Bhat's Notes

Excerpts from the fireside AI chat between Mukesh Ambani and Jensen Huang at the NVIDIA AI Summit at Jio Convention Centre in Mumbai on 24th Oct 2024

India. AI. NVIDIA. AI. Reliance. AI. A powerful convergence of innovation. It’s Generative AI for Indian enterprises. It is NVIDIA’s pioneer with Reliance’s visionary. Showtime between Huang and Ambani.

Jensen Huang, Jensen Huang was born to Taiwanese immigrants, co-founded Nvidia in 1993 after earning a master’s degree from Stanford. By June 2024, Nvidia became the world’s largest company by market cap, and as of October 2024, Huang’s net worth reached $125.5 billion, making him the 10th richest person globally.

Mukesh Ambani: Mukesh Ambani is the chairman of Reliance Industries, transforming it into a global leader in energy, telecom, and retail. He launched Jio, revolutionizing India’s telecom sector. As of October 2024, he is the richest person in Asia, with a net worth of $119.5 billion, and is driving India’s push towards digital and green technologies.

The 10 AM start was delayed to 10:30. It’s now 11 AM, and the clock is still ticking. The main hall is full. Thousands of AI enthusiasts are still waiting—could the payoff be worth it? With two big shots on stage, 5,000 corporate man-hours, and 10,000 cumulative hours of queue time, the anticipation continues to build.

The chat started at 11:11AM. Finally.

Jensen Huang of NVIDIA: India is dear to the world’s computer and IT industry. Both my industry and your industry are undergoing a fundamental seismic change. Let’s talk about that today.

There are two fundamental shifts happening at the same time. The last time we witnessed something of this magnitude was in 1964, the year after I was born. IBM introduced the world to the concept of IT, describing the CPU and the separation of hardware and applications through a layer called the operating system. They highlighted the architectural benefits across generations, recognized the importance of the installed base, and underscored the significance of software investments and building systems that run applications.

The same industry that the Indian IT sector grew from—the general-purpose computing that has existed for the past 60 years—has been powered by Moore’s Law. We’ve seen performance double every year while reducing costs by half. It’s been the most incredible depreciation force any technology has ever seen, making IT more accessible to society.

The invention of System 360 and Moore’s Law, combined with Windows PCs, drove industries with IT. But now we know that CPU scaling has stalled. The depreciation we relied on has ended, We can no longer expect that doing nothing with software will lead to automatic improvements in performance and the ability to tackle bigger challenges.

At NVIDIA, we started our company to delay this challenge. We rethought computing workloads with the CUDA model to accelerate applications. This acceleration provides the same advantage as Moore’s Law. One example is real-time computing graphics using GPUs, which democratized computer graphics as we know it. Over the past 30 years, we’ve been on a journey of acceleration.

There’s no magical processor that accelerates everything in general. We had to reinvent the entire computing stack. One domain after another, we extended the CUDA architecture to industries like simulation, computer-aided engineering, 5G radio, quantum computing, gene sequencing, computational lithography, data processing, and decision optimization. For each of these fields, we worked closely with industry experts, software engineers, and stack developers to create custom NVIDIA products and libraries, which is why it took time.

These acceleration libraries now cover every industry. CUDA has reached a tipping point. In 2012, AlexNet made a massive leap in computer vision, and we took a step back to examine the power of deep learning.

In the traditional approach, Software 1.0, programmers wrote algorithms that produce a specific output. The process of coding and programming became a central industry during our generation. But that approach has been disrupted by Software 2.0—machine learning. We’ve moved from human coding to machines learning. This is the shift from software to AI, from CPUs to GPUs running neural networks. The entire technology stack has been reinvented, and what software can do is now fundamentally different.

Today, NVIDIA has built incredible systems designed to study data at an enormous scale. This is the breakthrough. Over the last several years, we’ve learned the meanings of images, videos, proteins, particle physics, amino acids, text, and more. We can now translate one modality into another—pixels to words, videos to captions, proteins to chemicals, genes, and beyond. This is essentially a universal translator across modalities. It has sparked the rise of many startups, and applications of Generative AI have exploded globally, all thanks to the ability to process massive amounts of data.

Take Blackwell, for instance. It’s a production system with 144 GPUs connected via NV-Link backbone, linked to an advanced switch, and running sophisticated software. It includes Megatron for training large language models (LLMs) and TensorRT for model distribution, all on top of NVIDIA AI.

The Blackwell system is extraordinary. Each rack weighs 3,000 pounds and offers the highest computing density the world has ever seen. It operates under scaling laws: the more data you have, the larger the model. Every year, data and model size double, which means computation increases fourfold. We’re seeing progress faster than Moore’s Law ever delivered—four times more computation each year compared to Moore’s 100x every decade.

We’ve also discovered the importance of thinking compute—not just processing intelligence. The more time the system spends “thinking” through complex problems, the higher the quality of results. This concept of longer thinking time, as seen with OpenAI’s GPT, brings about a new way to scale AI.

Planning trips, managing logistics, and solving complex multi-step tasks require this thinking compute, and that’s what led to the creation of Blackwell. NVIDIA is AI in India. When you hear “NVIDIA” and “India,” it sounds almost mystical, like great places for innovation. In one year, we’ll have 20 times the compute power we had the previous year.

The first step in building an AI ecosystem is infrastructure. The next is the operating system of AI: LLMs. We’re building Hindi LLMs, aiming to cover 25 languages and their dialects across India. It’s the hardest LLM project in the world, but if India can figure it out, it can lead the way for the rest of the world.

AI-native companies are creating new applications, and our service partners—like Indian consulting giants—are taking these models to enterprises globally.

At NVIDIA, we differentiate between NVIDIA Enterprise and NVIDIA Omniverse. LLMs help understand the data, from reception to reasoning and task execution. Agents break down tasks into steps, sometimes coordinating with other AI models. These agents are connected to a central LLM and handle tasks like marketing, chip design, software engineering, supply chain management, and more. They act as supercharged employees.

We onboard these agents using a “training curriculum,” company-specific vocabulary, evaluation systems, guardrails, and more. We’ve integrated this entire process into the NeMo platform, where agents are created, onboarded, and continuously improved. The API microservice, called NIMs, helps operate AI efficiently. NeMo helps you build, manage, and operate AI at scale

Vishal, India Head NVIDIA: In India, we have SARVAM as a prime example of how AI can make India communicate at scale. They were able to achieve this because of the advanced AI infrastructure available today. Then there’s Bharat GPT, which is pushing the boundaries even further. The work being done at IITs and across other enterprises is a testament to how collaboration can solve critical challenges that India faces.

I see NVIDIA playing a crucial role in helping India revolutionize sectors, enabling us to build well-rounded solutions that promote communication, health, and beyond. It’s about doing it in India, for India, and eventually, in India for the globe.

Huang: The next generation of IT is all about delivering AI that is far more impactful and exciting than ever before. NVIDIA’s role is to build and deploy AI while working with India’s vast IT workforce to create intelligent agents tailored for India’s needs.

From first principles, we reinvented the computing stack. There are two most general forms of AI: agents that assist us in performing superhuman tasks, and robots, the physical embodiment of those agents. NVIDIA’s role is to create the technologies that enable India to develop and deploy these agents.

India has a history of exporting software, and in the future, India will export AI. When you export AI, you need machinery that transforms energy into tokens or floating-point numbers, which then reconstitute into AI. The future will witness the rise of a new industry—the large-scale production of intelligence. I look forward to partnering with India to place it at the center of this new industrial revolution.

The Omniverse is a virtual world that follows the laws of physics, allowing robots to learn and behave like real robots. AI can now be embedded in any robotic system, whether it’s a picking arm, a car, or an entire factory, a system we refer to as AGX. From Omniverse to DGX to AGX, we are entering the era of physical AI, which will transform industries in ways we could never have imagined.

AI may not do 100% of your job, but it can handle 20-50% of it a thousand times better. The real threat to jobs isn’t AI itself—it’s the people who know how to use AI effectively. In the long term, we’ll all be working alongside AI agents, integrating their capabilities into our daily tasks.

Our fundamental mission is to develop AI that is systematically safe. While some may attempt to use AI for nefarious purposes, it is our responsibility to ensure that the AI we create is secure and used ethically.

Welcome to the creator of the modern digital fabric of India. No one has contributed more than you in making India a leader in high tech and deep tech. So, why is AI the pivotal moment for India?

Mukesh Ambani of Reliance: Huang, Welcome to our city of Mumbai. My wife asked me to mention that she designed this venue, Jio Conventional Centre. In India, the word Vidya holds great significance—it means knowledge. When I think of NVIDIA, it reminds me of Vidya. In our culture, Vidya is represented by Saraswati, the goddess of knowledge. When you sincerely devote yourself to Her, you gain knowledge, and then the goddess of prosperity follows.

That’s exactly what has happened to you, Huang. You are leading a knowledge revolution, transforming it into intelligence that is driving prosperity for the 8 billion people of the world. Thank you for your immense contribution to the world and for ushering in the intelligence age in our lifetime.

Huang: It’s an incredible joy to contribute in this way. The IT industry in India is massive, and its most valuable resource is its talent in IT. Over the last couple of years, we’ve been working together to upskill India’s IT workforce into AI. The big question now is: How do we transform India into a global center of AI? What steps must we take to achieve this?

Ambani: As our Prime Minister said, this is a new aspirational India. With an average age below 35, our youth and aspirations are driving our technology forward. Our government has played a crucial role as a leader in this transformation. Third, we as Indians have raw talent. India has become home to the world’s leading companies, and we are growing the fastest in terms of Global Capability Centers (GCCs). We are advancing in space research, pharmaceutical research, and now even exploring chip manufacturing. Energy industries are driving their innovations in India. It’s clear that India is fast becoming an innovation hub.

NVIDIA now has 10,000 engineers in India, and we’ve built the necessary infrastructure, equivalent to roads in the digital world. Today, apart from the USA and China, India has the best 4G/5G digital connectivity. Jio has been a key player in this transformation, taking India from 158th to number 1 in the world in just 8 years. Despite having no prior knowledge in this field, today we are the largest data company in the world, with volumes larger than AT&T and Verizon combined.

What’s most satisfying is that Jio delivered about 16 exabytes of data this year. In the USA, you pay $5/GB, but in India, Jio delivers it at $0.15/GB. This showcases the perpetual value that technology has delivered to customers, truly democratizing access to the digital world.

This signifies the power of technology in reaching the people of India. We can bring prosperity not only to India but to the world. This is why India is becoming one of the biggest intelligence markets. Indians won’t just export CEOs to the world’s largest companies—we will have hundreds of millions of Indians delivering AI to the world.

This transformation can’t be achieved by any single individual. It’s a collective effort, and together, we must ensure that we deliver AI safely to the world.

Huang: The immense data available in India brings us to this important announcement. We need data in massive quantities and the corresponding AI infrastructure to truly harness its potential. Today, we are announcing that Reliance and NVIDIA are partnering to build this AI infrastructure. With AI data, AI infrastructure, and the talent of the Indian people, this idea becomes a powerful flywheel for innovation!

Six years ago, Prime Minister Modi asked me to address his cabinet about AI—long before AI became a global discussion. I explained to him why AI infrastructure is as essential as roads and energy. He understood and said that India should manufacture its own AI. He told me, “Don’t expect to import intelligence through data. India shouldn’t export flour to import bread!” His vision was clear—Indians should partner with NVIDIA to build AI right here.

He also said back then, “AI has the ability to elevate the entire population of India.” Programming is difficult, but Indians are great at it. However, everyone can program AI. If AI is placed in the hands of every citizen, this incredible capability can benefit society. Even six years ago, PM Modi had a profound understanding of how AI could transform India.

Ambani: We have a visionary leader who believes not only in the vision but also in its execution. Now that GPT has matured, I look forward to seeing India leverage the absolute best that NVIDIA has to offer. We are building infrastructure for 1GW, with the capacity to scale even further. At Jamnagar, we already have green power, and we are working to expand this significantly.

India and Indians need to repeat the success of Jio, making AI and intelligence accessible and affordable for the common people. To achieve this, we are designing and building the necessary infrastructure, ensuring that our customers won’t need to change their phones or computers—we’ll handle the heavy lifting of putting that infrastructure together.

I have great respect for Mark Zuckerberg and what he has done with LLAMA by promoting openness. Just as Linux was open-sourced, we can use LLAMA as a base model and develop new models on top of it. We will drive forward with an Indian AI model in the future.

We also look forward to seeing Indians trained in Omniverse, applying AI in practical, real-world ways. Just as we surprised the world with Jio’s impact in data, we will do the same in the Intelligence industry, showcasing what India can achieve on the global stage.

Huang: We are leveraging the India advantage to drive the flywheel—turning intelligence into data and data into intelligence. Let India take full advantage of the intelligence revolution that lies ahead of us.

Ambani: Huang, Diwali, the festival of lights and also our new year is around the corner. wish you happy Diwali.

Huang: Happy Diwali.