India's AI startup Sarvam has announced plans to build a trillion-parameter foundation model while introducing a managed inference service aimed at helping developers and enterprises deploy AI applications more efficiently. The announcements, made during the company's Epoch conference in Bengaluru, underscore Sarvam's ambition to strengthen India's sovereign AI ecosystem by expanding both its model capabilities and AI infrastructure.
The proposed trillion-parameter model represents the next stage in Sarvam's frontier AI roadmap. While the company has not disclosed a timeline for its release, it said the model is intended to support more advanced reasoning, multilingual understanding and enterprise-grade AI applications. If completed, it would place Sarvam among a limited number of AI developers globally pursuing models at the trillion-parameter scale.
Alongside the model announcement, Sarvam introduced Sarvam Inference, a fully managed inference platform designed to simplify access to large language models. The service enables developers to run open source and proprietary AI models through APIs without having to manage GPU infrastructure. The company said the platform has been built to deliver lower latency, cost-efficient deployment and production-ready scalability for enterprise AI workloads.
The inference service supports Sarvam's own models as well as a growing ecosystem of open models, allowing enterprises to choose deployments that match their performance, compliance and cost requirements. The offering is expected to benefit startups, developers and organisations looking to integrate AI into customer support, software development, document processing and multilingual applications without investing in dedicated infrastructure.
Sarvam has increasingly positioned itself as a full-stack AI company with products spanning foundation models, speech AI, developer tools, cloud infrastructure and enterprise applications. Earlier this year, the company open sourced its 30-billion and 105-billion parameter reasoning models, both trained in India under the IndiaAI Mission. It also introduced products including the multilingual assistant Indus, conversational AI platform Samvaad and on-device AI initiative Sarvam Edge.
The latest announcements follow Sarvam's recent Series B funding round, in which the company secured the first close of a planned $300 million raise led by HCLTech and other investors. The capital is being used to expand research into next-generation foundation models, scale AI infrastructure and accelerate enterprise deployments across sectors including financial services, government and healthcare.
The move also aligns with India's broader efforts to build sovereign AI capabilities through locally developed models, domestic compute infrastructure and multilingual AI systems. As organisations increasingly seek alternatives that offer greater control over data residency, compliance and deployment, Indian AI companies are investing across the full technology stack rather than focusing solely on model development.
Industry observers view inference infrastructure as an increasingly important layer of the AI ecosystem, particularly as enterprises shift from experimenting with models to deploying production-scale AI applications. By combining foundation model research with managed inference services, Sarvam is seeking to address both AI development and operational deployment, strengthening its position in India's fast-growing enterprise AI market.