AI asked: “What came first: use case or foundational model?”

Big tech companies like Google disagree with technologist Nandan Nilekani’s advice that India’s AI ecosystem should prioritise building use cases over foundational models. Mr Nilekani’s reasoning is sagacious. India should focus on spending on computing, infrastructure and AI clouds. Building foundational models on the lines of GPT-5, Claude, Llama, or Gemini requires enormous compute, GPUs, datasets, talent, and retraining worth ₹5000 – ₹10,000 crore. The Indian government does not yet possess such funds or facilities. It is dependent on big tech.

By contrast, millions of applications exist because they are relatively easy to build and monetise. Most do not require huge capital or cutting-edge research. Across digital ecosystems, economic value is captured at the application layer, not the foundational layer. AI applications in law, medicine, languages, governance, agriculture, education, and other fields require domain-specific data, product design, workflow integration, and fine-tuning existing models rather than expensive frontier research. 

Use cases are favourable because of three structural advantages: faster innovations, local relevance, and lower capital. For a country like India, these advantages are significant. A realistic AI ecosystem evolves in three increasingly complex and resource-intensive phases: i) Booms (applications, startups, etc.), ii) Tools (databases, data pipelines, evaluation tools), iii) Foundations (specialised, languages, domain-specific). 

If the goal is building a real AI ecosystem, we must prioritise use case-based application development first. This approach ensures that value and innovation are spread broadly, instead of being concentrated in a few cities as with foundational models. It is particularly important for Bihar, which has suffered from geographical disadvantages and policies like Freight Equalisation. If the Indian state policy prioritises expensive foundational models over applications, regions like Bihar risk remaining captive users of technology.