India AI: As DeepSeek and ChatGPT surge, is Delhi falling behind?

The dramatic reduction in generative AI application development costs by China's DeepSeek has sent shockwaves through the tech industry. Two years later, the global race for AI supremacy is intensifying, and India appears to be lagging, particularly in creating indigenous foundational language models to power functions like chatbots.

The Indian government claims that a homegrown DeepSeek alternative is imminent. They are providing startups, universities, and researchers with thousands of high-end chips to develop one within 10 months. Recently, many global AI leaders have also been discussing India's capabilities, adding to the optimism.

OpenAI CEO Sam Altman initially dismissed the idea but stated this month that India should play a leading role in the AI revolution. India is now OpenAI's second-largest market in terms of users. Companies like Microsoft have also invested heavily, pledging $3 billion for cloud and AI infrastructure. Nvidia's Jensen Huang has also stated that India's "unparalleled" tech talent is key to unlocking its future potential.

Despite having key ingredients for success, experts say India risks falling behind if it doesn't fundamentally address issues in education, research, and national policy. Tech analyst Prasanto Roy told the BBC that China and the US are already "four to five years ahead," having invested heavily in research and academia and developed AI for military applications, law enforcement, and now, large language models.

While India ranks in the top five globally in Stanford University's AI Vibrancy Index, which ranks countries based on metrics like patents, funding, policy, and research, it still lags far behind the two superpowers in many key areas. Between 2010 and 2022, China and the US accounted for 60% and 20% of total global AI patents, respectively. India accounted for less than 0.5%. In 2023, private investment in Indian AI startups was also a fraction of that in US and Chinese companies.

Meanwhile, India's nationally funded AI mission is worth just $1 billion, while the US has allocated a whopping $500 billion for Project Stargate, aimed at building massive AI infrastructure in the US, and China has reportedly invested $137 billion with the goal of becoming an AI hub by 2030. While DeepSeek's success shows that AI models can be built on older, less costly chips, Jaspreet Bindra says a major problem is the lack of "patience" or long-term capital from industry or government. Bindra is the founder of a consulting firm that works to improve AI literacy in organizations.

Bindra emphasized, "While people hear about DeepSeek developing a model for $5.6 million, there's a lot more money behind it." Another problem is the lack of high-quality, India-specific datasets to train AI models for regional languages like Hindi, Marathi, or Tamil, especially given India's linguistic diversity.

Despite the challenges, India excels in talent, with 15% of AI practitioners globally coming from India. However, Stanford University's AI talent flow research shows that a growing number of Indian talent are choosing to leave the country. Bindra says this is partly because "foundational AI innovation usually comes from deep R&D in universities and corporate research labs." India lacks a supportive research environment, and its academic and corporate sectors rarely have deep tech breakthroughs.

The huge success of India's payments revolution was attributed to strong government-industry-academia collaboration, he said, adding that a similar model needs to be replicated to advance AI. The Unified Payments Interface (UPI), a digital payment system developed by a government organization, has revolutionized digital payments in India, allowing millions to transact with the click of a button or a scan of a QR code.

Bangalore's $200 billion outsourcing industry, with its millions of programmers, should have been at the forefront of India's AI ambitions. But these IT companies never really shifted their focus from cheap, service-based work to developing foundational consumer AI technologies. "That's a huge gap, and they've left it to startups to fill," Roy said. But he is not sure if startups and government missions can do the heavy lifting fast enough, adding that the 10-month timeline set by ministers is a rushed reaction to DeepSeek's sudden emergence.

"I don't think India will be able to produce something like DeepSeek for at least the next few years," he added. Many share the same view. However, Bhavish Agarwal, founder of Krutrim, one of India's earliest AI startups, recently wrote on X that India can continue to build and fine-tune applications on existing open-source platforms like DeepSeek "to leapfrog our own AI progress."

In the long run, experts say that developing foundational models is crucial for strategic autonomy in the field, reducing import dependence and the threat of sanctions. India also needs to increase its computing power or hardware infrastructure to run these models, which means manufacturing semiconductors, but progress on this front has been slow. A lot needs to fall into place before it can close the gap with the US and China.