India’s Bold Push to Build Its Own AI Future Amid Global Competition

India’s AI Landscape: A Tale of Two Startups

In Bengaluru, the launch of DeepSeek, a Chinese language model rivaling Western benchmarks but created with far less capital, sparked mixed feelings among Indian AI entrepreneurs. Adithya Kolavi, founder of CognitiveLab, saw it as proof that disruption could happen with limited resources. Meanwhile, Abhishek Upperwal, creator of India’s early foundation model Pragna-1B, felt frustration over underfunding that stalled his multilingual AI project designed to tackle India’s unique linguistic challenges.

Challenges of Indian AI Development

India’s AI ecosystem is constrained by historic underinvestment in research and development, scoring just 0.65% of GDP on R&D compared to China and the US. The country’s linguistic diversity—with 22 official languages and hundreds of dialects—adds complexity to training language models, as Indian languages constitute less than 1% of online content and suffer from poor tokenizer support. This makes building effective multilingual models much harder than in English-dominant regions.

Government Initiatives and Industry Response

Following DeepSeek’s launch, India’s Ministry of Electronics and Information Technology (MeitY) quickly mobilized resources by partnering with cloud providers to access nearly 19,000 GPUs for AI research. This sparked a wave of proposals and led to plans for six large-scale models and 18 AI applications targeting key sectors by the end of 2025. Notably, Sarvam AI received funding to develop a 70-billion-parameter model optimized for Indian languages.

Innovations in Indian Language AI

Startups like Sarvam AI and Soket AI Labs are pioneering solutions for Indian languages. Sarvam developed OpenHathi-Hi-v0.1, a large open-source Hindi model trained on 40 billion tokens. Upperwal introduced “balanced tokenization” to enable smaller models to perform like much larger ones by addressing the agglutinative and script complexities of Indian languages. Krutrim-2 aims to support multimodal and voice-first applications across 22 Indian languages.

Funding, Infrastructure, and Ecosystem Growth

India’s AI ambitions are backed by a $1.25 billion IndiaAI Mission, deploying over 18,000 GPUs to startups and funding foundational model development. Despite this, debates continue over open-source versus closed models, with concerns about transparency and accessibility. Access to compute remains a bottleneck, but government programs and lower data center costs offer advantages.

Strategic Focus and Future Direction

India’s strategy emphasizes building sovereign AI adapted to local linguistic and cultural realities, rather than competing head-on with giants like OpenAI. Influential voices advocate focusing on application layers, talent, and niche strengths. The evolving collaboration between government and startups signals a determined effort to overcome longstanding innovation gaps and carve out a unique role in the global AI landscape.