The Chinese Deep-seek LLM AI model launch has disrupted the AI market overshadowing one of the best performing AI models such as OpenAI’s ChatGpt.
In a move to compete with Global AI models such as DeepSeek and ChatGPT, the Indian government has announced its plans to develop its own foundational AI models within the next 8-10 months.
To support this initiative, India AI Mission worth Rs. 10,000 crore, will establish state-of-the-art computing infrastructure powered by 18,693 graphics processing units (GPUs). These GPUs will be provided by companies such as E2E, Jio Platforms, NxtGen Datacenter, CtrlS, Locuz Enterprise Solutions, CMS Computers India, Tata Communications, and Yotta Data.
As per Digital India Programme, these models will be tailored to the Indian context, incorporating native languages, cultural nuances, and region-specific datasets while ensuring there’s no bias.
The rates secured for AI computing are kept highly competitive. The average cost per GPU hour is Rs. 115.85 ($1.34), compared to the global benchmark of $ 2.5 – $3 per hour. For high-end computing, the rate is Rs. 150 per hour. The common computing facility will be even more affordable, costing less than Rs. 100 per hour for startups, academia, Students, and researchers, as per the IT and Electronic Ministry.
To facilitate AI foundational model development, the government invited proposals from startups, researchers, and entrepreneurs. It will explore various models like initially providing direct funding to startups, with disbursements of payment on completion of contractual obligations.
Financial Support may come in the form of direct grants or in the form of Computational tokens or Credit which can be redeemed to procure various other computing resources such as CPU cycles, Memory allocation, Cloud Storage Capacity and Network Bandwidth etc these are often Quantifiable, Exchangeable also, which thus can be used for AI computing.
Additionally, an equity-based funding model is also under consideration, wherein the IndiaAI initiative may invest further by acquiring stakes in AI startups. For the shared AI computing facility, which will be accessible to startups, academia, and researchers, the government will subsidise 40% of the total cost.
Bidders and GPU Models:
10 companies (bidders) will provide 30 different GPU models. Some of these GPU models come from well-known brands like Nvidia, AMD, and Intel.
Preference to L1 Bidders:
The government will prioritize the 5 companies with the lowest bids (L1 bidders). These L1 bidders are: Jio Platforms, E2E, NxtGen Datacenter, Locuz, and CtrlS.
How will it work:
The L1 bidders will provide their GPU resources first. If their resources are utilized, then other bidders like Yotta and Tata Communications can participate. However, these other bidders must match the L1 price to offer their resources.
Special Case: Yotta has proposed the highest GPU capacity (9,216 units). However, they will only be able to offer their resources after the L1 bidders’ capacity is fully utilized.
Why are GPU’s Crucial to AI Computational Processes ?
GPU’s consist of thousands of cores that can perform massive amounts of highly intricate calculations simultaneously and make them ideal for parallel processing.
AI algorithms rely heavily on Matrix Operations which can be performed by GPU efficiently. With the ability to calculate billions of numerical values in decimal points instantly offering high performance in machine learning, deep learning and scientific simulation. GPUs possess high bandwidth facilitating fast data transfer and processing. National language processing for text summaries and translations, Computer vision for fast image and video processing and object detection and Autonomous Vehicles for AI system In self-driving Vehicles, enabling real-time processing and decision making are some of GPU’s Real-World Applications.
The government has assured that the Accreditation process will remain open, allowing additional entities to join in the future. A new procurement notice for additional GPUs is also expected soon.
More Importantly, DeepSeek is an open-source model and soon to be hosted on Servers in India, just as the government did with Meta’s Llama. In this context, to Safeguard the Data Safety and Privacy Parameters, IT Minister Ashwini Vaishnaw stated that Indian AI models will be hosted on Domestic Servers.
In a parallel effort to ensure ethical AI deployment, the government has announced an AI safety institution, which will oversee eight key projects focused on AI bias mitigation strategies, privacy-enhancing strategies, machine unlearning, AI Governance testing Frameworks, algorithm auditing tools, among others.
Another crucial component of the India AI Mission Mission is the deployment of AI applications to address large-scale societal challenges. In the first round of funding, 18 projects have been selected across three key themes: agriculture, climate change, and learning disabilities. With these ambitious initiatives, the government is setting the stage to become a global leader in AI development, ensuring technological sovereignty while fostering innovation in AI research and applications.
