Indian AI firms Sarvam AI and BharatGen are offering foundation models at significantly lower rates than global peers, aided by government GPU subsidies. While this aggressive pricing targets local adoption, questions regarding long-term cost sustainability and technical capability relative to global leaders like OpenAI remain.
Indian artificial intelligence startups Sarvam AI and the government-backed project BharatGen are aggressively pricing their foundation models to capture market share. These entities are currently charging rates that are a fraction of what global technology leaders like OpenAI, Anthropic, and Google charge for comparable services. This pricing strategy is heavily supported by government-led subsidies on graphics processing units (GPUs), which are the core hardware required for training and running large-scale AI systems.
Financial data indicates a stark contrast in pricing structures. BharatGen’s Param-2 model is available at ₹5 per million output tokens, while Sarvam AI charges between ₹10 and ₹16 depending on the model's parameter scale. In comparison, global alternatives such as OpenAI’s GPT-5 Mini and Google’s Gemini 3.5 Flash are priced at ₹191 and ₹858 per million output tokens, respectively. This trend continues for input token costs, where Indian firms charge as low as ₹1 to ₹4, compared to the significantly higher premiums commanded by global providers.
Strategic Focus on Local Markets
The business objective for these Indian firms is not necessarily to replace global frontier models but to create a specialized business advantage. By focusing on Indian languages and regional use cases, these companies are building smaller, more efficient models that cater to local enterprise needs. Industry analysts note that enterprise clients are currently in the early stages of AI adoption and are looking for cost-effective, localized solutions rather than the general-purpose, high-cost models developed by major global corporations.
Sustainability and Operational Risks
While the current pricing provides a competitive edge, the long-term viability of this model remains a critical monitorable for investors and stakeholders. The reliance on government subsidies creates a potential vulnerability. Once these support mechanisms are phased out, the companies will have to account for the full cost of data center operations, electricity, and the high depreciation costs of expensive GPU hardware.
Furthermore, there is a clear capability gap between these specialized local models and the general-purpose frontier models from international competitors. Sarvam AI, which has secured $234 million in funding, faces the task of proving that its models can provide enough value to justify enterprise adoption beyond just price advantages. Investors will likely track whether these firms can maintain low operational costs and improve technical performance once they move beyond the early stages of development and state support.