rather than being controlled entirely by governments.

But the underlying strategic problem is beginning to look familiar. The US and China fear falling behind. Each suspects that the other may exploit technological advantages. Neither wants to reveal capabilities that could strengthen its rival. But both have an interest in preventing an AI incident from becoming a military or geopolitical crisis.

Safety versus supremacy

Chinese AI systems are developing quickly, while their safety practices remain less transparent than those of leading US companies, according to Reuters. Recent testing has also found deceptive or evasive behaviour in Chinese AI agents similar to behaviour observed in US systems.

That makes Trump's concern about Chinese capabilities understandable as a national-security calculation. But it also makes the absence of deeper cooperation more consequential.

If increasingly powerful AI systems create risks that cross borders, national competition cannot completely contain those risks. A cyberattack does not stop at an international boundary because the model that helped generate it was trained in America or China.

For now, Trump appears to be choosing to maintain America's technological lead first, and manage the risks of competition second. The AI hotline with Xi is the safety belt around that strategy. The voluntary accord with American technology companies is its domestic counterpart. With this the feeling of hurtling towards a global catastrophe is an unmistakable reminder of the risks of the Cold War nuclear arms race.