TLDR
- The Trump administration is drafting rules requiring US government approval for AI chip exports worldwide, expanding current restrictions globally.
- Shipments of over 1,000 Nvidia GB300 GPUs would face review; deployments over 200,000 units would require host government sign-off.
- Nvidia has halted H200 chip production for China at TSMC, shifting capacity to its upcoming Vera Rubin chips.
- Nvidia CFO Colette Kress confirmed the company has generated zero revenue from China despite having US approval for some H200 shipments.
- CEO Jensen Huang said Nvidia’s $30 billion OpenAI investment may be its last, as he expects the AI startup to go public soon.
Nvidia $NVDA fell around 1.7% Thursday after two separate stories landed in quick succession — and neither one was good news for the stock.
Bloomberg reported that the Trump administration is drafting new export rules that would require US government approval for AI chip sales to virtually every country in the world. That sent NVDA and $AMD, which dropped around 2%, lower in afternoon trading.
The proposed rules would expand existing restrictions — currently covering around 40 countries — into a full global licensing framework. Under the draft, any shipment of up to 1,000 of Nvidia’s GB300 GPUs would go through a review process, with some exemption paths available.
Larger orders get more scrutiny. Deployments exceeding 200,000 GB300 units owned by one company in one country would require the host government itself to participate in the approval process.
The US would only greenlight those large-scale exports to allied countries that make security commitments and invest in American AI infrastructure — though the draft doesn’t spell out a specific investment ratio.
The rules aren’t a ban, but they would hand the US Commerce Department sweeping oversight over who gets to buy the chips that power services like ChatGPT and Gemini.
China Sales Have Already Hit a Wall
Separately, the Financial Times reported that Nvidia has quietly stopped producing its H200 chips for China at Taiwan Semiconductor Manufacturing Co., with that capacity now being reallocated to the next-generation Vera Rubin chips.
The two chip lines use different technologies and process nodes — H200 runs on CoWoS-S packaging and older high-bandwidth memory, while Vera Rubin uses CoWoS-L and the new HBM4 standard — so the production shift doesn’t necessarily affect supply of one or the other directly.
Nvidia’s China business has been stuck in limbo for months. Trump approved H200 sales to China in December, with a condition that the US government take a 25% cut. Before that, Nvidia had been selling the lower-powered H20 chip in China — until the Trump administration banned that too, last April.
But even with US approval in hand, sales haven’t happened. On last week’s earnings call, CFO Colette Kress said Nvidia has “yet to generate any revenue” from China and wasn’t sure if Beijing would allow any imports at all.
Chinese Rivals Are Gaining Ground
Kress flagged another concern: a string of recent IPOs from Chinese chip companies that she said “have the potential to disrupt the structure of the global AI industry over the long term.” Nvidia says it will keep engaging with both the US and Chinese governments.
On the OpenAI front, CEO Jensen Huang said this week that Nvidia’s $30 billion participation in OpenAI’s $110 billion funding round in late February “might be the last time” the chip maker invests in the AI company, as he expects OpenAI to go public soon. Huang added that a previously discussed $100 billion investment with OpenAI is “not in the cards.”





