S. Umamaheshwar | India’s IT Empire Lags in Research To Shape the Future of AI

Artificial Intelligence is one of the most used words in India. It became a buzzword in the late 1990s, almost akin to Y2K. Everyone knew this would herald a once-in-a-generation change. But towards the end of January, something unexpected happened that put an end to the complacency enjoyed by those who favored the status quo.

On January 30, US technology company Anthropic released 11 open-source plugins for its AI workplace suite Claude Cowork. Unlike traditional chatbots, Cowork functions as an autonomous digital colleague: reading files, drafting documents, reviewing contracts, and executing multi-step workflows across legal, finance, sales, and marketing with minimal human instruction. Days later, Anthropic released Claude Opus 4.6, a model that can coordinate teams of AI agents for financial research and due diligence.

New artificial intelligence agents have virtually eliminated the role of humans in technology, creating convulsions in the global corporate world. Approximately $285 billion in market value was wiped out worldwide. In India, the Nifty IT index fell 5.87 per cent, its sharpest decline since March 2020, when India announced the Covid-induced lockdown, wiping out nearly Rs 2 lakh crore. TCS and Infosys fell over 7 per cent each on the day; Tech Mahindra lost over 5 percent.

The underlying fear driving this meltdown: If a single AI agent can do the work of teams, India’s headcount-based outsourcing model will face existential repricing. This threatens India’s dominance in global IT and ITeS services, forcing Indian businesses and the government to take a fresh look at whether the country is ready to adapt and build domestic AI infrastructure.

Even before ChatGPT popularized AI in the country, the Central government had approved IndiaAI Mission, a 10,300-crore program under the Union ministry of electronics and information technology (MeitY) in 2024. The goal was audacious but clear: to build India’s own AI infrastructure, train its own models, and reduce the country’s dependence on foreign AI systems controlled by American and Chinese firms.

About 18 months ago, India was debating whether it needed 1,000 GPUs (graphics processing units) for AI research. But by 2025, the government had distributed more than 38,000 GPUs, surpassing the initial target of 10,000, and offered them to Indian startups and researchers at a subsidy rate of just £65 per hour. This target has increased to 50,000 GPUs by the end of the year.

India had built one of the most accessible public computing repositories in Asia almost from scratch. India has become one of the four countries in terms of computing capacity, up from seventh place two years ago, a truly excellent performance.

India’s most ambitious goal was to develop indigenous large language models (LLMs), which are artificial intelligence systems trained from scratch with Indian data, capable of reasoning in Indian languages, and deployed within India’s borders.

The government selected Bengaluru-based startup Sarvam AI in April 2025 to launch India’s dominant LLM. In parallel, the government-led BharatGen initiative, anchored at IIT Bombay, has built Param 2, a multilingual model with 17 billion parameters. Both projects train in all 22 planned Indian languages. The goal is to eventually scale to a trillion parameters, serving use cases in agriculture, law, healthcare, finance, and education.

Beyond flagship projects, the IndiaAI Mission has expanded to include 12 organizations creating LLMs: from Fractal Analytics (developing India’s first major reasoning model) to Tech Mahindra (creating the first model in Hindi for education), Gnani.ai (speech-to-speech AI handling 10 million voice interactions per day) and Socket AI (building open-source core models with permissive licenses).

The Mission has also established Centers of Excellence in health, agriculture, sustainable cities and education. It founded the IndiaAI Security Institute, created a dataset platform, and launched a skills program to retrain government officials and young professionals for an AI-powered economy.

So how is India performing globally in AI technology? While success far outweighs the time spent, it is still far from the limit.

According to the Stanford AI Index, India ranks among the top four countries globally in terms of AI skills, capabilities and policies; This marks a rise from seventh place just two years ago. India is the second largest contributor to AI projects on GitHub. The adoption rate of enterprise AI stands at 89 percent, well above the global average of 69 percent. A wave of foreign capital is coming to India, indicating that the country will not be a peripheral market but ultimately an artificial intelligence market.

However, in frontier model development, India is yet to develop a system that can compete with GPT-4, Gemini or China’s DeepSeek. While India’s graduate studies are promising, culturally relevant and affordable, they are not yet world-class.

This difficulty is the most severe. There are less than 300 skilled AI researchers in India; which is much less than in England or France.

Although India produces more engineers every year than any other country, its portfolio of deep AI researchers is thin.

India has tremendous data from digital public infrastructure at the population scale (Aadhaar, UPI, ONDC) allowing companies to create cost-effective solutions for people. It spends only 0.7 percent of GDP on research and development. Without sustained and patient investment in basic research, India risks remaining a consumer and adapter of AI and creator. Companies also focus too much on paying dividends to shareholders rather than investing in cutting-edge R&D.

In an age where militaries are leveraging AI capabilities, India’s use of foreign AI capabilities is akin to dependence on a foreign military. Therefore, India needs to stop brain drain and race against time to achieve its goals.