The Nvidia AI summit has created a platform for Nvidia and India to work together to build technical skills and move India from exporting IT services to exporting AI. The development work has begun on building the AI infrastructure with 20x compute capacity.

🤔 But what is this computing capacity?

For AI to work, you need a large amount of data on which the computer can train, understand, generate, and respond. Large Language Models (LLMs) train the computer to understand human language and respond in the same language across various tasks and contexts. For instance, you can instruct your AI assistant to make a phone call, play a song, or even switch on the lights in your language, and it will correctly interpret, respond, and perform the task.

At present, AI understands English. But several Indian companies are looking to train AI on Hindi LLMs, which is challenging as there are also many Hindi dialects. Next, various applications will have to be created around it. If we look at Jio’s AI plans, it aims to launch applications like the HelloJio Voice Assistant, Jio Phonecall AI (that can convert voice calls into text), and the JioHome app to control your WiFi smart home devices.

Training LLMs requires massive computing power, huge costs, and electricity. That’s where Nvidia’s chips come in. From Ampere to Hopper to Blackwell, Nvidia keeps introducing new chips that double or triple the performance while reducing the computing cost by half or by a third and speeding up the processing of inference through algorithms.

📈 Nvidia’s India ambitions

So, when Nvidia says India’s compute capacities will grow 20x, it means Nvidia’s computing chips will power several data centers, while its algorithms will enhance their efficiency. The tech giant has partnered with several big and small Indian companies for this.

🤝 Joining hands with Reliance

Reliance already has the data thanks to its strong connectivity infrastructure. It now needs computing power to convert this data into intelligence. It aims to implement Jio’s AI “Everywhere for Everyone” by establishing gigawatt-scale AI-ready data centers in Jamnagar. And NVIDIA will supply its GH200 Grace Hopper Superchip and AI supercomputing cloud service called DGX Cloud to help Reliance build an AI infrastructure “more powerful than the fastest supercomputer in India.”

The GH200 Grace Hopper systems are already powering more than 40 AI supercomputers across global research centers, system makers, and cloud providers for robotics, big data, climate research, drug discovery, and more. Reliance will bring this supercomputing power and make it accessible to Indian tech startups to make AI applications. This partnership will help NVIDIA strengthen its presence in India.

📃 Nvidia’s other Indian collaborations

While the Reliance partnership was well-known, Nvidia is also partnering (for chips and software) with:

• Yotta to build a fundamental computing infrastructure in India.
• Tech Mahindra to develop ‘Indus 2.0’, an AI model in Hindi.
• Infosys and Wipro to build custom LLMs for their enterprise clients.
• Indian startups like Flipkart, Zoho, and CoRover.ai to roll out Hindi-based LLMs.
• IIIT-Delhi to develop an AI-powered data integration and predictive analytics tool called AMRSense that can detect antibiotic prescription patterns in local languages.