Eco Survey: AI strategy must be grounded in economic realities, not scale

Chasing scale, building large LLMs for own sake not useful for India

India’s strategy for artificial intelligence (AI) must be grounded in the country’s economic realities rather than chasing scale and building large language models like other advanced economies, the Economic Survey for 2025-26 has suggested. 

Cautioning against the rapid development of AI infrastructure, including the expansion of data centre capacity, the Survey said the recent phase of highly leveraged AI-infrastructure investment has exposed business models that depend on optimistic execution timelines, narrow customer concentration and long-duration capital commitments. 

“A correction in this segment would not end technological adoption, but it could tighten financial conditions, trigger risk aversion and spill over into broader capital markets,” the economic survey has noted, adding that the macroeconomic consequences due to such a crisis could be worse than the 2008 global financial crisis. 

In countries such as India, where resources such as electricity, water and capital remain scarce, indiscriminate scaling of energy-intensive AI infrastructure is not sustainable in the long run, the Survey said. These resources also compete directly with similar demands from households and other industries. 

According to Nasscom, despite generating nearly 20 per cent of the world’s data, India hosts only about 3 per cent of global data centres, around 150 out of 11,000 worldwide. 

With the growing adoption of AI technologies and regulatory developments such as the Digital Personal Data Protection (DPDP) Act, 2023, and the RBI’s data localisation guidelines, demand for data centres is likely to rise in the coming years, according to an estimate by JLL. 

Instead of pursuing indiscriminate scaling, India should focus on “application-led innovation, the productive use of domestic data, human capital depth, and the ability of public institutions to coordinate distributed efforts”. 

“A bottom-up strategy anchored in open and interoperable systems, sector-specific models, and shared physical and digital infrastructure offers a more credible pathway to value creation than a narrow pursuit of scale for its own sake,” the survey said. 

The first phase of implementing this AI policy should involve operationalising institutions already announced by the central government under various schemes, including the India AI Mission. 

The government should also incentivise experimentation with AI and encourage grassroots-level research by expanding access to shared AI infrastructure onboarded under the India AI Mission. There should also be a clear focus on applications and on sector-specific, small and open-weight models to enable efficient resource utilisation. 

Once these mechanisms begin to function and startups that are allowed to experiment with AI generate evidence of public and economic utility, medium-term policy should focus on selective scaling. 

“Shared and certified domestic computing infrastructure should expand, with voluntary participation by large and resourceful firms linked to regulatory facilitation and access to public datasets. At the same time, AI regulation should be formalised on a risk-based and proportionate basis,” the Survey noted. 

In the long term, India’s focus should be on building resilience by reducing dependence on imports of advanced hardware such as Graphics Processing Units, the Survey said. 

AI startup ecosystem: A snapshot

 

Deeptech companies saw a 78 per cent rise in funding during calendar year 2024 (CY24). The number of active Generative AI startups rose more than threefold to over 890 by the first half of 2025 from about 240 in the year-ago period. A majority of those Gen AI startups, 39 percent, are based in Karnataka, India’s technology capital, followed by Maharashtra, 14 per cent, and Delhi at 9 per cent.