Sarvam, the Indian AI startup, just cracked one of the industry's toughest problems - running sophisticated AI models on devices that most of the world actually uses. The company's new edge AI technology runs entirely offline on feature phones, cars, and smart glasses using models that take up only megabytes of space, a breakthrough that could bring AI capabilities to billions of users in emerging markets who've been left out of the generative AI revolution.
While tech giants chase ever-larger cloud-based models, Sarvam is zigging where everyone else zags. The Bengaluru-based startup revealed its edge AI strategy at the India AI Impact Summit 2026, announcing technology that could finally democratize AI access beyond smartphone-wielding urban elites.
The math is staggering. India still has over 300 million feature phone users, devices that cost as little as $15 and lack the processing power or constant connectivity that today's AI applications demand. Sarvam's edge models solve this by compressing AI capabilities into tiny packages that run entirely on-device, no internet required.
"We're not just shrinking models, we're rethinking how AI should work in markets where connectivity is expensive and unreliable," the company explained during its presentation. The technology works on existing processors found in basic phones, eliminating the need for specialized hardware upgrades that would price out most potential users.
The timing couldn't be more critical. While OpenAI and Google dominate headlines with increasingly powerful cloud models, the vast majority of the world's population can't access these tools. Slow internet speeds, data costs, and device limitations create an AI access gap that's widening as models get more sophisticated and resource-hungry.
Sarvam's approach flips this paradigm. By keeping models small enough to fit in megabytes rather than gigabytes, the startup enables AI features on the cheapest devices in circulation. The company's working with Qualcomm to optimize models for the chipmaker's entry-level processors, which power millions of budget smartphones and feature phones across Asia and Africa.












