At NDTV AI Summit, NVIDIA's Jay Puri Breaks Down 5-Layer AI Cake

NVIDIA's Executive Vice President Jay Puri on Wednesday said that AI's ability to manufacture intelligence at scale is a "miracle" and that it is unleashing a new industrial revolution.

Speaking at the NDTV AI Summit in New Delhi, Puri said that AI is going to be as essential infrastructure as electricity, transportation, and the internet.

The impact on society is going to be huge, Puri, who is leading the NVIDIA delegation at the India AI Impact Summit, said.

Talking about how it became a reality, he said there is actually a very fundamental change in the way computing and software are done, which is what made it possible.

"In the past, software was very explicit. You had to tell a computer exactly what to do. You wrote the program or the algorithm, gave it access to structured data, and then it would essentially do what you asked it to do. AI and the computers now are very different. You can basically talk to them and ask them questions; they understand what your intent is and what the context is. And then they're able to reason about things and, you know, give incredible responses or take action and so on," he said.

'Energy, Chips, Infrastructure, Models and Applications': Executive Vice President Jay Puri speaks to @ShwetaRKohli about NVIDIA's 5-layered cake@nvidia | @SPF_India pic.twitter.com/9ShYw5AlmL

— NDTV (@ndtv) February 18, 2026

"It's really amazing because now the types of things you can do literally are what humans could do before with our intelligence," Puri said.

He said everything now is being done in real time.

"That requires a tremendous amount of compute to be able to do that, and NVIDIA essentially pioneered that method of computing... In the past 10 years, since AI came to the fore, NVIDIA has increased computing by 1 million times," he said.

He also spoke about NVIDIA CEO Jensen Huang describing the AI infrastructure stack as a five-layer cake.

"At the foundation, it's about energy. The amount of energy, the amount of compute or intelligence, and the amount of intelligence that you will be able to produce fundamentally, of course, depend upon how much energy you have. And so that is the most basic of constraints," he said.

"On top of that, we have chips. That's where NVIDIA builds the most advanced chips when it comes to parallel processing and graphics processors. You need high bandwidth memory. You need very fast interconnect and so forth," Puri said.

"These chips, you have to make sure they're all co-engineered to work well together, because ultimately they determine the horsepower of the system," he said.

The next layer of the stack is what we call infrastructure, he said.

"This is actually building the AI factories. So you need the land. You need to build the shell for the data center. You need to provision it with the appropriate power and cooling and connectivity and so on. Then you take the chips that I talked about. You put them in servers, and you connect them with very high-speed networking. You have a whole software layer that is on top to make it work. So these are AI factories," he said.

NVIDIA has reference architectures that help you design the data center appropriately, he said.

On top of the infrastructure, the next layer then is the models, he said.

"These models are like ChatGPT, Gemini, Claude, and Grok," he said.

He said the other set of models that are there are the open models like Deepseek and Quinn and indigenous sovereign models like India's Sarvam.

Speaking about the top layer, he said it is the most important.

"And those are the applications that are built on top. Because these are the applications where the real economic value is brought to industry. So whether it be healthcare, manufacturing, or finance, every single industry is going to be transformed," he said.

"So that's the five layers: energy, chips, infrastructure, models, and applications. Each layer sort of pulls on the layer before beneath, and all are extremely important," he said.