Nvidia Is Officially on the A.I. Agents Train With New Family of LLM Models
Nvidia is going all-in on A.I. agents as it seeks to cash in on what CEO Jensen Huang calls a “multi-trillion dollar opportunity” that could change how IT teams operate. During his keynote speech at the 2025 Consumer Electronics Show (CES) yesterday (Jan. 6) in Las Vegas, Huang unveiled Nvidia’s Nemotron family of large language models that developers can use to build A.I. agents—autonomous bots that can perform complex tasks without human control.
Built with Meta (META)’s Llama model, the open-sourced Nemotron models, including Nano, Super, and Ultra, allow developers to “create and deploy” A.I. agents they can fine tune to specific business needs, Nvidia said. Applications include customer support, fraud detection, and inventory management optimization. By deploying A.I. agents, companies can “achieve unprecedented productivity,” according to the company.
“In the future, these A.I. agents are essentially a digital workforce that are working alongside your employees doing things for you on your behalf,” Huang said during his keynote. “The way you would bring these specialized agents into your company is to onboard them just like you would on-board an employee.”
In a live demonstration, Huang showcased examples of A.I. agents with digital replicas of his face. The A.I. research assistant agent for students, for instance, is designed to take documents like lectures, academic journals and financial results and synthesize them into an interactive podcast for “easy learning.” The software security agent can scan software and alert developers if vulnerabilities are detected so they can take immediate action. Other A.I. agents displayed included roles like financial analyst, employee support and factory operations.
“The IT department of every company is going to be the HR department of A.I. agents in the future,” Huang said.
Major companies are already lining up to use Nvidia’s Nemotron models. One early user is the enterprise software giant SAP. Philipp Herzig, chief A.I. officer at SAP, said in a release the company expects “hundreds of millions of enterprise users” to interact with Nvidia’s A.I. agents to “accomplish their goals faster than ever before,”
Another enterprise software giant, ServiceNow, will use Nvidia’s new models to “build advanced A.I. agent services” that can solve a complex range of problems in a move to “achieve more with less effort,” Jeremy Barnes, the company’s vice president of platform A.I., said in a release.
Nemotron is just one of many products Huang announced during his nearly two-hour-long keynote. The CEO also unveiled a new family of its latest Blackwell chips which the company claims is three times more powerful than previous generations of GPUs. The Blackwell family is expected to create a $100 billion market opportunity for Nvidia, Stifel analyst Ruben Roy wrote in a note last November. On top of that, Huang announced the GB200 NVl2, a super chip that it claims will supercharge the capacity for data centers to run A.I. workloads.
Nvidia isn’t just betting big on its highly sought-after chips. Huang also announced the Cosmos foundation model to advance physical A.I. like humanoid robots, the DRIVE Hyperion platform to train self-driving vehicles, and Project Digits, a $3,000 personal supercomputer that Nvidia says is 1,000 times more powerful than a typical laptop. Nvidia also announced two autonomous driving partnerships, one with Toyota and the other with Aurora, a self-driving truck startup.
The latest A.I. developments cements Nvidia’s position as a global leader in the A.I. revolution. Prior to Huang’s speech, Nvidia’s stock jumped more than 3.4 percent to a record high of $149.43 per share after market close in anticipation of the CES announcements. That surge sent Nvidia’s market value up to $3.47 trillion, briefly surpassing Apple’s. Nvidia’s market value grew by over $2 trillion last year.