India's AI market is already crowded with global players, but Sarvam AI is betting that it can carve out a space of its own by building models and infrastructure specifically for the country. The question is whether localisation can give Sarvam an edge over companies with far greater capital, computing power and distribution.
What Sarvam is building
Founded in 2023 by Vivek Raghavan and Pratyush Kumar, Sarvam describes itself as a full-stack sovereign AI platform, spanning large language models, speech recognition, translation, document processing, conversational agents and coding across 22 Indian languages.
The company raised $234 million in its first Series B close, led by HCLTech, followed by a $75 million Nvidia extension, taking the round to about $309 million and its valuation to $1.51 billion. Total funding since 2023 is close to $349 million.
At its July developer conference, Epoch, Sarvam launched Sarvam Inference, hosted in India, and said it handles more than two million conversations and 10 million API calls a day. Its document systems have digitised more than 35 million pages, while its speech models transcribe over 500,000 hours of audio every month.
Sarvam has access to about 2,000 Nvidia Blackwell chips and aims to scale this to 10,000, alongside a $1.5 billion HCLTech data centre partnership in Odisha. It released Sarvam 30B and 105B under an open-source licence and is building a model with more than one trillion parameters.
Its technology powers Aadhaar voice services in 10 languages, while SBI Life Insurance is deploying Samvaad across 80 million customers and 350,000 distributors.
The company's revenue rose from Rs 1.5 crore in FY25 to Rs 45.1 crore in FY26.
What Big Tech has
The difference in distribution is large. Recent data from Sensor Tower showed ChatGPT had 330 million monthly active users in India by May 2026, up from 297 million in December, while Google's Gemini had 229 million and Anthropic's Claude had 72.3 million, up from 13.3 million in December.
In the April-to-June quarter, Claude's downloads in India rose 30-fold year on year to 8.5 million, while consumer spending on the app rose nearly 19-fold to about Rs 62.2 crore. ChatGPT's India spending in the same quarter was about Rs 112.8 crore, compared with Rs 20.4 crore for Grok and Rs 5.8 crore for Perplexity.
The global companies also bring established cloud, search, productivity, developer and consumer ecosystems.
Where Sarvam differs
Sarvam's strategy is to make localisation and deployment part of the product. Its models are trained from scratch in India, while its inference platform serves open-weight models from Indian infrastructure. The company also offers cloud, private-cloud and on-premise deployment.
That matters where organisations want control over sensitive data, local language support or deployment inside their own infrastructure.
Jibu Elias, responsible computing lead for India at the Mozilla Foundation, said during a Stimson Center discussion on US-India AI collaboration that open-weight models allow Indian teams to focus on "localization, multilingual systems, and domain-specific adaptions". He cited Sarvam as an example of a "middle path on sovereignty", where models can be hosted, fine-tuned and governed locally.
Economics is also part of Sarvam's pitch. Sarvam said its 105B model is priced at $0.80 per million tokens, against $4.50 for OpenAI's GPT-5.4 Mini and $9 for Google's Gemini 3.5 Flash.
India's AI market is expected to provide a substantial opportunity. Boston Consulting Group has projected India's AI market will more than triple to cross Rs 1,45,384 crore, or about $17 billion, by 2027, driven by enterprise technology investment and India's base of AI professionals.
But Sarvam's access to compute under the IndiaAI Mission is time-limited. Once the government allocation ends, the company will need to fund its own infrastructure.
Sarvam's advantage, therefore, is not consumer reach. It is localisation, domestic deployment and the ability to build AI around Indian-language and institutional use cases. Its harder task is turning those advantages into repeatable products and revenue while continuing to improve model performance.