Washington’s first comprehensive AI diffusion framework highlights the need for India to solidify its role as a key US ally amid rising global demand for AI chips
Image Source: Getty
The growing technology war between the United States (US) and China has led to a growth in export control restrictions across a wide range of critical and emerging technologies, starting with the US’s export control regulations on semiconductors in 2019 under the first Trump administration. The US Artificial Intelligence Diffusion Framework, announced in January 2025, is the latest attempt by the US government to curb China’s evolving prowess in the domain of Artificial Intelligence (AI) and is bound to have far-reaching consequences for both India and the world at large.
The Framework for Artificial Intelligence Diffusion is the first comprehensive framework put in place by Washington to manage the diffusion of AI technology at a global level. It builds on the export controls on advanced AI chips or Graphical Processing Units (GPUs) passed in October 2022, and constitutes a broader attempt to limit the technological capabilities of the countries the US perceives as a threat, prominently China. The framework modifies existing US export controls on AI technology primarily in two ways:
| Tier 1 | Tier 2 | Tier 3 |
| Includes 18 major US partners and allies including Australia, Canada, Japan, South Korea, Taiwan, the United Kingdom and most North Atlantic Treaty Organization (NATO) members. | Comprises of all countries not included in Tier 1 and Tier 3, which includes most nations, including India. | Consists of all US arms-embargoed countries, including China, Russia and North Korea. |
Tier 1 possesses unrestricted access to chip imports if the entity importing them is headquartered within Tier 1 countries. Exports are effectively banned to Tier 3 countries, in line with the export controls incorporated in October 2022. Tier 2 nations are the ones who must primarily contend with most of the changes put in place.
The US Bureau of Industry and Security had previously established the Data Center Validated End User (DC VEU) programme in September 2024, which provided licenses on a case-by-case basis to Tier 2 countries. This has been expanded upon and streamlined by bifurcation into two DC VEU categories: Universal Validated End Users (UVEUs) and National Validated End Users (NVEUs). Only Tier 1 companies can avail of UVEU for deploying data centres in Tier 2 nations. They are, however, subject to the restriction that at least 75 percent of their controlled computing must be maintained in Tier 1 countries while it must be less than 7 percent in Tier 2 countries.
Figure 1: Computing Power Allocations for UVEU Companies.
Source: RAND
Tier 2 companies need to obtain a separate NVEU authorisation for each Tier 1 or Tier 2 country they intend to export to, which are subject to absolute caps.
Figure 2: Absolute caps for each NVEU company based on Total Processing Power (TPP) in terms of NVIDIA H100-equivalent units.
Source: RAND
Tier 2 countries can also receive additional chips through standard one-time export licenses for which country caps have been set at 50,000 H100-equivalent chips. Small-quantity exemptions have also been set for individual companies in Tier 2 countries at 1,700 H100-equivalent chips, which do not count towards country caps.
Finally, the framework imposes stringent security requirements on data centres, including cybersecurity, and physical and personnel security, to restrict theft, unauthorised access and illegal export of chips to Tier 3 countries.
The primary motivation behind the framework is to maintain US dominance in the domain of AI, which is increasingly acquiring its position as one of the most critical emerging technologies in the world today. It attempts to do so by curbing access to AI technology by Tier 3 countries, particularly China, which has been making steady progress in the field, including in military applications and surveillance, and has emerged as the most prominent threat to US hegemony in the technology sphere. The primary strategy to achieve this is by limiting compute power in these countries via three main facets: advanced AI chips, cloud access and model weights.
The US Commerce Department has also sent letters to major chip-makers, including Taiwan Semiconductor Manufacturing Company and Samsung, informing them about the restrictions.
The framework is also a response to recent reports of smuggling and backdoor channels being employed by China to circumvent export controls through Southeast Asian nations like Singapore, Hong Kong and Vietnam. Additionally, Chinese entities had set up subsidiaries in some of these countries to sidestep these rules and purchase chips. The US Commerce Department has also sent letters to major chip-makers, including Taiwan Semiconductor Manufacturing Company and Samsung, informing them about the restrictions. There have also been reports that Russia has been illegally importing AI chips through India.
Long-term effects
The US is the global leader in AI compute capacity through companies like Nvidia and the framework is likely to cement this position. Although China has made significant progress in the field, shifting to a Chinese AI ecosystem might prove cumbersome and unattractive for most countries on account of their distinct software requirements and limited performance. China’s GPU alternatives, such as the Huawei Ascend series still lag behind Nvidia by at least one or two generations. Moreover, China’s GPU production capacity now is also significantly less than the US. Consequently, the framework will likely enable a lock-in of US AI infrastructure and technology across the globe, effectively phasing out Chinese competition.
Another possible impact of the framework is that there might be a global shift toward public models and a renewed focus on increasing compute efficiency rather than power alone, as exemplified by the recent release of China’s DeepSeek’s open-source Large Language Models. While this is certainly a viable alternative, the fact is that even open-source models need compute power, for which they are ultimately dependent on AI chips. For instance, DeepSeek itself was trained using Nvidia’s H800 GPUs. With the security protocols put in place by the US framework, even training public models will likely turn out to be problematic.
The US is the global leader in AI compute capacity through companies like Nvidia and the framework is likely to cement this position.
Short-term effects
In the short term, most Tier 1 countries are likely to benefit from the framework. For instance, Australia stands to benefit enormously, thanks to its prowess in providing colocation facilities through industry giants such as AirTrunk, NextDC, and Canberra Data Centres. On the other hand, most Tier 2 countries are likely to suffer, particularly those in Southeast Asia, including Malaysia, Singapore, and Indonesia. Malaysia, for instance, has ambitious plans to raise its data centre capacity significantly by 2027 to about 3.5 GW. Consequently, it has incurred significant investments from US tech giants such as Nvidia and Oracle, while simultaneously being a hotspot for Chinese investments through firms such as ByteDance, the parent company of TikTok. Thus, it will likely suffer significant setbacks due to the stringent requirements of the framework.
However, there is a silver lining for some Tier 2 nations. The creation of the UVEU and NVEU categories will streamline access to AI chips by Tier 2 countries located in the Middle East and Central Asia such as the United Arab Emirates (UAE) and Vietnam, which had to procure licenses on a case-by-case basis earlier.
India’s National AI Mission aims to develop infrastructure with over 10,000 GPUs over the next five years. India is also one of the few Tier 2 countries with a 3 GW (Giga Watts) planned data centre capacity. Reliance Industries has announced a 3 GW mega data centre at Jamnagar, Gujarat, intended for AI workloads. The US framework is likely to pose a challenge to these endeavours. To ensure India’s AI ambitions do not get disrupted, New Delhi will need to adapt quickly. Some possibilities include:
Therefore, in the long run, and with the ever-increasing demand for AI chips, it would be prudent for India to maintain its position as a strategically important ally of the US, since this would enable its AI ecosystem to grow seamlessly under the aegis of the framework. Further building on collaborative initiatives such as the US-India Initiative on Emerging Technology (iCET) and the establishment of a semiconductor assembly and test plant by Tata and Micron, are bound to have a positive effect in this direction.
Prateek Tripathi is a Junior Fellow at the Centre for Security, Strategy and Technology, at the Observer Research Foundation
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Prateek Tripathi is an Associate Fellow at the Centre for Security, Strategy and Technology. His work focuses on an emerging technologies and deep tech including quantum ...
Read More +
PREV
