India’s AI Ambitions at a Crossroads: Should We Develop Our Own LLM in Light of US and China Advances?

In the U.S., major companies like OpenAI, Softbank, Oracle, Microsoft, and Nvidia are banding together to create AI infrastructure for OpenAI. They plan to invest a whopping $500 billion over the next four years in a project called Stargate.

Meanwhile, in China, a company called DeepSeek has launched a new large language model (LLM) that is said to measure up to—or even surpass—OpenAI’s model in various areas like math and coding. This model is open-source and available at a much lower cost than OpenAI’s offerings.

These developments have put India in a pivotal position in the global AI landscape. Now, India faces some important questions. Can it afford to scale up AI-specific hardware? Should it focus on creating a foundational model or build applications on existing models?

One major debate is whether India should invest in developing its own LLM from scratch. Some argue it should use existing open-source models to build applications instead. Nandan Nilekani, co-founder of Infosys, believes that India should not aim to create another LLM. He suggests that India should leverage existing models to generate synthetic data and develop smaller language models suitable for Indian languages.

On the other hand, some experts disagree with this viewpoint. Aravind Srinivas, founder of Perplexity AI, argues that focusing only on reusing existing models could limit India’s AI capabilities. He believes India should aim for both building foundational models and developing applications on top of them. The recent success of DeepSeek indicates that creating AI models might be more affordable than previously thought.

There’s a caveat, though. Building applications on top of existing models makes companies dependent on those models. If access to them changes, it could be a risk for businesses.

Another key issue is computing power. Nvidia currently dominates the market for the graphics processing units (GPUs) used to train advanced AI models. Companies like OpenAI and Meta have invested heavily in acquiring these GPUs, making the cost of AI operations a significant hurdle for Indian startups.

To support local companies, the Indian government has launched the Rs 13,370 crore IndiaAI Mission. This initiative aims to help companies acquire 10,000 GPUs and build AI data centers at affordable prices. However, plans could be complicated by recent U.S. regulations on AI hardware exports, which limit how many GPUs India can import.

The U.S. government has classified countries into tiers, each with different restrictions on AI chip exports. India is placed in the middle tier, meaning it will face some limitations on GPU imports unless they are used in secure environments. The law also includes provisions that could allow Indian companies to use exported technology for both civilian and military purposes—but not for nuclear applications.

Source link

stargate, LLM, artificial intelligence India, LLMs in india, deepseek, china ai, India AI, indian express explained