India is facing a technological crisis. Even as US and Chinese firms have released a parade of increasingly capable AI models this year, India has no competitive frontier model and no realistic prospect of producing one without significant policy shifts. The anticipated impact of AI can be separated into two categories. First, AI can diffuse through the economy by automating some routine jobs. Each industry will require specialised applications built on foundation models. India has an active start-up ecosystem devoted to building such applications, and businesses have rapidly adopted AI tools. However, AI will also have a strategic impact on research, cybersecurity and defence. AI models have been used to solve some of the most important open problems in mathematics. Anthropic’s Mythos model has formidable cybersecurity capabilities; it has not been released publicly, and is available only to selected organisations. Currently, consumers have access to other frontier models, including Chinese open-weight models. But it is unrealistic to expect this to continue indefinitely. Earlier this year, the US temporarily imposed export restrictions both on Mythos and on a version of Mythos with guardrails called Fable. It is likely to restrict and regulate AI to “achieve global dominance”. US industry leaders and academics have advised India to Focus on applications rather than the frontiers — a position supported by sections of the Indian IT industry. But countries that fall behind in frontier AI risk suffering the fate of those that missed the Industrial Revolution: A small business elite might find a niche and prosper, but ordinary people will suffer disempowerment. India can still catch up to the frontier with a focused effort. The algorithms underlying modern AI models are widely understood. The industry has exploited “scaling laws” that predict how a model’s performance improves with its size and the computing power used for its training. So, while better algorithms improve efficiency, the basic formula for producing frontier models is to scale compute and data. Lack of computing power has bottlenecked sovereign Indian model development. The IndiaAI mission has a pool of 45,000 GPUs. But this is only a fraction of the capacity controlled by a single US frontier lab. The mission allocated 4,096 GPUs to Sarvam AI to train India’s flagship model. This is about 50 times smaller than what is used to train frontier models. No amount of ingenuity can compensate for this resource gap. A number of data centres, with significant computing capacity, are coming up in various states. But these will primarily serve MNCs, and their location in India offers no tangible benefits. Most of the announced capital investment will be directed to electronic equipment, and the employment they create will be limited to a few construction and maintenance jobs. Large data centres have a significant environmental impact. In India, this impact will be borne disproportionately by local communities. India could use these data centres to alleviate its compute shortage with a “compute tax”: Any data centre established in India would be required to reserve, say, 25 per cent of its computing capacity for a publicly administered national compute pool. This capacity would remain physically within the data centre, but could be allocated by a central scheduler to Indian institutions. MNCs are likely to resist, but their bargaining position is weak given the growing hostility to these installations elsewhere. Such a tax would not obviate the other data-centre concerns. But together with environmental safeguards and welfare measures, it might open a narrow route to building a frontier model in India. The writer is professor of Physics, International Centre for Theoretical Sciences, Bengaluru . Views are personal
