Enablers of AI innovation in India2 India can harness the AI opportunity through innovation by leveraging its core strengths outlined in Section 1.1. However, a few enablers need to be put in place for this to happen. The World Economic Forum held consultations with Indian AI start-ups as well as industry leaders and policy-makers in workshops and one-on- one interviews to identify the basic enablers of AI innovation in India. These consultations focused particularly on priority sectors such as agriculture, healthcare, education and MSMEs. The findings can be broadly grouped into five categories. Data enablers: AI start-ups often lack access to the local, AI-ready datasets that are crucial for building models and applications suited to the Indian context. The availability of recent and real-time data enhances efficiency and accuracy in model training and validation. The addition of vernacular-language datasets enhances local relevance, while clear governance frameworks on privacy and data sharing further smooth the path by increasing compliance and trust. Ease of access to real- world (anonymized) data would also accelerate the development of these datasets. Data quality and local data (not translated content) through good governance practices would eliminate bias due to Western ethnicities and create models that suit the local population. Together, this set is referred to as data enablers: availability, access, data quality, local content, data governance, data protection. Computational infrastructure: Access to affordable, scalable compute capability enables the training of sophisticated AI models. The IndiaAI Mission initiative’s initial aim of providing more than 18,000 graphics processing units (GPUs)13 at subsidized rates to start-ups has been significant in addressing this challenge. In fact, this was increased to 34,000 GPUs as of June 2025 and several AI start-ups have been onboarded. Further, more kinds of publicly available and accessible compute infrastructure, such as DPI-enabled clusters, could enable and democratize AI experimentation. Model availability and contextual representation: The reliance on AI models trained on non-Indian data significantly increases costs and latency, with many lacking robust vernacular capabilities, restricting their applicability and effectiveness in addressing local market needs. Most are trained primarily on English-language data14 (for example, OpenAI’s GPT-3 uses ~92.7% English-language sources), limiting their performance and fairness in Indian contexts and introducing bias when deployed locally. Focusing on models trained on Indian content/languages is a great enabler.Data access is a challenge across sectors – start- ups often lack high-quality, real-time and vernacular datasets, and data-sharing mechanisms among the systems of government, industry, academia and start-ups remain limited. Fabian Bigar, Secretary-General, Ministry of Digital, Government of MalaysiaAI sandboxes can act as launchpads for entirely new use cases by enabling developers, start-ups and researchers to safely test and refine cutting-edge solutions. From precision agriculture and AI-assisted diagnostics to intelligent public service delivery, these environments can accelerate the transition from promising ideas to scalable applications with real impact across India. Shankar Maruwada, Co-Founder and Chief Executive Officer, EkStep Foundation Latency remains a significant challenge. Establishing an AI sandbox environment in India – with either indigenous models or direct integration of global models – can substantially reduce latency and enhance performance. Keshav Reddy, Founder and Chief Executive Officer, Equal Shaping the AI Sandbox Ecosystem for the Intelligent Age 10

Shaping the AI Sandbox Ecosystem for the Intelligent Age 2025