Reforms 3.0: Charting India's Path to an AI-Powered Economy
As India targets sustained high growth, a new policy push aims to replicate the success of its digital public infrastructure in the age of Artificial Intelligence by making AI compute and 'tokens' abundant and accessible. Here’s what the proposal entails, the challenges it faces, and the risks involved.
Pre-requisite: Understanding the AI Policy Landscape
To grasp the debate around India's next wave of economic reforms, termed 'Reforms 3.0', one must understand the foundational concepts, historical context, and institutional players shaping the country's technology policy.
(1) KEY TERMS
- Artificial Intelligence (AI) Tokens: The basic units of data, such as words or parts of words, that Large Language Models (LLMs) process to understand and generate text. The cost of using an AI model is often measured by the number of tokens consumed.
- Large Language Models (LLMs): A type of AI system, such as Sarvam or Llama, trained on vast amounts of text data to understand, summarise, generate, and predict new content.
- Digital Public Infrastructure (DPI): A set of shared digital systems, such as digital identity (Aadhaar) and real-time payments (UPI), that serve as a foundational layer for economic and social services, built and managed with a public-good orientation.
- Sovereign AI: A nation's capability to develop, deploy, and control its own AI infrastructure and models, reducing dependency on foreign technology providers and ensuring data residency and security.
(2) BACKGROUND & TIMELINE
The current discussion is rooted in India's economic history. For decades after 1947, the economy grew at a modest 3%, often termed the 'Hindu rate of growth'. A severe balance-of-payments crisis in 1991 triggered the first wave of major economic liberalisation under Prime Minister P.V. Narasimha Rao, ushering in an era of higher growth.
A second transformation began in the 2010s, focused on digital infrastructure. The Aadhaar project, enabled by the Aadhaar Act of 2016, enrolled 1.38 billion people. The Unified Payments Interface (UPI) launched in 2016, revolutionising digital payments. In September 2016, the launch of Reliance Jio led to a dramatic fall in mobile data prices from approximately $3 per GB to $0.10 per GB in under three years, creating a mass market for digital services. Proponents of 'Reforms 3.0' argue that AI presents a similar transformational opportunity, requiring a policy playbook analogous to the one that created data abundance.
(3) INSTITUTIONAL FRAMEWORK
Several government bodies are central to shaping India's AI strategy. The Ministry of Electronics and Information Technology (MeitY) is the nodal ministry for IT policy and oversees the IndiaAI Mission. NITI Aayog (National Institution for Transforming India), the government's premier policy think-tank established in 2015, has been instrumental in drafting national strategies. Its June 2018 discussion paper, 'National Strategy for Artificial Intelligence: #AIforAll', outlined a vision for inclusive AI development. These institutions work alongside academic centres of excellence like the Indian Institutes of Technology (IITs) and the Indian Institute of Science (IISc) to foster research and development.
The discourse on 'Reforms 3.0' posits that just as the 1991 liberalisation unshackled the economy and the 2016 data revolution democratised internet access, a concerted push towards AI can unlock the next phase of high growth. The central thesis is that India must treat AI compute as a strategic national resource, akin to power or roads, and make it widely accessible.
What is the core proposal for an AI-powered economy?
The central proposal is to make the fundamental units of AI computation—tokens—as abundant and inexpensive as mobile data became after 2016. The argument, articulated by policy proponents like IAS officer Srivatsa Krishna, is that India should shift from subsidising 'calories, chemicals, and carbon' to subsidising 'cognition'. The plan suggests an initial annual investment of approximately $2 billion, or 0.06% of GDP, to provide free AI token access to the country's top 100 research institutions and 5,000 high schools. This figure is about one-fourteenth of India's food subsidy and one-tenth of its fertilizer subsidy, which has grown at an 11% compound annual growth rate over the past decade (Source: The Hindu).
How does the government plan to achieve this?
The strategy does not advocate for direct government subsidies but a three-pronged approach modelled on the success of DPI and the telecom sector. First, it suggests reallocating funds by freezing the growth of large subsidies, such as for fertilizers, for a single year. Second, the government would forge public-private partnerships (PPPs) with global hyperscalers like Amazon Web Services (AWS), Google, and Microsoft, offering incentives like land and power subsidies in exchange for free or low-cost AI inference capacity. According to this view, India's market of 1.4 billion users provides significant leverage. Third, the model relies on cross-subsidisation, where paid enterprise-tier AI services would fund free access for students and researchers. The government's role is envisioned as a facilitator, creating a regulatory environment similar to the one that enabled the data boom through policies like the National Digital Communications Policy, 2018.
Why is developing sovereign AI capability a key focus?
A critical component of the proposed reforms is building the national capacity to host and operate LLMs, rather than merely consuming them via APIs from foreign companies. This emphasis on 'Sovereign AI' is driven by four key considerations: geopolitical autonomy, cost-effectiveness, cultural customisation, and security. Hosting models on domestic infrastructure reduces reliance on foreign APIs that could be restricted. Eliminating per-token licensing fees for open-source models makes widespread free access for education economically viable. Sovereign capability also allows for fine-tuning models for India's 22 official languages and diverse cultural contexts. Finally, having auditable model weights is crucial for sensitive government and defence applications. This mirrors global trends, with nations like France backing domestic models like Mistral AI to ensure digital sovereignty.
What are the strategic challenges in building this infrastructure?
The primary challenge is the high concentration of power in the AI hardware supply chain. US-based firm NVIDIA currently controls over 80% of the market for the specialised GPUs required for AI training, creating a strategic vulnerability and high financial costs (Source: The Hindu). To mitigate this 'vendor lock-in', the proposed strategy is to diversify India's compute hardware portfolio. Instead of relying solely on NVIDIA, it suggests a '40:30:30' mix: 40% of infrastructure built on cost-effective inference chips from firms like AWS and AMD; 30% on Google's TPUs for research; and the remaining 30% on NVIDIA for specialised tasks. This diversification is deemed essential to build sovereign AI infrastructure at a scale suitable for 1.4 billion people without incurring prohibitive costs.
What are the broader risks and ethical concerns?
While the economic potential is significant, experts caution against several risks. Saumitra Bhaduri, a professor at the Madras School of Economics, highlights the danger of 'artificial wisdom'—the societal misconception that AI generates knowledge when it merely generates statistically probable text based on its training data. This creates a systemic risk where critical decisions could be influenced by intelligence that no one is qualified to verify. Another major concern is the concentration of economic and political power in the hands of a few corporations that control frontier AI models. Critics argue that without a robust governance architecture, including a new legal framework like the proposed Digital India Act to establish clear liability for AI-induced harm, the widespread deployment of AI could lead to manipulation and misinformation. Therefore, the push for AI adoption must be accompanied by a parallel effort to establish strong regulatory guardrails, potentially including sandboxing environments for high-risk AI applications, similar to provisions in the European Union's AI Act.
The Way Forward: Balancing Ambition with Governance
Proponents view this as a pivotal 'WhatsApp moment' for India, an opportunity to leapfrog developmental stages and achieve a sustained 'Bharat rate of growth' of 8% and beyond. The argument is that AI offers the same scale of transformational leverage that economic liberalisation did in 1991. Missing this window could mean falling behind in a technology set to redefine global economic and strategic power balances.
The likely trajectory, should this policy be adopted, is a phased, 24-month implementation. A proposed 'National AI Token Policy' would begin with pilot programs providing free research tokens to premier institutions like the IITs and IISc. This would be followed by launching an API sandbox for 500 startups and expanding to 100 universities, with the goal of making India a top-five global consumer of AI tokens and fostering over 10,000 AI-native startups within a few years.
The governance implications are profound. The strategy's success hinges not just on technological implementation but on creating a new social contract for the AI age. This involves massive investment in reskilling the workforce through programs like the Skill India Mission to adapt to cognitive automation. It also requires establishing clear legal liability for AI systems and building institutional capacity to distinguish between authentic and synthetic information. The ultimate challenge for India will be to pursue its ambition of becoming a leading AI power while simultaneously developing the institutional wisdom to govern it. How India navigates this dual challenge will not only shape its own economic future but also offer a potential model for democratic governance in an era of intelligent machines.