The United States is preparing to further tighten its control of Nvidia and AMD AI chipsWhat until now were restrictions focused on a few countries could transform into a global licensing framework that determines who can build the next big AI data centers in the world.
This regulatory change doesn't just target China, although that is the main objective, but This would affect virtually every international AI hardware operation.For Europe, and in particular for Spain, the debate is not insignificant: access to these advanced GPUs is key for sovereign cloud projects, supercomputing, and the deployment of generative AI models in universities, companies, and public administrations.
A new global licensing system for AI chips
According to various drafts and leaks cited by media outlets such as Bloomberg and Reuters, the US Department of Commerce is working on a mandatory licensing regime to export AI acceleratorsThe change would mean moving from a limited approach to around 40 jurisdictions to a much broader scheme, in which Almost any shipment of advanced chips would require prior authorization from Washington.
In practice, this would mean that companies like Nvidia and AMD would have to request authorization for each relevant transaction for the sale abroad of high-performance GPUs. This is not a simple bureaucratic formality: the government would evaluate both the destination of the chips and their intended use, introducing transparency and security requirements which can affect the timelines and cost of the projects.
US authorities frame this shift within their strategy of protection of critical technologiesAI accelerators and supercomputing processors are considered sensitive components with enormous potential in both scientific research and military and national security fields. The stated objective is to prevent massive computing capabilities from ultimately strengthening the infrastructure of countries considered strategic rivals.
This approach is integrated into the export controls system overseen by the Bureau of Industry and Security (BIS), the Department of Commerce agency responsible for monitor the international transfer of strategic goodsIn recent years, the BIS had already tightened the rules for high-end chips, especially regarding China, but now the ambition of the framework would be much greater.
Tiered volume controls and caps on China
One of the key aspects of the proposal is the introduction of control levels depending on the volume of chips involvedThe drafts handled by the Administration contemplate a tiered scheme in which surveillance increases as the size of the order grows and, therefore, the computing power it concentrates.
To relatively small shipments, under a thousand chipsThere is talk of simplified and faster review procedures. The idea would be to avoid blocking smaller projects or research initiatives that don't represent a massive leap in computing power.
The tone changes when talking about large or bulk ordersThe documents mention six-figure thresholds: for example, transactions above 100.000 units would require formal guarantees between governmentswhile sales exceeding 200.000 GPUs could lead to the possibility of on-site inspections by US authorities and additional commitments on investment in AI infrastructure within the United States.
In this context, the idea, filtered through various sources, of establishing a limit of 75.000 H200 chips Nvidia by Chinese buyerIt wouldn't be a total ban, but rather a kind of rationing: sales are allowed, but only up to a certain point, so that no company can suddenly build a mega-cluster of AI using only American hardware.
The cap wouldn't only affect Nvidia. The same sources indicate that AMD accelerators They would be included in the same calculation: if a company in China acquires H200 chips and, at the same time, latest generation MI chips (such as the MI325 or successors), Everything would add up to calculate the allowed limit.This is intended to prevent restrictions from being circumvented simply by changing providers within the U.S.
China in the spotlight and the shadow of gray markets
China remains the central focus of these measures, even though the rules are designed with global reach. Washington has been trying for years to to curb Chinese companies' access to the most advanced chips from Nvidia and AMD, both because of the battle for leadership in AI and because of the fear that those resources will end up in military or surveillance projects.
According to the leaks, the US government doesn't just want to know How many GPUs are being shipped and wherebut also what they will be used for. Hence the proposal to require companies to build large AI clusters disclosure of their business models and the possibility of US officials visiting the facilities where those systems are assembled and operated.
One striking aspect of the drafts is the responsibility they attempt to shift onto the manufacturers themselves. Nvidia, AMD, and other exporters would have to monitor the deployment of their chips and even implement software that prevents the grouping of a large number of units into megaclusters that exceed certain capacity thresholds.
However, while Washington tightens official oversight, gray chip markets They are gaining prominence in ChinaThe restrictions, far from eliminating demand, push it toward less transparent channels. US authorities are aware of this phenomenon and, according to sources consulted, are trying to design a licensing system that also hinders resale and triangulation through third countries.
The problem for Nvidia and AMD is obvious: China is one of the world's largest semiconductor marketsLimiting direct sales while simultaneously pursuing alternative routes has a potentially huge impact on current and future revenues, although for the moment companies continue to show very solid growth supported by other customers, especially large American and European technology companies.
From case-by-case review to structural control of AI infrastructure
Until now, many operations were being managed through licenses on a case-by-case basisespecially in the case of Nvidia's H200 chips. Authorities reviewed each application based on the technical configuration and the end recipient, granting partial permits that allowed exports limited in volume and capacity.
The new development is that the White House is considering moving from that reactive approach to a preventive and structural control modelUnder this new framework, the priority would no longer be so much which specific chip is shipped, but who is allowed to build large computing infrastructuresIn practice, the United States aspires to play a decisive role in the global distribution of computing power for AI.
Sources at the Commerce Department indicate that this shift is based on a review of previous regulations, known internally as "AI diffusion" frameworks. The current trend within the Administration is centralize control more and reduce broad exemptions that were granted to certain allies, although these continue to enjoy preferential treatment compared to countries considered to be at risk.
This approach opens the door to another key element: linking the granting of licenses to investment commitmentsSome drafts stipulate that countries wishing to import large volumes of AI chips must invest in data centers and AI ecosystems within the United States, thereby strengthening their own national digital infrastructure.
In political terms, this formula transforms export controls into a broader negotiating tool. Allies and partners could find themselves in a position of having to accept it. economic and security considerations in exchange for maintaining a stable flow of advanced hardware for their AI projects.
Impact on Nvidia, AMD, and the global semiconductor industry
The market's reactions were swift. After the proposals were revealed, Nvidia and AMD shares registered declinesThis reflects investor nervousness about a more complex regulatory environment. Although these are still drafts and the final rules could be softened, the mere discussion already introduces an additional element of uncertainty.
In Nvidia's case, the restrictions add to the pressure already facing its business in China. The company had managed to maintain growth by relying primarily on large clients in the United States and other developed markets, but any regulatory ceiling on global sales volume may affect their medium-term forecasts.
AMD is no exception to this trend. Its data center accelerators, which compete directly with Nvidia's, are poised to be one of the company's growth engines in the coming years. The possibility that Sales of AI GPUs will be subject to government authorizations This poses an additional risk to their plans, especially in markets not considered full allies of Washington.
Aside from the stock market reaction, experts emphasize that we are facing a possible turning point in the semiconductor industryThe chip supply chain is deeply global, and any substantial change in export rules by the United States has effects on designers, foundries, systems integrators, and cloud service providers worldwide.
Several governments have responded to this growing tension with programs to support domestic semiconductor productionThe United States, Europe, and Asian countries are promoting incentives to attract factories, design centers, and R&D projects, seeking to reduce dependence on third parties in a context of increasingly marked technological rivalry.
What does all this mean for Europe and Spain?
For the European Union, Washington's move comes in the midst of the deployment of its own strategic autonomy strategy in chips and artificial intelligenceThe EU has long warned of the need to ensure a stable supply of advanced semiconductors, both for economic reasons and for technological sovereignty.
If the new US rules are applied broadly, European projects on sovereign cloud, supercomputing and large AI models They could be forced to adapt to a more stringent licensing system. Although EU countries are considered allies and would have an advantage over other markets, this would not exempt them from complying with additional security, transparency, or investment requirements.
In Spain, the issue is directly connected with initiatives such as the PERTE, dedicated to digitization and semiconductorsas well as with plans to strengthen the computing capacity of public bodies, universities, and companies. Access to advanced GPUs from Nvidia and AMD is a critical factor for training proprietary models, boosting AI projects applied to healthcare, industry, and government, and attracting international data centers.
Stricter regulation from Washington could, on the one hand, to increase the cost or slow down the arrival of hardware for certain projects within European territory. On the other hand, it can also serve as a catalyst to accelerate the implementation of chip manufacturing and design infrastructures within the EU, reinforcing the community's commitment to reducing dependence on foreign countries.
In this balance between security and openness, European technology companies are forced to rethink their procurement and deployment strategiesChoosing where to build new data centers, which technology partners to prioritize, or how to distribute AI workloads across different regions becomes not only an economic issue, but also a regulatory and geopolitical one.
The ongoing debate in the United States makes it clear that the The race for AI chips is being fought both in laboratories and in regulators' offices.What Washington decides in the coming months could redefine who has access to the computing power needed to lead the next wave of innovation in artificial intelligence, and Europe, including Spain, will have to move carefully to avoid being left behind or excessively tied to decisions made elsewhere.
