Sarvam AI has made a breakthrough in the tech sector of India by raising $234 million in the first close of its Series B financing. This recent fundraise brings the valuation of the company after the fund to $1.5 billion, making it India’s 130th unicorn. In addition to this achievement in fundraising, Sarvam AI has unveiled its future technological plan that is focused on creating a 1-trillion-parameter foundation model in India.

Capital infusion and token sovereignty

Established in 2023, the startup is rapidly becoming the backbone of the country’s push towards developing its own indigenous generative AI capabilities. This is a step towards bringing India among global artificial intelligence leaders by developing a world-class artificial intelligence model for complex enterprise and research applications, unveiled at the developer conference of its flagship platform Epoch 2026 in Bengaluru.

The investment round valued at $234 million demonstrates institutional confidence in Sarvam AI’s vision of sovereign artificial intelligence. The round was led by venture capital firm Bessemer Venture Partners and IT services major HCLTech, alongside continued investors such as Khosla Ventures and Peak XV Partners. 

In the financing, HCLTech also obtained 10.5% ownership of Sarvam AI for ₹14.27 billion (approximately $150.7 million). Through this strategic partnership, HCLTech will help to co-create specialized language models and customized artificial intelligence solutions for its enterprise customer base across the globe, and ensure that government-based and regulated companies can rapidly deploy sovereign solutions. 

The capital will directly support the heavy computational framework, data curation, and research power necessary for building future models within India. Sarvam AI has been identified as the leading startup in the Government of India’s national AI technology mission to develop India’s first sovereign large language model, with an allocation of ₹10,370 crore. 

This initiative connects Sarvam AI to a compute pool of graphics processing units, which include 4086 NVIDIA H100 GPUs, providing enough computational power to support large-scale training runs. The two co-founders of the company, Pratyush Kumar and Vivek Raghavan, emphasized that this infrastructure will facilitate the idea of token sovereignty. 

Token sovereignty will ensure that AI queries, sensitive information, and computations made in India are processed by cloud computing in India itself rather than being outsourced. This mechanism would guarantee privacy, compliance with regulations, and autonomy in technology for Indian organizations.

Ecosystem expansion and 1-trillion-parameter frontier model

Sarvam AI announced that its engineering team is currently training a 1-trillion-plus-parameter foundation model from the ground up at the Epoch 2026 conference. Co-founder Pratyush Kumar announced an aggressive model launch timeline of 6 months. 

While the classical conversational chatbot is focused mainly on text generation, this 1-trillion-parameter architecture is trained to tackle highly technical tasks, such as multi-step research reasoning, scientific simulation, cybersecurity analysis, and complex computer coding. Through building the model from scratch using a range of Indian and international datasets, Sarvam AI aims to demonstrate that a local team is capable of developing competitive frontier technology without the assistance of fine-tuned external models.

In addition to its flagship model roadmap, Sarvam AI announced 12 new courses to bring AI to the masses and create budget-friendly options across India. Key features are Bulbul V4, a more expressive text-to-speech engine for multiple regional languages; Sarvam Vision 2.0, a powerful optical character recognition tool for document understanding; and the Epoch Builder Edition platform for enterprise model fine-tuning. 

The company also released open-source versions of its Sarvam 30B and Sarvam 105B mixture-of-experts models. At the core of Sarvam AI’s business model is cost efficiency, with domestic inference costing approximately five times less than its international counterparts. 

While rival models such as GPT-5.4 Mini and Gemini 3.5 Flash cost $4.50 and $9.00 per million blended tokens, respectively, Sarvam AI serves its 105B model at approximately $0.80 per million blended tokens. The company also expanded into the United States by opening an office in San Francisco and hiring Devendra Chaplot, a former researcher at xAI and Mistral AI, as an advisor for its global talent sourcing and strategic outreach efforts.

Conclusion

The fundraising of $234 million and the guarded declaration of a 1 trillion parameter model is one important milestone in India’s technical evolution. The financial backing from founders such as HCLTech, along with government-backed compute power through the IndiaAI Mission, is enabling Sarvam AI to build an entirely indigenous value chain.

The startup is building a vertical of its own with low-cost inference hosting, support for Indian languages, and frontier technical capabilities that go head-to-head with the rest of the world for establishing a self-reliant, competitive artificial intelligence.