The Trump administration says another round of AI chip export rules is coming, and that could shift the global AI race all over again.
Jeffrey Kessler, the Commerce Department official in charge of U.S. export controls, told lawmakers Tuesday that new regulatory action involving AI and advanced semiconductors is on the way. He also said the administration doesn’t plan to scrap the existing framework that limits how many advanced AI chips certain countries can receive, according to Reuters.
The actual rules haven’t been released yet, and that matters. Washington hasn’t announced a finished policy, a start date, or a complete list of the companies and countries that’ll be affected.
What it has done is put the industry on notice: the rules governing the world’s most important AI hardware are still being rewritten.
Why AI chip restrictions matter
The AI race usually gets framed as a fight between models like ChatGPT, Claude, Gemini, and Grok.
But underneath all of those systems is a much more physical battle involving data centers, electricity, networking equipment, and enormous clusters of specialized processors. It’s the same infrastructure crunch behind the White House push to make Big Tech pay its own AI power costs.
Advanced GPUs and AI accelerators are the machinery used to train and run frontier models. Access to enough of them can determine how quickly a company or country can experiment, scale a model, and deploy it to millions of people.
That makes export controls one of Washington’s most powerful tools in the technology competition with China. The U.S. can’t regulate every model developed overseas, but it can try to control access to the American-designed chips and supporting technology needed to build them at the frontier.
The Commerce Department’s Bureau of Industry and Security is responsible for administering and enforcing those controls.
The strategy is pretty straightforward: slow the expansion of rival AI infrastructure while protecting America’s lead in advanced computing.
Actually pulling that off is where things get messy.
Nvidia has begun limited H200 shipments to China
Kessler’s comments came on the same day he confirmed that Nvidia had started shipping a small number of its powerful H200 AI chips to approved customers in China, Reuters reported.
The shipments are still limited, even though U.S. approvals reportedly cover several major Chinese companies. Approved buyers include firms connected to ZTE, Alibaba, Tencent, ByteDance, JD.com, Kingsoft, and others.
The Nvidia H200 is a high-performance data center GPU built for demanding AI and high-performance computing workloads. Allowing even restricted sales to China creates an uncomfortable balancing act for Washington.
On one side, officials don’t want advanced American technology strengthening China’s military, surveillance, or strategic AI capabilities.
On the other, completely shutting American chipmakers out of China could cost U.S. companies billions of dollars while speeding up demand for Chinese alternatives.
This isn’t a simple blanket ban anymore. It’s more like a tightly controlled valve that Washington can open or close depending on who’s buying the chips, where they’re going, and how they’ll be used.
Some sales are allowed. Others require licenses. Buyers get screened. Shipment volumes stay limited. And the rules can change whenever security concerns or diplomatic priorities shift.
The loophole problem
One of the biggest challenges is making sure chips sold somewhere else don’t eventually reach restricted Chinese companies.
Advanced processors can be purchased through subsidiaries, middlemen, cloud providers, and data centers outside mainland China. That’s forced regulators to look beyond the shipping address and ask who ultimately owns, controls, or uses the hardware.
Nvidia has reportedly tightened its screening of customers across Asia, including companies in Singapore, Malaysia, and Japan. It’s reduced its authorized buyer list and expanded compliance checks meant to prevent advanced chips from being rerouted to China.
Those checks reportedly include customer interviews, contract verification, and visits to data centers.
That tells you how deeply export policy is starting to reach into the everyday operations of chipmakers and cloud infrastructure companies.
Selling the hardware isn’t the end of the transaction anymore. Companies increasingly have to prove where it’ll be installed, who’ll control it, and what it’ll be used for.
The rules are still moving
The administration’s approach to AI chip exports has changed repeatedly.
Officials are trying to balance national security, commercial interests, relationships with allies, and the risk that overly aggressive restrictions could push global customers toward Chinese technology. Earlier proposals have been withdrawn, reconsidered, or replaced with narrower licensing policies and new enforcement guidance.
Tuesday’s testimony makes one thing clear: that process isn’t finished.
The next round of rules could affect Nvidia and AMD, cloud providers building overseas data centers, foreign governments negotiating access to American AI infrastructure, and Chinese companies trying to secure enough compute to stay competitive.
It’s also pushing major AI companies to reduce their dependence on a single supplier. OpenAI is already developing its own custom inference chip, while companies like Arm are moving deeper into AI data-center silicon.
Until the Commerce Department publishes the actual text, we don’t know whether the new action will tighten shipment caps, expand licensing requirements, close subsidiary loopholes, require tougher customer checks, or combine several of those ideas.
What we do know is that advanced AI chips have become a strategic resource, and Washington plans to keep using access to them as leverage.
Tony’s Take
The AI race isn’t just about who builds the smartest model anymore.
It’s about who gets the chips required to build one.
Everyone loves arguing about benchmark charts, coding scores, context windows, and which model won the week. Meanwhile, the companies building those models are fighting over something even more important: compute.
If you don’t have the chips, you don’t get to play at the frontier. It’s really that simple.
That’s why a Commerce Department hearing can matter almost as much as a major model launch. A change to export rules can reshape who gets frontier-level compute, where new data centers get built, and how quickly an entire country’s AI ecosystem can grow.
For now, the biggest detail is also the simplest: new rules are coming, but we don’t know exactly what they are yet.
The moment that text lands, the chip war starts all over again.
