The government’s selection of Bengaluru-based start-up Sarvam to develop the country’s first indigenous large language model (LLM) under the `10,370-crore IndiaAI Mission is certainly a commendable step. The announcement regarding the intent to build a home-grown foundational artificial intelligence (AI) model was made in January, and Sarvam’s selection from 67 applicants followed within just two to three months. This rapid decision-making reflects a sense of urgency and ambition to position the country as a global AI contender, especially amid the disruptive rise of China’s low-cost, open-source DeepSeek model. The government’s investment in 18,693 graphics processing units (GPUs), an AI safety institution, and societal applications in agriculture and education signals a holistic approach. By fostering local talent and infrastructure, the government is laying the foundation for technological sovereignty. This initiative not only advances AI innovation but also heralds the country’s transition from a service-driven to a product-driven nation. While monetisation and global competitiveness remain daunting, a strong beginning has been made. Sarvam’s journey will test India’s ability to carve a niche in the AI ecosystem, proving that ambition, backed by swift action, can reshape the nation’s technological destiny and inspire a new era of innovation.
Sarvam, backed by access to 4,000 GPUs for six months, is tasked with building a 70-billion-parameter model optimised for voice, reasoning, and fluency in Indian languages. As information technology minister Ashwini Vaishnaw noted, innovations in programming and engineering will enable this model to rival global leaders’. Sarvam’s three variants: Sarvam-Large, Sarvam-Small, and Sarvam-Edge, promise versatility for advanced, real-time, and on-device applications. By prioritising voice-based AI and local data sets, Sarvam aims to tap into the country’s linguistic diversity, potentially enabling farmers, rural communities, and non-tech-savvy users to interact seamlessly in their native languages. This focus aligns with the government’s vision of democratising AI and has the potential to be a game changer.
However, while the swift progress made by the government deserves applause, the road ahead is fraught with challenges, particularly in marketing and monetising a closed-source model in a competitive global landscape. The harder battle lies in monetising this closed-source model. Unlike DeepSeek, which leverages open-source accessibility to gain traction, Sarvam’s proprietary approach, as confirmed by co-founders Vivek Raghavan and Pratyush Kumar, aims for strategic autonomy and enterprise appeal. While this offers monetisation potential through subscriptions or application programming interface access, the global AI market is fiercely competitive. Sam Altman, CEO of OpenAI, recently highlighted the financial strain of proprietary models, stating that OpenAI was losing money on ChatGPT Pro subscriptions due to unexpectedly high usage outpacing the $200-per-month pricing he set. This underscores the challenge Sarvam faces: balancing development costs, user demand, and sustainable pricing in an open economy where global giants like ChatGPT, Gemini, and DeepSeek dominate.
Surely, Sarvam’s objective of a voice-based, culturally nuanced model offers a unique value proposition, but differentiation alone won’t guarantee success. Past attempts like Hike and Koo struggled against global competitors, and even the 4G telecom stack has so far found limited domestic adoption. Sarvam would need to navigate an open market where Indian users, as Altman noted, form OpenAI’s second-largest base yet overwhelmingly prefer foreign models. High operational costs, including compute and talent, combined with the need for a robust go-to-market strategy, pose significant hurdles. Additionally, closed-source models risk limited transparency, potentially deterring trust among enterprises wary of biases or data privacy issues, especially in sensitive sectors like healthcare or finance.