UPI-like AI, credit-based fellowship for kids, caution for IT: Economic Survey on AI

Instead of expensive frontier models and large language models, the Survey advocates for India to focus on application-specific small models tailored to defined sectoral needs. These computationally efficient models can run on locally available hardware like smartphones or personal computers, enabling distributed innovation across firms, sectors, start-ups, research institutions, and public agencies without requiring proportionate expansion in expensive, resource-intensive data centres.

The Survey bats for a “bottom-up” approach to AI development, a decentralised path that allows AI capabilities to spread widely across sectors while avoiding the high capital expenditure, energy intensity, and hardware dependencies that characterise Western frontier model development.

Despite being late to the AI race, the Survey notes that India enters the AI era with formidable strengths that make a bottom-up approach viable. Stanford University’s Artificial Intelligence Index Report 2025 ranks India as having the world’s second most AI-literate workforce, behind only the United States.

AI for students

The survey proposes a radical restructuring of education through an ‘Earn-and-Learn’ initiative where students would begin earning both academic credits and paid work experience through apprenticeships and project placements as early as Class 11. The structured, credit-bearing industry fellowships could be co-designed by the private sector and academic institutions, with both practical experience and wages contributing to formal degrees. This would be a key way to unlock India’s demographic dividend, as per the Survey.

The Survey cites international examples, including China’s Young Thousand Talents Program (which has boosted domestic research productivity significantly), EU member countries’ industry-academia collaborations, and emerging models like Palantir’s Meritocracy Fellowship offering high-school graduates immediate industry experience.

Creating new jobs

The Survey calls for a comprehensive sectoral mapping of jobs outside the white-collar workspace, which have a high-skill requirement but are understaffed, which is an often-overlooked source of new jobs within the economy. This could also subside some of the impending risks that AI can have on current jobs on offer.

For instance, the Survey says that nursing and geriatric care are already understaffed, and the doubling of India’s dependency ratio in the next decade will create additional demand for skilled labour in this sector. Other tasks involving high-skill with long apprenticeship curves include culinary sciences, advanced metalwork, experiential hospitality designers, surgeons, physiotherapists, advanced electricians, early childhood educators, among many others. Sectors like these should be identified, and the education and skilling infrastructure must be upgraded to impart the necessary knowledge to fill the labour supply gaps. Opportunities in the physical, human-centric space and hands-on jobs hold immense potential for creating meaningful jobs in the coming decade.

India’s structural challenges, and a note of caution for the IT sector

The Survey notes that India faces some structural challenges. The country’s access to cutting-edge compute infrastructure remains severely limited, with Indian start-ups focused on curating training data representing only 2% of the global total, compared to 40% in the United States, 21% in the European Union, 9% in the United Kingdom, and 5% in China. Over 70% of all data centres by count are located in high-income countries, with India accounting for just 3%, according to World Bank data cited in the Survey.

Financial resources for large-scale model training are scarce, and private participation in foundational AI research is relatively muted in India compared to global leaders. Most critically, the Survey says that even with abundant domestic capital and strong demand, India’s AI infrastructure expansion would be severely constrained by global GPU supply chains.

Since control over data and compute needed for AI is highly concentrated, it raises concerns about market power, technological dependence, and the resilience of supply chains. It also raises a substantial question about the future of India’s IT sector, as firms that once relied on India’s comparative advantage to handle a bulk of their work may no longer need to do so, the Survey said. “It risks hollowing out India’s core value proposition if adaptation lags. If the country is to sustain its competitive edge in IT, a comprehensive evolution is necessary, one that takes full advantage of the potential embedded in AI development and deployment,” it added.

The trade-offs

The Survey also highlighted some trade-offs that India would need to make in its AI journey:

LLM vs applications: India would need to make a trade-off between expending scarce resources to chase frontier-scale models or deploying those resources more effectively towards domain-specific AI systems aligned with domestic economic priorities.

Labour displacement vs efficiency: For labour-abundant economies such as India, AI creates a tension between aggregate productivity gains and employment absorption. Rapid, uncalibrated deployment of AI may boost output but risks displacing segments of the workforce faster than the economy can reabsorb them. Conversely, delaying adoption to protect jobs may risk locking firms into a low productivity equilibrium. The policy challenge, therefore, is not whether to adopt AI, but how to pace its diffusion so that labour augmentation can be facilitated.

Open vs closed AI models: The most widely deployed AI models are proprietary and opaque, limiting transparency around their training data, internal logic, and update mechanisms. Open-source models and open-weight models offer lower entry barriers, greater adaptability, and reduced vendor lock-in, with some trade-offs. India’s challenge is to strike a balance between openness and stewardship, leveraging shared innovation while ensuring that the economic value created from domestic data and intellectual property accrues within India rather than being captured abroad.

Compute strength vs resource allocation: The Survey notes that developing AI compute by means of data centres is extremely taxing on shared resources like electricity and water, with some firms projected to burn half a trillion in cash by 2030 while pursuing compute infrastructure. For India, as power, finance, and especially water remain binding constraints, scaling compute indiscriminately carries opportunity costs. This creates a trade-off between centralised scale and distributed efficiency, strengthening the case for smaller, task-specific models that can run on limited hardware and decentralised compute networks.

Innovation vs regulation: Regulatory compliance, aimed at ensuring safety, controlled proliferation, auditing, transparency, or mitigating liability exposure, imposes fixed costs that may scale poorly for smaller firms and those involved in early-stage experimentation, the Survey said. And while rich countries and firms can absorb these costs, India’s more fragmented and resource-constrained innovation landscape could be stifled if the same degree of regulations were binding. However, very minimal or completely absent regulatory clarity can undermine trust, slow adoption, and create systemic risks, particularly as AI is deployed in critical sectors such as healthcare, education, governance, or finance. “The challenge for India, therefore, is how and when to regulate AI,” it added.

Strategic autonomy vs global integration: AI has emerged as a geostrategic asset, and as the technology sees adoption in critical sectors and public institutions, an overdependence on foreign systems could carry systemic risks, the Survey said, even as it acknowledged that complete technological self-sufficiency is neither feasible nor efficient. For India, the trade-off would be preserving openness where it enhances capability while insulating critical functions from external shocks.