Sarvam AI | All about the Bengaluru startup’s plans for a trillion-parameter model
Bengaluru-based startup Sarvam AI announced, at its developer conference "Epoch 2026," plans to build a trillion-parameter frontier AI model developed entirely in India.
The company also launched a platform allowing developers, enterprises, and government bodies to build custom large language models (LLMs) tailored to Indian languages and use cases, with inference workloads to be hosted within India.
The new models are expected to be released within roughly six months of the announcement, targeting capabilities in coding, cybersecurity, simulation, and scientific reasoning.
The initiative is positioned alongside the Union Government's IndiaAI Mission, which is building sovereign compute and foundation-model capacity in India.
IndiaAI Mission and Sovereign AI Capacity
The IndiaAI Mission is a Union Government initiative approved by the Cabinet in March 2024 with an outlay of ₹10,372 crore over five years, aimed at building a comprehensive AI ecosystem in India — from compute infrastructure to foundation models to safe/trusted AI governance. It is implemented through the IndiaAI Independent Business Division under the Digital India Corporation, functioning under the Ministry of Electronics and Information Technology (MeitY).
Key Details
- Total outlay: ₹10,372 crore, approved by the Union Cabinet in March 2024.
- Seven pillars: Compute Capacity, Foundation Models, AIKosh (data platform), Application Development, FutureSkills, Startup Financing, and Safe & Trusted AI.
- Compute pillar: public-private partnership model providing subsidized GPU access; over 18,000 GPUs empanelled, with tens of lakhs of GPU-hours sanctioned to startups and researchers building indigenous models.
- FY2025-26 budget allocation for the mission: ₹2,000 crore, a sharp increase over FY2024-25 revised estimates.
- Over 85% of compute-pillar subsidies disbursed have gone toward development of indigenous generative AI (LLM/SLM) models — a mix of large language models and small language models selected under the mission.
Sarvam AI's trillion-parameter model is among the flagship efforts aligned with the IndiaAI Mission's Foundation Models pillar — an attempt to reduce India's dependence on foreign frontier AI models (e.g., from US or Chinese labs) and build "sovereign AI" capability, including data residency (inference hosted within India) and language coverage for Indic languages.
Data Sovereignty and Foundation Model Parameters
"Parameters" in a neural network are the learnable numerical weights that determine a model's behavior; larger parameter counts generally (though not always) correlate with greater model capability, at the cost of proportionally larger compute, energy, and data requirements for training. "Sovereign AI" refers to a nation's capacity to independently develop, host, and control AI models and the compute/data infrastructure underlying them, reducing reliance on foreign technology providers for a strategically important capability.
Key Details
- Frontier global models (e.g., from OpenAI, Google, Anthropic) are estimated in the range of hundreds of billions to over a trillion parameters, though most frontier labs do not officially disclose exact parameter counts.
- Building and training trillion-parameter-scale models requires very large compute clusters (thousands of high-end GPUs), significant capital, and large curated datasets — a major technical and financial barrier for any single country or company.
- Data sovereignty concerns: keeping AI training data and inference workloads within national borders addresses privacy, security, and strategic-autonomy concerns tied to relying on foreign-hosted AI infrastructure.
Sarvam AI's claim of building a trillion-parameter model entirely from scratch in India, with in-country inference hosting, is directly a data-sovereignty and strategic-autonomy play — situating the company as a private-sector complement to the government's IndiaAI Mission goal of sovereign AI capability.
- IndiaAI Mission: ₹10,372 crore outlay, approved March 2024, under MeitY.
- FY2025-26 IndiaAI Mission budget allocation: ₹2,000 crore.
- IndiaAI Mission compute infrastructure: 18,000+ GPUs via public-private partnership.
- Sarvam AI's model timeline: expected to go live within about six months of the August 2026 announcement.
- Sarvam AI's stated target domains: coding, cybersecurity, simulation, and scientific reasoning.