Sarvam AI, India's IndiaAI Mission-backed startup, released two open-source MoE reasoning models — Sarvam-30B and Sarvam-105B — under the Apache 2.0 licence on Hugging Face and AIKosh, both trained from scratch on government compute. For builders, this means production-ready Indic-language model weights are now freely available to fine-tune or deploy, eliminating the need to rely on OpenAI or Google APIs for Indian-language use cases.
On 15 September 2025, the RBI issued the Reserve Bank of India (Regulation of Payment Aggregators) Directions 2025, consolidating and superseding all earlier PA/PG regulatory frameworks into a single operative document. Any team building a checkout, embedded payments, or merchant-facing collection flow now has one definitive compliance reference — non-compliance is no longer a grey-zone risk but a clear regulatory breach.
₹300 Cr
India's DPIIT issued Gazette Notification G.S.R. 108(E) on 4 February 2026, replacing the 2019 startup framework and creating a formal 'Deep Tech Startup' category with up to 20 years of recognised startup status and a ₹300 crore turnover ceiling for benefits — double the old 10-year window and 3x the revenue threshold. Founders in AI, semiconductors, biotech, or space building long-gestation products can now retain tax benefits, grants, and regulatory privileges far beyond what the old framework allowed.
42%
The IndiaAI Mission has onboarded over 38,000 GPUs on its compute portal offered at roughly ₹115–₹150 per GPU-hour, about 42% below commercial cloud rates, and has simultaneously opened applications for Cohort 2 of its Startups Global programme — a 4-month funded stint in Paris via Station F and HEC Paris for 10 selected Indian AI startups. Builders training Indic or vertical AI models can now slash their compute burn significantly, while the Paris cohort offers a direct route to global enterprise and investor networks.