The agreement, which has been leaked to the public these days, changes the way US export controls are implemented.
According to Reuters, citing a senior US official, Nvidia and AMD have agreed to pay the US government 15% of revenue from the sale of certain AI chips in China. In return, the US Department of Commerce has begun granting them licences to export these products to the Chinese market.
There has never been a model like this before. Export controls have always worked in a binary way: technology is either allowed or banned, and the decision is based on safety assessments.
Now, in addition to the technical restrictions and a list of authorised products, a fiscal element has also been introduced.
The right to market access is being transformed into an agreement in which the state becomes a partner in every respect.
Technology at the centre
At the heart of this deal are the H20 from Nvidia and the MI308 from AMD – both products are powerful AI accelerators designed for data centres.
The H20 is a version of Nvidia's GPU system that has been technically "trimmed" (modified to stay under the performance thresholds set by US export regulations).
Despite these limitations, the H20 still offers enough performance for sophisticated data processing tasks and training of smaller AI models, making it attractive to Chinese buyers.
The ability to offer these products in the Chinese market means that they are present in one of the largest and fastest-growing user groups of AI infrastructures
AMD's MI308 is part of the Instinct line, which is designed for processing and deploying artificial intelligence models at scale used in cloud services and large research environments.
The Instinct line is AMD's series of GPU accelerators - GPUs for general data processing - designed specifically for data centres, machine learning and high-performance computing (HPC) tasks.
For both companies, the ability to offer these products in the Chinese market means that they are present in one of the largest and fastest-growing user groups of AI infrastructures in the world.
The value of the Chinese market
The figures explain why Nvidia and AMD have accepted the additional costs.
According to financial reports, Nvidia made around 17 billion dollars in sales in China in the last financial year, which corresponds to around 13% of total sales. For AMD, China's market share is around 6.2 billion dollars, or 24% of total sales.
It is not yet known which products and services will be subject to the allocation obligation or how it will be calculated
Applying the agreed percentage to all qualified products could potentially generate billions of dollars annually for the US budget.
This is, of course, a hypothetical calculation, as it is not yet known which products and services will be subject to the allocation obligation or how it will be calculated – by invoice, quarterly or through a special calculation mechanism.
A legal and political issue
The agreement has also raised a legal dilemma: Is this 15% really a hidden tax on exports, which is prohibited under US law, or a legitimate administrative fee related to the granting of export licences?
Critics warn that the security argument of export controls can be called into question if the impression is created that the risk can be "bought off" by paying a certain percentage of the revenue.
There is also a political motive behind this: President Donald Trump's administration wants to avoid two extremes - a complete ban, which would accelerate China's technological independence, and unlimited exports, which would undermine US security policy.
Such logic creates the impression that US security is negotiable on a market basis
The fiscal "valve" in the form of 15% acts as a compromise: companies retain access to the market, and the state retains control and receives additional revenue.
Critics see this approach by the Trump administration as a dangerous departure from the basic principle of security policy: that national security is not for sale.
By introducing a model in which access to the most sensitive technologies can be secured by paying a certain percentage of revenue, sends a signal that even strategic risks have a price.
Such logic undermines the argument that export controls are a matter of principle, not trade, and creates the impression that US security is negotiable on a market basis.
Not only does this weaken US credibility in the eyes of its allies, but it also provides a propaganda weapon for opponents who have long claimed that Washington uses "security reasons" as a cover for economic interests.
Beijing's possible reactions
Beijing can respond in a variety of ways, none of which require direct formalisation.
Chinese state media have already labelled H20 as "unsafe" for the domestic environment, pointing to the possibility of technical "backdoors" and externally imposed restrictions. Such claims carry political weight regardless of their technical merit and can influence government procurement.
This could direct some of the demand to domestic manufacturers, such as Huawei, Biren, or Moore Threads
Informal recommendations to government companies and institutions to restrict purchases of products subject to the agreement are also possible, which could direct some of the demand to domestic manufacturers, such as Huawei, Biren, or Moore Threads.
Although these players cannot currently replace American accelerators in all segments, such measures would give them time and resources to develop competitive products.
Business impact for producers
In the short term, the 15% cost can be compensated in three ways: by increasing prices for Chinese customers, by reducing profit margins, or by redesigning services and software that fall outside the scope of the licences.
Regardless of which model they implement, both Nvidia and AMD will have to invest more in monitoring shipments, audits, and licence compliance, which will increase operating costs and business complexity.
For Nvidia, maintaining the CUDA development environment in China remains a priority
For Nvidia, maintaining the CUDA development environment in China (Compute Unified Device Architecture, Nvidia's software platform and a set of tools that enable its GPUs to be used for artificial intelligence processing and other sophisticated computing, thereby locking users into Nvidia's technology for the long term) remains a priority, as software dominance ensures long-term hardware sales.
AMD, with a weaker share in the AI segment, sees an opportunity to position itself as a sustainable alternative in the data centre sector, especially where compatibility with existing software environments is less of an issue.
The bigger picture and possible consequences
This agreement has implications beyond the bilateral relationship between the US and China. This precedent could extend to other sensitive technology areas such as quantum devices, advanced materials, and optical systems.
The question for allies, particularly in Europe and Japan, is whether US export controls are solely a security tool or also a fiscal policy instrument.
Security risks are no longer an absolute category but an item that can be quantified and included in trade agreements
If the 15% model proves to be sustainable, its expansion can be expected. This would mean that security risks are no longer an absolute category but an item that can be quantified and included in trade agreements.
In such an environment, companies would have to develop two strategies – one to operate in a system without additional fiscal burdens, the other to operate in an environment where market access is linked to a "percentage".
What happens next?
In the coming months, it will depend on the details that the US Department of Commerce will publish - the precise scope of products, the calculation and payment method and the reporting obligations.
At the same time, Congress and possible legislative initiatives that would call into question the constitutionality and legality of the model should be monitored.
American products are still technically irreplaceable in many segments, but each new round of restrictions and fiscal burdens gives more arguments to those in favour of China's technological independence - Jensen Huang
On the Chinese side, it will depend on how quickly domestic manufacturers can offer competitive solutions and how willing the government is to direct consumers towards them.
American products are still technically irreplaceable in many segments, but each new round of restrictions and fiscal burdens gives more arguments to those in favour of China's technological independence.
This regulation is not just a formality for export licences. It changes the rules of the game. It introduces a hybrid model that intertwines security, politics, and fiscal policy, replacing the traditional black-and-white regime of authorisation and prohibition.
Its sustainability will depend on legal resilience, political will and market realities - but it is already clear that the consequences will be felt by producers, buyers and governments on both sides of the Pacific.
