Key Takeaways
- Trump administration develops new framework requiring federal approval for AI chip exports to nearly all global markets, moving beyond current 40-country restriction list.
- Proposed regulations would mandate review for shipments exceeding 1,000 Nvidia GB300 GPUs; orders surpassing 200,000 units need host nation involvement in approval.
- Production of H200 chips for Chinese market has ceased at TSMC, with manufacturing resources redirected toward Vera Rubin chip development.
- CFO Colette Kress disclosed that China operations have produced no revenue for Nvidia, even with partial US authorization for H200 distribution.
- Jensen Huang indicated Nvidia’s $30 billion OpenAI stake could mark the company’s final investment round before anticipated public offering.
Nvidia $NVDA experienced a 1.7% decline Thursday following dual announcements that put pressure on shares throughout the trading session.
Bloomberg disclosed that the Trump administration is developing export regulations requiring federal approval for AI chip transactions across nearly every global market. The announcement drove NVDA and $AMD, which fell approximately 2%, lower during afternoon hours.
The incoming regulations would transform current restrictions — affecting roughly 40 nations — into a comprehensive worldwide licensing system. The proposed framework would subject any shipment containing up to 1,000 of Nvidia’s GB300 GPUs to a formal review process, though certain exemption channels would remain available.
Higher-volume orders face intensified scrutiny. Any deployment involving more than 200,000 GB300 units under single company ownership within one country would necessitate direct participation from that nation’s government in the authorization process.
The US would grant approval for such large-scale exports exclusively to partner nations willing to provide security assurances and commit to American AI infrastructure investments — although the proposed framework lacks specifics regarding investment thresholds.
These regulations establish oversight rather than prohibition, granting the US Commerce Department extensive authority over distribution of chips powering services including ChatGPT and Gemini.
China Sales Have Already Hit a Wall
The Financial Times revealed separately that Nvidia has discontinued H200 chip manufacturing for China at Taiwan Semiconductor Manufacturing Co., redirecting that production capacity toward next-generation Vera Rubin chips.
The two product lines employ distinct technologies and manufacturing processes — H200 operates on CoWoS-S packaging combined with earlier high-bandwidth memory, whereas Vera Rubin utilizes CoWoS-L alongside the latest HBM4 standard — meaning the production reallocation may affect each line’s supply independently.
Nvidia’s Chinese market operations have remained stagnant for several months. Trump authorized H200 distribution to China during December, establishing a requirement for 25% US government revenue share. Prior to that authorization, Nvidia had marketed the reduced-performance H20 chip within China — until Trump administration officials prohibited those sales last April.
Despite receiving US authorization, transactions have failed to materialize. During last week’s earnings presentation, CFO Colette Kress revealed Nvidia has “yet to generate any revenue” from China while remaining uncertain whether Beijing would permit any imports whatsoever.
Chinese Rivals Are Gaining Ground
Kress identified an additional challenge: multiple recent IPOs from Chinese semiconductor firms that she characterized as having “the potential to disrupt the structure of the global AI industry over the long term.” Nvidia maintains its commitment to ongoing dialogue with both US and Chinese government officials.
Regarding OpenAI developments, CEO Jensen Huang stated this week that Nvidia’s $30 billion participation in OpenAI’s $110 billion funding round completed in late February “might be the last time” the chipmaker backs the AI company financially, citing expectations for OpenAI’s imminent public offering. Huang clarified that a previously discussed $100 billion investment arrangement with OpenAI is “not in the cards.”
