Artificial intelligence (AI) capabilities are based on four main pillars: algorithms, data, talent, and AI computing, the latter of which is often considered the most important. Graphics processing units (GPUs) have become the standard measurement for AI computing, spurring a global race to build massive GPU capacity.
In this context, the government's decision in March to deploy more than 10,000 GPUs as part of the India AI Mission was a major step forward. The effort, accomplished through a public-private partnership (PPP), reflects India's willingness to leverage the strengths of the private sector and marks a departure from past practice of confining such capital-intensive national capabilities to the public sector.
Click here to connect with us on WhatsApp
India's AI computing capabilities are modest compared to leaders such as the US and China. The global GPU market is estimated to be valued at $23 billion in 2022, with 1-2 billion GPUs worldwide, the majority of which are in the US and China. Both countries have been aggressive GPU buyers over the past decade and have bigger plans to increase their AI computing capabilities. The US launched its National AI Research Resources Program in January, while China aims to increase its total computing capacity by more than 50% by 2025, including 10 exascale systems. India cannot and does not need to copy their strategies. Instead, it should adopt a smart fast-follower approach, which I call BharatCompute.
Why is AI computing so important? In AI, “compute” refers to the computational power required to perform complex operations like training and running AI models. This is essential for optimizing deep learning models that often have millions or billions of parameters with large datasets. Additionally, compute power is also needed for real-time inference to enable predictions from new data.
Over the past decade, the computational needs of large language models (LLMs) have driven a surge in global demand for AI computing. However, as LLMs have become more widely available, this demand has stabilized and the focus has shifted to less computationally intensive models such as image processing, gaming, and multimodal AI. BharatCompute can prioritize the development of foundational models in the economic, social, and strategic domains while leveraging existing open source LLMs that have already been developed. This approach promises significant benefits even when computational resources are scarce.
BharatCompute should also aim to leverage India's existing CPU or central processing unit-based computing power for AI advancements. India's strong information technology (IT) industry and large user base provide ample CPU-based computing power to make up for the shortage of GPUs. Although CPUs are slow for AI tasks, leveraging this resource can increase accessibility and reduce costs.
The BharatCompute initiative advocates making government-built AI computing infrastructure available to those who cannot afford it at market prices. Currently, AI computing costs between $1-4 per GPU hour, plus additional costs for memory and IT infrastructure. Government subsidies could encourage more startups to enter the AI space, and the growth of AI startups could be an indicator of the success of the initiative. The government could also prioritize development of foundational models in areas like agriculture, education, healthcare, and water to align with Indian requirements.
BharatCompute must support India's sovereign AI needs, especially in the strategic and security domains. Currently, the AI computing market is heavily dependent on cloud infrastructure managed by US and Chinese hyperscalers. Research shows that US-based Amazon Web Services, Microsoft Azure, and Google Cloud account for around 70% of the global public cloud market, while Chinese tech giants Alibaba, Huawei, and Tencent control most of the rest. Public cloud raises privacy and security concerns. To ensure security and sovereignty, BharatCompute must mandate that computing infrastructure be physically located in India and use Indian cloud platforms. The government can allocate a portion of computing infrastructure in a secure non-cloud environment for sensitive or classified applications.
Building AI Computing on a PPP model is a welcome step. It can be operationalized by a panel of Indian vendors contributing to a government guaranteed pool of 10,000 GPUs. Such a process is much quicker than building greenfield facilities and ensures a competitive ecosystem that delivers cost-effective, high-quality services. The private sector brings resources, skills, and innovation. However, a PPP framework needs to address potential pitfalls such as continuity in case of vendor change, avoiding vendor lock-in, data security and privacy issues, and ensuring access for MSMEs and start-ups. It is therefore envisaged that the governance of BharatCompute should be an autonomous organisation led by industry leaders with expertise, with government representation.
The Indian AI computing market is still in its infancy. While Indian industry is ramping up AI computing infrastructure, some developing for in-house use and others offering it as a service, BharatCompute’s installation of 10,000 GPUs is not enough. We should stimulate the larger AI computing market and encourage the industry to significantly expand its capacity. Giving preferential access to domestic companies for 10,000 GPUs would give the industry a much-needed start. We recommend encouraging private investment in AI computing infrastructure and proposing tax incentives to support customized AI chip development for foundational models. The burgeoning market for customized AI chips highlighted by Nvidia’s growth, OpenAI’s Sam Altman’s $7 trillion chip manufacturing venture, and Microsoft and Amazon’s entry into AI chip design suggests significant new opportunities. With an abundance of chip design talent, India is well positioned to capitalize on it. India is rich in talent and data. BharatCompute could be a play to leverage these strengths to establish an AI advantage.
The author is Distinguished Visiting Professor at the Indian Institute of Technology, Kanpur and former Secretary of Defence.
