Overview
- On June 28–29, brokerage reports from Bernstein and Jefferies publicly flagged an urgent choice for India between staying an applications-first market or investing in homegrown foundational language models.
- Jefferies highlighted China’s GLM-5.2 as an example of a low-cost, enterprise-grade model that it says can deliver near-leading performance at about one-quarter the per-token cost of top US models.
- Jefferies also cited OpenRouter data showing Chinese models processed about 21.37 trillion tokens in the week ending June 21 compared with 5.76 trillion for leading US models, indicating rapid adoption of cheaper Chinese stacks.
- Bernstein says India’s services-led tech model and lack of massive proprietary consumer datasets leave it structurally disadvantaged for training frontier LLMs and notes the country’s AI Mission currently spreads funding across many priorities rather than concentrating on foundational models.
- Analysts recommend practical steps such as building sovereign, domain-specific LLMs for healthcare, finance and defence, incentivising domestic foundational-model work, or requiring India-hosted stacks, and they say the key near-term thing to watch is whether policymakers commit to large-scale sovereign investments, which have not yet been reported.