AI unicorn Sarvam announced plans to build a trillion-plus parameter AI model in India and launched Sarvam Inference, a domestically hosted inference platform, at its Epoch 2026 developer conference. The Bengaluru-based startup aims to create an end-to-end AI stack covering models, infrastructure, and enterprise software, marking its biggest product roadmap yet, according to inc42.com.
Sarvam cofounder Pratyush Kumar said the new model is being built from scratch to compete in areas like coding, cybersecurity, simulation, and science. The company did not specify a launch timeline. Alongside the model announcement, Sarvam introduced Sarvam Inference, which currently supports its own 105 billion parameter model and open-source models such as GLM 5.2 and Gemma 4, all hosted on infrastructure located within India. The startup also opened a San Francisco office and appointed Devendra Singh Chaplot, formerly of Mistral AI and Thinking Machines Lab, as an advisor.
The move to develop a trillion-parameter model domestically positions Sarvam among global AI developers focusing on large-scale models. Hosting inference infrastructure within India addresses data sovereignty and latency concerns for local enterprises. Sarvam’s approach mirrors trends by other AI firms building competitive models and platforms to serve diverse enterprise needs in AI-driven coding, cybersecurity, and scientific applications, as noted by inc42.com.
Sarvam’s Epoch 2026 event marked a significant milestone with the launch of its India-hosted inference platform and the announcement of its trillion-parameter model project. The startup’s expansion includes a new San Francisco office and strategic advisory appointments, signaling its intent to scale both domestically and internationally, according to inc42.com.