Gnani.ai launched Artha, a 30-billion-parameter AI stack, with the Vice President pitching "sovereign AI." The move positions Artha as a direct India-based alternative to reduce reliance on foreign AI infrastructure. Only 5 out of 20 enterprises have shown early interest in the stack, with no deployments yet.
How We Got Here
India's push for "sovereign AI" gained urgency due to rising costs and reliance concerns over US and Chinese AI infrastructure. The IndiaAI Mission, launched in 2024, has been backing domestic models like Sarvam AI with subsidised GPU compute.
The Numbers
- Artha integrates Evon v3.3, a 30-billion-parameter language model trained in 11 Indian languages, with the Plexus AI agent platform.
- Gnani.ai co-founder Ganesh Gopalan stated 5 of 20 enterprises at a Pune meeting showed interest, starting builds on Evon for underwriting or payments.
- Evon v3.3 model weights are available via Hugging Face under an Apache 2.0 licence, upon request.
- The Vice President's Secretariat frames Plexus as connecting AI intelligence to real-world work and institutions.
- Gnani.ai targets data control, AI scaling costs, and multi-Indian language operation as Artha's core problem-solving areas.
What Happens Next
🇮🇳 Why This Matters for India
For Indian deep tech founders building agentic AI, Artha promises a local stack to reduce compute costs and data sovereignty concerns, especially for BFSI clients in Mumbai and Bengaluru.
The Take
The "sovereign AI" narrative got a big government push, but Gnani.ai’s real battle isn't with foreign models; it's with converting early interest into widespread, deployed use cases. Without concrete customer traction and successful pilots beyond the Pune meeting, the political framing remains a narrative, not market validation.
Source:
MediaNama ↗