India’s domestic AI Foundational Model

Context: In part of its ambitious India AI Mission, the government has selected three more start-ups- SoketAI, Gnani AI and Gan AI to build indigenous AI foundational models.
This is in addition to selection of Sarvam AI to build India’s indigenous foundational AI large language model (LLM), an open source 120 billion parameter AI model. The start-up has launched two models- Sarvam-1 model (2 billion parameters) and Sarvam-M (24 billion parameters) model with hybrid reasoning capabilities. 

Relevance of the Topic:Prelims: Key Terms related to Artificial Intelligence; India and AI- Government Efforts. 

Key Terms related to Artificial Intelligence

• Artificial Intelligence:
  • AI is the capability of a machine to imitate intelligent human behaviour and perform complex tasks similar to how humans solve problems. 
  • E.g., Perform cognitive tasks like thinking, perceiving, learning, problem-solving and decision-making. 
• Machine Learning: 
  • Machine learning techniques, including Artificial neural networks (ANNs), are used to achieve the goals of AI. 
  • Machine learning is a subfield of AI, and ANNs are a specific type of machine learning algorithm that uses interconnected nodes to learn from data.
• Artificial Neural Networks (ANNs):
  • Algorithms that are inspired by the structure and workings of human brains, and have the capability to identify and learn patterns in data. 
  • Breakthroughs in ANNs have enabled the development of LLMs and AI-tools like AlphaFold (An AI tool that can predict protein structures).  
• Artificial General Intelligence (AGI): 
  • Machines capable of ‘thinking’ and ‘acting’ autonomously through a process of self-learning or artificial general intelligence (AGI).
• Foundational AI: 
  • Large-scale AI models that are trained on very large datasets and over which numerous specific applications can be built, including generative AI. 
• Generative AI: 
  • AI models that use machine learning algorithms to create new and original content, such as images, text, code, audio, or even video with the help of natural-language prompts. E.g., DALL-E for image generation, ChatGPT for text generation. 
  • Generative AI models can be Large Language Models (LLMs) or Small Language Models (SLMs).
• Large Language Models (LLMs):
  • LLMs are a type of Foundational AI model trained with vast datasets with at least one billion or more parameters. 
  • LLMs have shown an exceptional proficiency to understand and interact in human languages in a meaningful way.  
  • E.g., AI-powered tools like ChatGPT, Gemini, Perplexity, DeepSeek, Grok. 
• Small Language Models (SLMs): 
  • SLMs are compact AI systems typically having fewer than 1 billion parameters (ranges from millions to a few billion parameters).
  • Cheaper to run and maintain, and ideal for specific use cases. 

India’s challenges to build its own Foundational Model

India has been focused on building AI-based applications for specific work, like in healthcare or drug discovery. But it aims to develop its own foundational model. 

• Building foundational models is an extremely resource-heavy and expensive exercise.
  • It involves massive computational infrastructure, enabled through specially designed state-of-the-art chips called Graphics Processing Units (GPUs).
    • Training advanced deep learning models demands substantial GPU clusters (thousands of GPUs run in hyperscale data centres, as big as one million square feet) and high-performance computing (HPC) facilities.
    • Shortage of GPUs (currently in high demand and short supply). AI Mission seeks to procure at least 10,000 of GPUs, which will require high skill and expertise.
  • The Model has to be trained on very large datasets which consume an enormous amount of electricity. E.g., LLMs like GPT-3 consumed nearly 1,300 megawatt-hours (MWh) of power. 
• Building applications on top of other country’s models can bring in layers of vulnerabilities. E.g.,
  • Models trained on global datasets often lack local nuances and can insert foreign biases, thereby producing unwanted or erroneous results.
  • In applications related to defence or national security, a foreign model always carries potential dangers of sabotage or leaking sensitive data.

India and AI: Government Efforts: 

• For India work on AI is at a relatively nascent stage. In 2024, India had launched a Rs 10,000-crore IndiaAI mission to build capabilities in AI. 
• India aims to build its own LLM within 10 months (till the end 2025).
  • The central government had received at least 67 proposals to build the India-specific models.
  • A high-level technical committee will evaluate the proposals. 
  • The intellectual property of the models will remain with the entity, with provision for a perpetual license for use by the government for public use.
• The government has also selected 10 companies to supply 18,693 graphics processing units or GPUs (high-end chips needed to develop machine learning tools) crucial for developing a foundational model.
  • The initial aim of the IndiaAI Mission was to procure 10,000 GPUs.
• The government will launch a common compute facility from where startups and researchers can access the computing power. To ease access to these services, the government will give a 40% subsidy to end users on the total price.

Also Read: Inclusive AI: AI Action Summit 2025 

India needs to create a centralised AI infrastructure, allocate substantial funding, including on procuring GPUs, and private-public Industry-academia participation.