Can India carve out a place in the AI race without owning frontier models or large-scale compute? That question anchored the debut episode of Bloomberg's new podcast "Emerging," where experts weighed India's options.
Emerging · India's AI Strategy
Can India Compete in AI Without Frontier Models or Large-Scale Compute?
The world's most populous nation and fifth-largest economy is betting on talent, multilingual data and application-layer innovation — not chip ownership or massive compute — to carve out its place in the global AI race.
~3%
of global data-center capacity held by India
100%
of advanced AI chips are imported
22+
scheduled languages powering multilingual data
1,000+
H100 GPUs on Indian infrastructure for training
Sarvam AI — Indic Open-Source Models (Apache 2.0)
Mixture-of-Experts models trained on India's own infrastructure
~30B
Sarvam 30B Real-time conversational AI · Samvaad
~105B+
Sarvam 105B Complex reasoning · agentic · Indus
Column heights scaled to parameter count (~30B vs ~105B)
Sarvam Vision OCR — Reported Scores (English subset)
Self-reported; the 105B's Math500, MMLU and LiveCodeBench scores lack independent verification.
STRENGTHS & PRAISE
Strong on Indic, Hinglish & code-mixed tasks
On-prem / local GPU deployment for data privacy
Open-sourced + API access for enterprise customization
Deep engineering talent & UPI transactional data
GAPS & CRITIQUE
Some models seen as fine-tuned, not "from scratch"
Rated "mediocre" vs DeepSeek, Qwen, GLM by some
High training/inference costs, scale & data gaps
Reliance on foreign architectures & imported hardware
The Crux
Rather than full-stack self-sufficiency, India's bet is on differentiating through talent , multilingual data and the application layer — a race intensified since the rise of DeepSeek.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…