Google's India AI push: Soket AI and Gnani.ai to develop foundation LLM with Gemma

In a bid to deepen its AI presence in India, Google announced its partnerships with two more IndiaAI Mission startups aside from Sarvam. At the Google I/O Connect India 2025 in Bengaluru on Wednesday, the tech giant revealed that Soket.AI and Gnani.AI will develop next-generation foundational AI models based on Google’s open-source Gemma family.

“Indian developers are literally writing the next chapter of India’s success story, using AI capabilities to build real-world applications that are reaching millions of businesses and people across India and the world. We have been inspired by their solutions, and proud that many have chosen our AI, including Gemini, Gemma and GenMedia models like Veo and Imagen, to build their innovations,” said Manish Gupta, Senior Director for India and APAC at Google DeepMind.

Gemma is a lightweight open language model that has been built from the same technology as Google’s Gemini models. This extends model access to developers and researchers for a wider range of applications, making their deployment possible even on laptops and mobile devices.

Sarvam has used Gemma to build Sarvam-Translate for long-form text translation. Meanwhile, Soket.AI is developing a 120-billion-parameter open-source model optimised for India’s linguistic diversity, as part of Project EKA. Gnani.AI is building a multilingual, voice-first AI system focused on emotional intelligence and enterprise use cases.

Google also reported that the Android and Google Play ecosystem generated ₹4 lakh crore in revenue in India in 2024 and supported 35 lakh jobs. Around 85% of Indian users access public digital services via Android, and 69% reported their first AI experience came through apps on Android.

Google also shared that it has brought onshore processing of Gemini 2.5 Flash to India, enabling low-latency, high-speed AI services that are critical for regulated sectors such as healthcare, finance, and public services.

This means Google has set up local data processing for its AI model Gemini 2.5 Flash within India. So instead of sending data to servers abroad, the AI can now process information within the country, making the entire process faster and more reliable. The move is also aimed at meeting data residency requirements and improving responsiveness for Indian developers.

Additionally, Google introduced new agentic tools in its Firebase Studio, an agentic cloud-based development environment, including Optimised AI templates and backend service integrations, to help Indian developers build and deploy full-stack AI applications faster.

To support India’s growing game development ecosystem, Google has launched the ‘Google Play x Unity Game Developer Training’ program. Developed with Unity and the Game Developer Association of India (GDAI), the program offers over 30 hours of training across game development, art, and programming tracks. The program will initially support 500 developers in partnership with Tamil Nadu, Andhra Pradesh, and GDAI.