Nvidia and AMD to Share 15% of China AI Chip Sales with US Government — A New Export License Rule
In a groundbreaking policy shift that could reshape the global semiconductor and artificial intelligence (AI) industry, Nvidia and AMD will soon be required to share 15% of their revenue from certain AI chips sold in China directly with the United States government. This rule applies specifically to chips that need special export licenses for sale in China, including Nvidia’s H20 AI chip and AMD’s MI308 chip.
Crucially, this 15% revenue share is not a tax or penalty. Instead, it’s a mandatory licensing condition set by the US government as part of its strategy to control the export of advanced AI hardware to countries where national security concerns are a factor. The regulation is aimed at ensuring that cutting-edge semiconductor technology does not end up strengthening adversarial military or surveillance capabilities, while still allowing American companies to engage in highly profitable sales under controlled conditions.
---
Why This Policy Matters
For years, US regulators have been tightening export controls on advanced semiconductors and AI hardware to China, citing concerns about dual-use technology — hardware that can be applied to both civilian and military applications. GPUs (graphics processing units) and specialized AI chips have enormous computational power, making them valuable for training large AI models, running advanced simulations, and even aiding in military-grade systems.
The US already restricts the export of certain high-end Nvidia and AMD chips to China. But instead of imposing an outright ban, Washington is now offering companies an option: pay a 15% revenue share to the US Treasury on these sensitive sales and obtain a special export license.
This approach provides three strategic benefits for the US:
1. Economic Gain: The US government secures a direct revenue stream from high-value exports.
2. Control & Monitoring: By licensing, Washington gains better insight into who is buying these chips and for what purpose.
3. Selective Access: It allows limited exports without completely cutting off US companies from the lucrative Chinese market.
---
Nvidia’s H20 AI Chip — The Focus of the Rule
The Nvidia H20 is one of the most advanced AI chips ever produced for data center and machine learning applications. Although it is a China-specific version of Nvidia’s flagship AI GPU — designed with reduced performance to comply with earlier US export rules — it still offers high computing power for AI model training, cloud computing, and advanced analytics.
The H20 is part of Nvidia’s H-series lineup, designed to handle massive parallel workloads for AI inference and training. It is widely used in:
AI-powered natural language processing (NLP)
Autonomous vehicle research
Large-scale scientific simulations
Financial modeling
Even though the H20 has been "performance-limited" for compliance reasons, the US still considers it sensitive enough to require an export license.
---
AMD’s MI308 — A Rival in AI Computing
AMD’s MI308 is another AI powerhouse — a high-performance accelerator card aimed at both training and inference workloads. It is part of AMD’s Instinct series, which is designed to compete directly with Nvidia’s H-series and A-series chips in high-end AI markets.
The MI308 offers:
Extremely high memory bandwidth
Optimized architecture for AI workloads
Compatibility with major AI frameworks like TensorFlow and PyTorch
Scalability for data centers and cloud providers
Like the H20, the MI308 also falls into the category of controlled AI chips, meaning AMD must seek special permission before exporting it to China.
---
How the 15% Revenue Share Works
The 15% payment is not a traditional tax collected at the border. Instead, it is part of the export license terms. The mechanics are as follows:
1. Application for License: Nvidia or AMD applies for an export license for a controlled AI chip sale to China.
2. License Approval: If approved, the license includes a clause requiring the company to pay 15% of the gross sales revenue from that chip to the US government.
3. Direct Payment: The company must transfer this amount directly to the US Treasury.
4. Ongoing Compliance: The license is reviewed periodically, and sales data must be reported for verification.
This policy does not apply to other Nvidia or AMD products — such as gaming GPUs, older AI chips, or non-controlled computing hardware. It is narrowly focused on AI chips that meet the export restriction criteria.
---
Impact on Nvidia and AMD
The policy creates a mixed scenario for Nvidia and AMD.
Advantages:
They can continue selling in China’s huge AI market without a complete ban.
Revenue potential remains high despite the 15% deduction, especially given the high profit margins on AI chips.
It maintains their competitive presence against non-US rivals like Huawei or Biren Technology.
Challenges:
The 15% cut will directly reduce profit margins on affected sales.
More reporting and compliance overhead will increase operational costs.
Chinese customers may seek domestic alternatives to avoid the regulatory complexities.
---
The Bigger Geopolitical Picture
This rule is part of the US-China tech rivalry that has intensified in recent years. Washington sees AI chips as strategic technology — comparable to advanced fighter jets or nuclear components — because of their potential to power both commercial AI systems and next-generation military capabilities.
China, meanwhile, has been rapidly developing domestic AI chipmakers like Biren, Cambricon, and Huawei’s Ascend line to reduce dependency on US suppliers. This new rule could accelerate China’s push for semiconductor self-sufficiency, but it also shows the US is trying to balance security concerns with economic interests.
---
Reactions from the Industry
Nvidia’s stance: The company has not officially opposed the policy but is expected to seek ways to pass on some of the cost to buyers, potentially raising prices.
AMD’s position: Likely to comply while maintaining competitive pricing to retain Chinese customers.
Tech analysts: Many believe this is a “middle-ground” solution that allows US companies to profit while maintaining national security safeguards.
Chinese buyers: Some may shift to domestic chips to avoid the political and economic complications.
---
What This Means for the Future of AI Exports
The policy could become a template for future US tech export rules — allowing controlled sales of sensitive technology while monetizing the process for the US government. If successful, similar revenue-sharing license models could be applied to:
Quantum computing hardware
Advanced semiconductor manufacturing tools
Military-grade communication equipment
Space technology components
However, there is also the risk that such measures push foreign customers away from US suppliers in favor of neutral or domestic alternatives, potentially hurting American market share in the long run.
---
Key Takeaways
1. Nvidia and AMD will pay 15% of their China AI chip sales revenue to the US government for H20 and MI308 chips requiring export licenses.
2. This is not a tax or penalty but a license condition.
3. The rule applies only to controlled AI chips, not to gaming GPUs or other hardware.
4. The move balances national security concerns with the need to keep US companies competitive.
5. The policy could set a precedent for future export control frameworks.
Comments
Post a Comment