Representative image
Credit: Reuters Photo
As details about the budget allocation for the IndiaAI mission emerge, this is an opportune moment to reassess its objectives for fostering India’s AI ecosystem and to evaluate how the government can achieve these goals. The authors argue that the government should promote open-source initiatives, adopt funding mechanisms that enable the market to evaluate value creation and innovation, and fund research to understand AI risks in an Indian context.
A recent response to a question in the Lok Sabha outlined the allocation of the budgeted Rs 10,372 crore across the seven pillars of the IndiaAI mission. A significant share, around 44%, is directed toward building domestic computing capacity. An innovation centre and startup financing each account for 19% of the budget, while the remainder is distributed among future skilling, an application development initiative, a datasets platform, and a safe and trusted AI initiative.
Computing capacity pillar
Investments in computing infrastructure are among the most significant expenses in AI development. For instance, Meta reported spending $8.5 billion in the second quarter of this year alone on AI-related computing infrastructure. In comparison, the IndiaAI’s planned investment over a five-year period are modest. Big tech firms are estimated to have spent $189 billion globally in 2024 and are projected to spend an additional $500 billion over the next three years on capital expenditure. Given this scale, the IndiaAI mission must carefully consider where and why it allocates funds for computing and AI research.
The request for empanelment to provide computing services for the IndiaAI mission received 19 bids, including Jio, Tata Communications, and Yotta. IndiaAI will decide the authorised end users from academia, MSMEs, startups, the research community, government entities, and public sector agencies. These users are to receive services from the empanelled vendors at prices established during the selection process.
This process is unlikely to be efficient for price discovery or offering competitive computing services. In addition, the process may create a closed loop of providers who are no longer required to compete with all competitors. This set up could also hinder newer and smaller players from entering the market. While the compute pillar provides a boost to the domestic cloud computing industry, the authors argue that the mission could be more effective by providing compute credits directly to selected users. These credits could then be used to services from any vendor, encouraging competition and innovation.
Innovation centre and startup funding pillars
The second largest areas of expenditure are the innovation centre and startup funding.
The innovation centre aims to develop foundational models, including indigenous large multimodal models and domain-specific foundational models, with applications in governance, healthcare, agriculture, sustainability, and manufacturing. The startup funding initiative seeks to accelerate the growth of deep-tech AI startups.
These initiatives raise a few critical questions. A narrow focus on developing large language models (LLMs) risks sidelining other promising areas of AI research. AI encompasses far more than LLMs, and a broader approach could yield more diverse innovations. Further, the idea of “indigenous models” needs clarification, especially in an era where advanced models require massive computational resources -- of 500 billion to a trillion parameter large models trained on 100,000 GPU clusters-- and where excellent open-source models such as those from Meta and Mistral already exist.
While the focus on innovation centres and startup funding is timely and important, the government should promote open-source AI models. Rather than funding a single innovation centre or specific startups, the government should look to promote open-source AI by creating a grant fund with 10x support on private funding raised. This approach shifts the responsibility of identifying high-potential projects to the market, which is better equipped for this task. Furthermore, the mission should place greater emphasis on the datasets pillar by creating India-specific open datasets across various domains. These datasets could then be leveraged by firms to develop innovative, localised AI solutions.
In the race to innovate and disrupt, the market often lacks incentives to thoroughly analyse the risks associated the AI. This highlights the need for State-funded AI safety institutes dedicated to understanding and evaluating these risks. The UK, US, and EU have already established such institutions.
(The writers are researchers with the High-Tech Geopolitics programme at the Takshashila Institution)