India, March 27 -- India is pouring its energy into catching up with the US and China on building foundational Artificial Intelligence (AI) models. Government officials are discussing billion-dollar investments in data centres and computing infrastructure. Academics are calling for sovereign large language models (LLMs). Industry groups are lobbying for more public-private collaborations in model training. But all this focus is on yesterday's problem.

The world doesn't need more foundational models. It already has dozens. Open-source alternatives like DeepSeek, LLaMA, Qwen, and Mistral are freely available - and improving at a breakneck pace. DeepSeek recently released a model that rivals GPT-4 in reasoning benchmarks. Qwen, developed by...