India is accelerating its efforts to establish itself as a major player in artificial intelligence (AI) development, particularly as it faces increasing competition from Chinese advancements like DeepSeek. The Indian government is planning to host its own open-source AI models on Indian servers, marking a significant milestone in its AI strategy. According to the Union Minister for Information and Broadcasting, Ashwini Vaishnaw, the government has already determined how many servers and the necessary capacities for these models.
“We will very soon be hosting those open-source models on Indian servers,” Vaishnaw stated at his recent press conference. This initiative aligns with the government's broader aim to leverage the power of AI for population-scale problems, emphasizing applications across multiple sectors such as healthcare, agriculture, and disaster management.
The rise of DeepSeek, which has impressed many with its open-source reasoning model R1, has shifted the narrative about the requirements for building advanced AI systems. While traditionally, large language models (LLMs) have depended on vast computing resources—where R1 was trained on over 2,000 GPUs—the ChatGPT model required around 25,000 GPUs. This discrepancy brings forth the question of whether India should seek to replicate such approaches.
To support the local AI ecosystem, the Indian government is also working on launching a portal for startups and researchers to access GPUs at significantly reduced prices. Subsidies for students using these resources will allow their costs to fall to below Rs 100 per hour, ensuring broader accessibility to AI technologies.
Currently, India has approximately 10,000 GPUs available out of around 18,000, with 15,000 being high-end GPUs ready for deployment. Vaishnaw highlighted this capacity as part of the government’s plan to create AI applications capable of addressing large-scale challenges within the country.
Meanwhile, Krutrim, which has rapidly emerged as a unicorn startup led by CEO Bhavish Aggarwal, announced it is hosting DeepSeek's models on its cloud platform. Aggarwal touted the platform's pricing as the lowest globally, signaling Krutrim's commitment to making advanced AI accessible and pushing for innovation within India.
The AI models hosted by Krutrim, which range from 8 billion to 70 billion tokens, include the notable DeepSeek R1—highlighted for outperforming OpenAI’s GPT-4 across several metrics. This strategic partnership is poised to disrupt the AI model pricing structure, with comparison showing Krutrim pricing starting at INR 10 per million tokens for some models, significantly lower than alternatives available internationally.
“The foundational models made in India will be able to compete with the best of the best in the world,” Vaishnaw reinforced, indicating India's readiness to establish its footing against notable tech giants.
Part of India’s ambitious $1.2 billion IndiaAI Mission involves cultivating both large and small language models. The government has already approved 18 projects aimed at leveraging AI to tackle significant issues such as climate change and agricultural efficiency. Companies are anticipated to invest over $30 billion to improve data centers within the coming years.
With pressure mounting to create homegrown AI innovations, India's efforts are underscored by aspirations for technological sovereignty, particularly as concerns over data privacy and reliance on foreign systems remain prevalent.
Industry experts, including Dalberg Advisor’s Kunal Walia, advocate for enhanced research and development initiatives at both public and private sectors to optimize resources dedicated to designing these models. The importance of datasets unique to India's cultural and linguistic diversity is not lost on those involved; as Walia pointed out, fostering innovation relies heavily on access to such data.
Challenges nonetheless persist, with calls for stronger collaboration between academia and the tech industry to mobilize funding and resources effectively. This imperative clarity resonates deeply as India endeavors to position itself within the rapidly advancing global AI arena.
The emergence of foundational models has the potential to reshape the competitive dynamic within AI, and subsequent strategies being launched by India could offer valuable insights for other countries wrestling with similar technological ambitions. By engaging local organizations and prioritizing cultural relevance, India aims to create AI systems reflective of its unique social fabric, pushing against the linear path typically taken in technology development.
Negotiations and partnerships will play pivotal roles as India seeks to build upon its recent strides. The efforts undertaken by organizations like Krutrim demonstrate India's proactive stance in the AI race and the importance of localized models to meet specific societal needs. While reports indicate India is moving closer to having multiple foundational AI models ready for deployment within the next year, the industry remains vigilant, recognizing the considerable steps necessary to thrive within this transformative field.