The Indian government has initiated discussions with leading Indian-origin AI researchers working at global firms to encourage them to return and contribute to the country’s generative AI ecosystem. The move aims to reverse brain drain and bolster India’s AI capabilities, Moneycontrol reported, citing sources.
With China’s DeepSeek gaining traction, debates have emerged over whether India should develop foundational LLMs or focus on AI applications using existing models. Experts warn that hesitating on foundational development could be a strategic misstep.
“Once we create an environment for innovation, Indian-origin developers may be encouraged to return and contribute to building something significant in India,” the report quoted a government official as saying.
“We are already in talks with prominent AI researchers, including Niki Parmar, who has made major contributions to machine learning and deep learning. The government wants these AI experts to mentor Indian startups and young researchers. These innovators could leverage the IndiaAI Mission’s GPU computing facility to develop AI models tailored to India’s needs,” the official said on condition of anonymity, according to the report.
Some leading AI experts of Indian origin have reportedly shown interest in returning to India to contribute to building a strong foundation for the country’s AI development.
“This is the perfect opportunity for Indian-origin professionals working at OpenAI, Google DeepMind, Mistral AI, and Perplexity AI to return and contribute. If we are to build an alternative to China’s DeepSeek, we need their expertise,” the official reportedly said, adding that the government will encourage top academic minds to utilise the computing infrastructure.
“We will provide them with mentoring and other support to form their company to scale. We also want top India-origin folks to mentor them in developing AI applications and use cases. We want to nurture youngsters, and we need to find them. If you look at DeepSeek, it has all youngsters and people who have worked with companies like Google DeepMind,” the official reportedly said.
IndiaAI Compute Facility
India aims to develop its own Large Language Models (LLMs), positioning itself alongside global AI leaders such as OpenAI’s ChatGPT and China’s DeepSeek. This initiative will be driven by the India AI Compute Facility, which has secured 18,693 GPUs to power LLM development. Initially, the government aimed to acquire 10,000 GPUs but nearly doubled that number through competitive bidding.
The facility will include 12,896 Nvidia H100 GPUs, 1,480 Nvidia H200 GPUs, and other high-performance AI chips from AMD, Intel, and AWS. Of these, 10,000 GPUs are ready for immediate deployment, while the remaining units will be gradually installed. Starting January 3, startups and researchers can apply for GPU access through a government portal.
GPUs will be supplied by L1 and L2 bidders. L1 bidders include E2E Networks, Jio Platforms, Ctrls Datacenters, CMS Computers, Locuz Enterprise Solutions, NxtGen Datacenter and Cloud Technologies. L2 bidders include Orient Technologies, Tata Communications, Vensysco Technologies, and Yotta Data Services.
Tiered Approval System for AI Projects
To manage GPU allocations efficiently, the government has introduced a tiered approval system. AI projects will undergo different levels of scrutiny based on their scale, with large-scale initiatives requiring higher-level approvals from officials such as the principal scientific adviser, the minister, and the cabinet.
“If subsidies worth crores of rupees are involved, government compliance rules will be followed. The system is designed to ensure quick online approvals. Approval committees will include government representatives, industry bodies like NASSCOM, academicians, researchers, and AI experts,” a second government official reportedly said.
The committees will assess both the applicants and their intended use of GPU resources to prevent any misuse. The subsidy will be allocated directly to GPU resource providers, to support AI development.
Some people read for free. A few choose to support. If you found TelecomTalk useful, you can help keep us running.
Support Telecomtalk
