Published: June 04, 2026, 04:30 UTC
# Trump’s AI Order Tries to Have It Both Ways on China
Trump signed the toughest AI export restrictions on China yet, but also let Nvidia keep selling. The order is a compromise that pleases no one completely.
The scene inside the White House in the days before the signing was not one of unity. Elon Musk and David Sacks — two of the most influential voices in Trump’s tech orbit — were pressing the president to hold off. Sacks, who served as Trump’s AI czar before the role was quietly dissolved earlier this year, argued that the new restrictions would accelerate America’s own decline as much as China’s. Musk, whose companies depend on a global supply chain that includes Chinese manufacturing and customers, warned that cutting off Nvidia sales to Beijing would hand the market to Chinese competitors and fragment the global AI ecosystem.
Trump listened. Then he signed the executive order anyway.
The document that emerged on June 3, 2026, is a study in contradiction. Its stated goal is to protect America’s lead in artificial intelligence while preventing the People’s Republic from using American chips to build weapons systems, surveillance infrastructure, or military-grade AI. But the order is also full of carve-outs, exceptions, and compromises that leave its core question unresolved: can the United States wage a technology war against China without also wounding itself?
The answer the order gives is a muddled “yes, maybe, but not really.” And that may be the most honest thing about it.
What the Order Does
The executive order has three main components, and taken together they represent the most aggressive export-control measures the United States has imposed on China in the AI domain — more restrictive even than the Biden-era rules they partially replace.
First, the Commerce Department’s Bureau of Industry and Security has been directed to expand the list of advanced AI chips subject to licensing requirements. The new categories cover not just the cutting-edge processors used to train large language models, but also the mid-range chips that power inference workloads in data centers. This closes a loophole that Chinese firms had been exploiting: buying slightly less powerful chips that fell below the export threshold of earlier rules, then linking them in clusters to achieve the same computing power.
Second, the order targets the cloud infrastructure that supports Chinese AI operations overseas. Specifically, it cracks down on China-linked data centers in Southeast Asia — facilities in Malaysia, Singapore, and Indonesia that Beijing has been using as backdoor access points for American chips and computing power. The Commerce Department can now revoke export licenses for any data center operator found to be providing Chinese firms with access to restricted American technology, even indirectly. This is aimed squarely at the gray-market infrastructure that has grown up around the region in the past two years as Chinese AI companies sought to evade earlier restrictions.
Third, the order puts America’s most advanced AI capabilities to work defending critical domestic infrastructure. The Department of Energy, the Cybersecurity and Infrastructure Security Agency, and the Department of Defense have been ordered to deploy federal AI systems to monitor and protect power grids, hospital networks, government databases, and financial systems from Chinese state-sponsored cyber attacks. This is not a defensive measure in the traditional sense — it goes beyond patching vulnerabilities and into active threat-hunting using machine learning models that can detect Chinese intrusion patterns in real time. The order describes this as “asymmetric defense,” but the plain meaning is that the United States is now using AI to fight a continuous, automated cyber war against China on its own soil.
What the Order Does Not Do
The order’s teeth are real, but so are its missing ones. And the most conspicuous gap concerns Nvidia.
Trump permitted Nvidia to continue selling certain advanced chips — the H200 and H20 series — to Chinese customers, albeit with tighter caps on total volume. This is a retreat from the Biden administration’s more restrictive stance, which had sought to cut off almost all advanced chip exports to China. The rationale was business pragmatism: Nvidia’s revenue from Chinese customers is substantial, and a total ban would have cost the company billions and accelerated Chinese efforts to develop indigenous alternatives. But the effect is to undercut the order’s own logic. If the chips are dangerous enough to license, they are dangerous enough to ban. By allowing Nvidia to keep selling, Trump sent the message that American commercial interests still trump — no pun intended — American national security concerns.
The order also lifts some of the Biden-era restrictions on cloud computing services. Under the previous framework, American cloud providers like Amazon Web Services, Microsoft Azure, and Google Cloud were required to police their customers for Chinese government links and cut off access to advanced computing capabilities. Trump’s order relaxes these requirements, arguing that they were too broad and hurt American technology companies without meaningfully hurting China. The new rule shifts the burden from the providers to the Commerce Department, which now must specifically identify and sanction individual bad actors rather than imposing blanket prohibitions.
For critics on the hawkish side — and they are many — this is a catastrophic concession. Senator Marco Rubio called the Nvidia carve-out “a green light for the Chinese military-industrial complex.” The Heritage Foundation issued a statement accusing Trump of “selling the future to Beijing for quarterly earnings.” But the pro-business wing of the administration counters that a total ban would be worse: China would simply buy from alternative sources, and the only thing the United States would accomplish is the destruction of its own semiconductor industry.
This is the central tension of the order, and Trump has not resolved it. He has merely papered it over with a document that gives both sides something — and neither side everything.
China’s Response
Beijing did not wait for the ink to dry. Within hours of the order’s release, the Ministry of Commerce issued a statement condemning the new restrictions as “a violation of international trade norms” and “an act of technological hegemonism.” The language was boilerplate, but the actions behind it were not.
China has been preparing for exactly this scenario for at least two years. The Biden-era export controls, which Trump inherited and now partially expanded, forced Beijing to accelerate its domestic chip development programs. State-owned semiconductor firms have poured billions into developing alternatives to Nvidia’s lineup, and while no one in the industry believes Chinese chips have caught up, the gap is narrower than it was in 2023. Huawei’s Ascend series now competes credibly in the mid-range AI chip market, and SMIC has made progress on advanced fabrication nodes despite being cut off from Dutch lithography equipment.
More importantly, China has been building circumvention infrastructure. The data center crackdown in Southeast Asia will hurt, but Beijing has already begun shifting its cloud operations to partners in Central Asia and the Middle East. Chinese AI companies have been stockpiling American chips — buying through shell companies, using third-country intermediaries, and exploiting the very loopholes the new order is meant to close. The Chinese government has also been pressing domestic tech firms to prioritize homegrown chips over Nvidia alternatives, even when the domestic chips are less performant. The message from Beijing is clear: dependence on American hardware is a strategic vulnerability, and it must be eliminated — whatever the short-term cost.
The Ministry of State Security has also signaled that it will retaliate against American companies operating in China that comply too aggressively with the new rules. This is the standard Chinese playbook: make compliance expensive and painful, so that American firms have an incentive to find workarounds. Nvidia, which maintains a significant presence in China, will be under enormous pressure to keep selling through whatever channels remain open.
The Broader Tech War Context
The executive order is not an isolated event. It is the latest chapter in a story that began long before Trump’s return to office and will continue long after he leaves it.
The US-China technology war has its own rhythm. The Biden administration established the framework: export controls, investment screening, entity lists, and multilateral coordination with allies. Trump inherited that framework and has been expanding it — but reluctantly, and with the constant pull of his own pro-business instincts. The result is a policy that lurches between confrontation and accommodation, never quite settling on either.
In this, the AI order mirrors the broader US-China relationship. The two countries are not at war, but they are not at peace. They are not decoupling, but they are not deepening engagement. They are locked in a managed conflict — a state of permanent competition that neither side can win and neither side can afford to walk away from. Every escalation brings a counter-escalation. Every restriction creates a new circumvention. Every attempt to slow the other side accelerates the other side’s independent development.
The order will slow China’s access to American AI chips, but it will also accelerate China’s drive to build its own supply chain. It will protect some American industries, but it will hurt others. It will make the United States marginally safer from Chinese cyber attacks, but it will not stop those attacks, because China is simply investing more in its own offensive AI capabilities.
The language of the executive order is grand: “preserving the United States’ strategic advantage,” “protecting the American people from foreign adversaries,” “ensuring that artificial intelligence serves freedom, not authoritarianism.” But the reality is more mundane. The order is a compromise between competing factions in the administration, a balancing act between national security and corporate profits, a document that tries to have it both ways because having it one way is politically impossible.
Neither War nor Peace
The AI executive order is not a turning point. It is not a decisive blow or a strategic masterstroke. It is a snapshot of a moment in a long, grinding conflict — a conflict that has no end in sight and no clear path to resolution.
Trump’s pro-business allies wanted him to do nothing, to let the market sort out the China problem. His hawkish advisors wanted him to do everything, to cut off every chip, every license, every sale. He did something in between, and that is probably what an American president will always do on China — enough to look tough, not enough to cause real pain at home, just enough to keep the conflict simmering without letting it boil over.
Whether that is strategy or indecision is a question the order itself cannot answer. But it is the pattern. It has been the pattern since the trade war of 2018, since the Huawei ban, since the CHIPS Act, since every previous escalation that was supposed to be the last one. The US-China technology war is not going to be won or lost by any single executive order. It is going to be managed, incompetently and interminably, by a series of them — each one slightly tougher, each one slightly more compromised, each one a little more exhausting than the last.
The order of June 3, 2026, is the latest of these. It will not be the last.