Indian AI Startup Sarvam Targets Feature Phones and Edge Devices with Lightweight Models
Sarvam, an Indian AI startup, has unveiled two new large language models designed to run efficiently on low-power devices like feature phones and smart glasses. By optimizing models to under 100MB and enabling offline functionality, the company aims to democratize AI access across India’s vast unconnected populations.
Indian artificial intelligence startup Sarvam is revolutionizing AI accessibility by developing lightweight, edge-optimized language models capable of running on feature phones, automotive systems, and smart glasses—devices traditionally considered too underpowered for advanced AI. The company recently unveiled two new large language models, Sarvam-30B and Sarvam-105B, designed not for cloud-dependent processing but for on-device inference with minimal memory footprint, according to The Economic Times.
Unlike conventional AI models that require high-end GPUs and constant internet connectivity, Sarvam’s innovations leverage advanced model compression and quantization techniques to reduce its largest model to under 100MB, enabling deployment on devices with as little as 2GB of RAM and standard ARM processors. This breakthrough allows users in rural and semi-urban India, where smartphone penetration is still evolving and data costs remain prohibitive, to access AI-driven services such as voice assistants, real-time translation, and local-language content generation—all without an internet connection.
"The goal isn’t just to make AI smaller, but to make it relevant," said a company spokesperson, as reported in Moneycontrol. Sarvam’s models are specifically fine-tuned for Indian linguistic diversity, supporting 12 regional languages including Hindi, Tamil, Telugu, Bengali, and Marathi. This linguistic customization addresses a critical gap in the global AI landscape, where most models are trained predominantly on English and lack contextual understanding of India’s cultural and linguistic nuances.
By focusing on edge computing, Sarvam sidesteps infrastructure challenges that have hindered AI adoption in India’s underserved regions. Feature phones, which still account for over 30% of mobile device usage in the country, can now run sophisticated AI applications thanks to Sarvam’s architecture. The company has already partnered with two Indian automotive manufacturers to integrate its models into in-car infotainment systems, enabling voice-controlled navigation and local-language commands without relying on cloud servers. Similarly, pilot programs with wearable tech startups are testing AI-powered smart glasses that offer real-time visual translation for tourists and small business owners navigating multilingual environments.
The implications extend beyond convenience. In healthcare, Sarvam’s offline models could empower ASHA workers to use AI-assisted diagnostics in remote villages. In education, students with limited connectivity can interact with AI tutors that understand regional dialects. The technology also enhances privacy, as sensitive user data never leaves the device—a crucial advantage in a country with evolving digital privacy regulations.
While global tech giants focus on scaling up AI with massive data centers, Sarvam’s approach represents a paradigm shift: AI for the many, not just the connected. The startup’s strategy aligns with India’s Digital Public Infrastructure initiatives and could serve as a blueprint for other emerging economies grappling with similar connectivity and affordability barriers.
Industry analysts note that Sarvam’s success hinges on continued optimization and ecosystem partnerships. Yet with backing from prominent Indian venture funds and growing interest from global hardware makers, the company is poised to become a key player in the next wave of inclusive AI. As smartphone manufacturers increasingly prioritize cost-effective models, Sarvam’s lightweight AI could become a standard feature—not a luxury—in the next generation of Indian tech devices.
