India faces a strategic decision over whether to rely primarily on proprietary AI models, open-weight models, or a combination of both. The debate has gained urgency as Chinese open-weight models have demonstrated capabilities approaching leading US closed models while being developed and deployed at significantly lower costs. For India, however, the issue goes beyond model performance—it is closely connected to digital sovereignty, national security, and long-term economic competitiveness.
Some Indian technology leaders argue that India needs to develop capabilities across the entire AI stack, rather than becoming dependent on foreign companies for critical AI infrastructure and intelligence. Sarvam co-founder Vivek Raghavan warns that without a full-stack Indian alternative, the country could become overly dependent on external technology. Open-weight models are particularly attractive because once their model weights are released, they can be adapted and deployed without depending entirely on the original provider.
At the same time, the article argues that India does not necessarily need to choose one side. Industry leaders such as Cognizant CEO Ravi Kumar S and Google DeepMind's Manish Gupta believe both approaches have a role. Closed frontier models can provide cutting-edge capabilities for complex applications, while open-weight models can offer lower-cost customization for Indian languages, industries, and enterprise workflows. This flexibility could be especially valuable for startups and organizations operating with limited resources.
The broader recommendation is that India should focus less on simply building the world's largest foundation model and more on creating useful, specialized AI products and capabilities. Open models can lower the cost of experimentation and adaptation, while proprietary models can be used when advanced intelligence is essential. Ultimately, India's advantage may come from being able to choose the right model for each task rather than committing exclusively to open or closed AI.
