NVIDIA is looking to power the AI evolution—from tokens to agentic AI, says Jigar Halani

Modern enterprises are consuming and producing tokens at a rapid pace through application programming interfaces (APIs), Jigar Halani, NVIDIA’s Director - Solution Architect and Engineering, said during DevSparks 2025, YourStory's flagship developer summit held in Bengaluru. 

According to him, the era of experimentation is over as enterprises are now facing tough decisions around the choice of models, infrastructure, and return on investment (ROI). 

API tokens are consumed every time a user types in a query or when the model is put to work on a task. In simple words, tokens are being counted and billed as part of the AI workloads.

However, reasoning models generate a significantly higher number of tokens. This in turn places high pressure on the infrastructure that supports these models. Jigar noted that this gap signals for a change—away from GPUs to platforms. 

Also Read

YourStory launches Builders’ Guild to power India’s developer communities

Graphic Processing Units, or GPUs, are specialised computer chips that power AI and machine learning. 

“We’re no longer just a chip company,” he stated. “We’re a full-fledged platform company.”

Showcasing how NVIDIA is adapting to this shift, Jigar pointed out that the firm’s platform houses 72 GPUs, integrated networking, and orchestrated layers built to support wider AI training. But, the hardware is just one side of the coin. 

Jigar was talking about NVIDIA’s agentic AI vision—digital workers that can break down tasks, collect data, and execute missions. Some of the examples showcased include weather forecasters, research assistants, and more—all powered by NVIDIA’s tools like NeMo, Cosmos, and Metropolis. 

The company also introduced robotics and factory simulations that are running on NVIDIA’s Omniverse, indicating the scope of its usage for a wide variety of developing technology including self-driving cars and humanoids. 

“The age of agentic AI is here,” Jigar concluded. And, NVIDIA is giving you the blueprint, he added.