After a long hiatus, let’s checkout a recent policy paper by the folks at XKDR Forum on Assessing ‘Compute’ Capacities and Policy Frameworks for Artificial Intelligence in India
Here are a few key excerpts from the policy paper to help understand the AI compute landscape in India and policy recommendations to improve the situation
“Indian policies regarding the promotion of “compute” can be broadly divided into two: policies relating to computing infrastructure, and policies relating to development of semiconductors.”
Policies on computing infrastructure - the National Super Computing Mission “70 high-performance computing facilities at various research institutes and universities with a budget outlay of INR 4500 crores spread across 7 years”
Policies on semiconductors - The India Semiconductor Mission (ISM), initiated in 2021 with a budget of INR 76,000 crores across 5 years, is intended to be “a groundbreaking effort aimed at transforming India into a worldwide center for semiconductor manufacturing and design”
The production of semiconductors in India is a sign of how the Indian economy is gradually climbing the value-addition scale. Producing large chips that are required for non-intensive uses like household appliances, etc. at scale will certainly help generate technical expertise provided India is able to do so at low costs. While this does not have a direct bearing on high-performance computing in general, it does enable the transfer and development of skills and know-how.
The impact of external factors on “compute”
Regd the Framework for AI Diffusion and the recent Interim Final Rule issued by the Dept. of Commerce, USA - The intent of the rule is clear: it seeks to restrict AI technology diffusion by restricting imports of GPUs. The rule, in fact, specifically targets Chinese access to AI technologies. At the same time, they also have the effect of limiting the scope of deploying these technologies in IndiaThere are also new reporting requirements for US companies to detect and report “unauthorized training” of models in Tier 2 and Tier 3 countries. Some of these restrictions (such as those on unauthorized training) would apply even if a company from a Tier 2 country is “renting” a set of servers in a Tier 1 country.
How should Indian policymakers analyze these developments? Firstly, we note that in India, most AI products are applications and not models. To put the Indian state of play in the global context, it would be The impact of external factors on “compute” Assessing ‘compute’ capacities and policy frameworks for artificial intelligence in India 09 5 useful to use Lehdonvirta et al. (2024)’s positioning of countries with significant prowess in AI into “Compute North” and “Compute South” countries. Compute North countries are “positioned to use their territorial jurisdiction to intervene in AI development at the point at which models are sent to their local public cloud regions for training. For instance, they could require algorithms and data sets to be audited and certified for compliance with their local rules before training is permitted to commence, shaping what kinds of AI systems can enter the global market”. Compute South countries are better suited for “AI system deployment than for development”. Using this classification, we could put India somewhere in the middle — a country that is ahead of its peers when it comes to the deployment of models and applications to be generated out of it, but is not quite ready to make meaningful changes to the direction of the field itself.
Two interesting responses have emerged to these developments from the government of India. The first and older response is that of the idea of “AI sovereignty” and the creation of a “National AI Stack”. In their concept paper, the Department of Telecommunications, Government of India (2020) identifies the strategic challenges regarding AI adoption in India (e.g., issues concerning access to technology, availability of trained researchers, etc.) It goes on to suggest that a National AI Stack could be a viable answer towards ensuring that all the functional requirements of AI research and services are met in India.
The newer response has been that of the Indian government’s recent RFP to build foundational models in India (Ministry of Electronics and IT, Government of India, 2025b). The developers selected to build an Indian LLM using various products developed under the other pillars of the IndiaAi mission: the developers would have access to GPUs acquired under the “compute” pillar and they would obtain research funding and training data for the same. This development marks a credible and serious response by the Indian government to strengthen its commitment as a serious AI power. At the same time, concerns have arisen over whether these products would run the risk of underutilization, and therefore, are premature.
Policy options to support improvement in “compute”
- “Buy” or “rent” should be a commercial decision
- Reforms in procurement
- Foreign policy that ensures the continued supply of “compute”
Given that the United States and its allies control the overwhelming majority of the supply chain for high-performance computing, it is best to find diplomatic avenues to improve India’s status to obtain preferential treatment for supplies of GPUs and other advanced processors. This has been done before — for example, in 1984, the United States relaxed certain supercomputer export restrictions on India
In an uncertain trade environment — when various countries are beginning to increase tariffs on important goods — it is crucial that the Government of India avoid any tariff rammifications on AI R&D. This would be done by negotiating reduced tariffs on essential AI hardware inputs through bilateral and multilateral trade agreements can lower production costs and encourage domestic manufacturing.
Strategic trade partnerships with the countries that are major players in AI and semiconductors can facilitate technology transfer, allowing Indian firms to access cutting-edge innovations and collaborate on R&D. Additionally, easing foreign direct investment (FDI) norms and including AI-specific provisions in trade deals can attract global tech giants to set up fabrication units and data centers in India. By integrating trade agreements with initiatives like the India Semiconductor Mission and Make in India, the country can create a robust ecosystem for AI-driven computational infrastructure, positioning itself as a competitive global hub.