Sarvam AI's Epoch Builder Edition launch highlights India's growing ambitions in sovereign artificial intelligence.
BENGALURU: Indian artificial intelligence startup Sarvam AI on Thursday made a series of major announcements aimed at strengthening India’s position in the global AI race. At its Epoch Builder Edition event, the company launched a new platform that allows developers, enterprises and government organisations to build custom large language models (LLMs) tailored for Indian languages and use cases. The company also announced plans to develop what it described as India’s first trillion-parameter AI model, marking one of the country’s most ambitious AI initiatives yet. The announcements align with the Government of India’s broader push to develop sovereign AI capabilities under the IndiaAI Mission.
The Bengaluru-based startup, founded by Pratyush Kumar and Vivek Raghavan, said the new platform is designed to help organisations build, fine-tune and deploy AI models without depending entirely on foreign AI technologies.
Epoch Builder Edition launched
The biggest product announcement was the launch of Epoch Builder Edition, a full-stack AI platform aimed at enterprises, developers, research institutions and government agencies.
According to the company, the platform enables users to:
- Build custom large language models
- Fine-tune models using proprietary datasets
- Deploy AI systems on cloud, hybrid or on-premises infrastructure
- Access GPU training infrastructure
- Use curated Indian datasets
- Train multilingual AI applications for Indian use cases
- Use reinforcement learning, benchmarking and safety testing tools
- Access APIs for model deployment and integration
The platform has been designed specifically for organisations working with Indian languages and domain-specific datasets, including banking, governance, healthcare, education and legal services.
Sarvam AI said the platform will enter private preview in August 2026, with broader commercial availability planned for the fourth quarter of 2026.
Built for Indian languages
Sarvam AI said the platform includes multilingual foundation models with 7-billion and 70-billion parameters, trained on nearly 2 trillion tokens.
These models support more than 10 Indian languages, including Hindi, Tamil, Kannada, Bengali and Marathi. The company also highlighted support for code-mixed languages such as Hinglish and Kannanglish, which are commonly used by Indian internet users but are often not handled well by global AI models.
Company reveals 100-billion-plus parameter model
During the event, Sarvam AI co-founder Pratyush Kumar said the company has pre-trained from scratch a 100-billion-plus-parameter foundation model in India over the past year using around 4,000 NVIDIA Hopper GPUs.
“Over the last year, we pre-trained from scratch in India on 4,000 Hopper Series GPUs a 100-billion-plus-parameter model,” Kumar said during the announcement.
Training a model of this scale requires enormous computing resources, high-speed networking infrastructure and vast multilingual datasets. The company said the milestone demonstrates that frontier AI models can be developed within India.
Plans for a trillion-parameter AI model
Perhaps the biggest announcement of the day was Sarvam AI’s plan to build what it described as India’s first trillion-parameter foundational AI model.
If completed, the project would place Sarvam AI among a small group of companies worldwide pursuing trillion-parameter AI systems. The company said the model will focus on advanced reasoning, multilingual capabilities, coding and enterprise applications while supporting India’s goal of developing sovereign AI infrastructure.
Although Sarvam did not disclose a launch timeline, executives described the project as the company’s next major research milestone.
Focus on enterprise AI
Sarvam AI said Epoch Builder Edition has been designed with enterprise customers in mind.
The platform offers:
- On-premises, hybrid and cloud deployment
- GPU clusters for training
- Reinforcement learning and instruction-tuning tools
- Curated domain-specific datasets
- Safety evaluation, red-teaming and benchmarking tools
- API access for developers
These features are intended to help banks, government departments, research institutions and large enterprises build AI systems using their own data while maintaining privacy and regulatory compliance.
Strengthening India’s AI ecosystem
The announcements align with India’s broader efforts to build indigenous AI capabilities under the IndiaAI Mission.
Sarvam AI also announced collaborations with academic institutions and state governments to expand AI adoption across education and public services.
During the event, the company reiterated its vision of reducing India’s dependence on foreign AI platforms by building technology that understands Indian languages, culture and enterprise requirements.
Global ambitions
The announcement reflects Sarvam AI’s ambition to build frontier AI systems alongside global companies such as OpenAI, Anthropic and Google DeepMind. However, those companies currently maintain significant advantages in frontier research, reasoning capabilities, multimodal AI, developer ecosystems and global computing infrastructure.
Sarvam AI is positioning itself by focusing on multilingual Indian AI, sovereign infrastructure and enterprise deployments tailored to local requirements.
Looking ahead
With the launch of Epoch Builder Edition and plans for a trillion-parameter AI model, Sarvam AI has laid out one of the most ambitious AI roadmaps announced by an Indian startup. If the company delivers on its plans, it could play a key role in India’s efforts to build globally competitive AI technology while strengthening support for Indian languages and enterprise applications.
About the Author
