The Biden administration’s AI chip export framework, first proposed in January 2025, is getting a second life under Trump — and it’s stricter than many in the semiconductor industry expected. Draft rules now circulating in Washington would require U.S. companies to obtain permits before exporting advanced AI chips to most countries, according to The Information. The permit system would replace the tiered country framework Biden originally envisioned, but the underlying intent is the same: keep the most powerful chips out of Chinese hands while maintaining some order in global AI supply chains.
This matters because the original Biden-era rule, known as the AI Diffusion Rule, was set to take effect on May 15, 2025, before the Trump administration delayed it by 120 days. That delay expires in September. What comes next will determine how Nvidia, AMD, and their customers do business across Asia, the Middle East, and beyond.
The core mechanism is straightforward. Companies wanting to ship advanced AI accelerators — think Nvidia’s H100, H200, or successor chips — would need government approval on a per-transaction or per-customer basis. No blanket authorizations for friendly nations. No automatic green lights. Every significant sale scrutinized.
For Nvidia, the financial stakes are enormous. The company generated $35.1 billion in data center revenue in Q4 of fiscal 2025 alone, with a significant portion coming from international customers. CEO Jensen Huang has repeatedly warned that overly restrictive export controls push foreign buyers toward Chinese alternatives. In March 2025, Huang told analysts that “different different different” approaches to export restrictions were creating confusion and chilling demand, as reported by Reuters.
He’s not wrong about the confusion. But the permit requirement isn’t coming from nowhere.
U.S. officials have watched chips diverted through shell companies in Southeast Asia and the Middle East end up in Chinese data centers. A New York Times investigation in early 2025 documented how brokers in Malaysia and the UAE were reselling American-made chips to Chinese entities, sometimes within weeks of purchase. The Commerce Department’s Bureau of Industry and Security has struggled to enforce end-use restrictions without a permit system. So now they want one.
What the permit system actually means for the industry
The practical impact depends entirely on how permits are administered. If the Commerce Department processes applications in weeks, the disruption is manageable. If it takes months — which is the norm for many export license applications — the system becomes a de facto ban for time-sensitive AI infrastructure deals.
Consider the timeline pressures. Hyperscalers like Microsoft, Google, and Amazon are racing to build data centers across the globe. Microsoft alone committed $80 billion in capital expenditure for AI infrastructure in fiscal 2025, with major builds planned in Japan, Indonesia, and the UAE. These projects operate on 12- to 18-month construction schedules. A three-month permit delay doesn’t just slow things down. It kills deals.
Sovereign AI programs are also at risk. Countries including Saudi Arabia, India, and France have announced national AI computing initiatives that depend on access to American chips. The UAE’s G42, which struck a deal with Microsoft in 2024 worth up to $1.5 billion for AI infrastructure, would face new uncertainty under a permit regime. As Financial Times reported, Gulf states have been lobbying Washington aggressively to secure carve-outs.
The semiconductor industry’s main trade group, the Semiconductor Industry Association, has pushed back hard. In a May 2025 statement, the SIA called broad permit requirements “unworkable” and warned they would “cede market share to foreign competitors without meaningfully improving national security.” That’s the standard industry line, but it carries weight when you look at the numbers. Chinese chip designers like Huawei’s HiSilicon have made real progress. Huawei’s Ascend 910C, while not matching Nvidia’s top chips in raw performance, is now being deployed in Chinese data centers at scale, according to Bloomberg.
And here’s the tension the Trump administration hasn’t resolved: stricter controls accelerate Chinese self-sufficiency. Every quarter that Nvidia can’t sell freely into global markets is a quarter where Huawei gains ground. The question isn’t whether China will build competitive AI chips. It’s when. Export controls buy time. They don’t buy permanent advantage.
Some in Washington argue that’s precisely the point — buying time is enough. Jake Sullivan, Biden’s national security advisor who architected the original chip controls in October 2022, framed it as maintaining a “relative” advantage, not an absolute one. The Trump team appears to agree on the goal while disagreeing on the method.
The permit approach also creates a new bureaucratic chokepoint that could be weaponized for trade negotiations. Want chips? Sign a bilateral agreement on AI safety. Or on trade terms. Or on something else entirely. This isn’t speculation. The Trump administration already used chip access as bargaining leverage in negotiations with the UAE and Saudi Arabia over defense and energy deals, as The Wall Street Journal reported in April 2025.
For companies planning their supply chains, the uncertainty is the damage. Not the final rule — the waiting. Nvidia’s stock dropped 4% on the day The Information published its report, though it recovered within 48 hours. That volatility reflects a market that’s priced in some restrictions but can’t model the specifics.
What should industry leaders actually prepare for? Three things. First, build compliance teams now. Permit applications require detailed end-use documentation, and companies that can’t produce it fast will lose to those that can. Second, diversify chip sources where possible. AMD’s MI300X and Intel’s Gaudi 3 may face the same restrictions, but having multiple supplier relationships gives negotiating flexibility. Third, engage directly with the Commerce Department’s rulemaking process. The comment period matters. Companies that submit detailed, data-backed objections have historically shaped final rules more than those that simply complain.
The draft rules aren’t final. But the direction is clear. Washington wants more control over where American AI chips end up, and it’s willing to impose administrative friction to get it. The companies that adapt fastest won’t just survive the new regime. They’ll shape it.
WebProNews is an iEntry Publication