Jensen Huang says agentic AI scales 100x

Agentic AI is the new sales pitch for the whole AI industry — and Jensen Huang is trying to make it sound less like a chatbot upgrade and more like a new operating model for companies. At ServiceNow’s Knowledge 2026 conference in Las Vegas on May 5, the Nvidia CEO argued that AI agents can take work humans currently do as services and turn it into software that scales far more cheaply and broadly. That is the core claim here. Not “AI helps workers.” More like “AI becomes the worker layer.” ### What does Huang mean by “agentic”? He means AI systems that do more than answer prompts. They reason through tasks, use tools, call software, follow rules, and complete multi-step work inside real business systems. Nvidia and ServiceNow are pitching that as an “autonomous workforce” — agents that can act across IT, customer service, HR, security, and operations instead of just drafting text on command. (nvidia.com) ### Why is the “100x” line getting attention? Because it reframes AI as an economics story. Huang’s argument is basically that once a service becomes software, the delivery cost drops and the addressable market explodes. A human service business scales by hiring. Software scales by replication. So when he talks about 100x, he is pointing to the jump from labor-bound output to software-bound output — a much bigger claim than ordinary productivity gains. (nvidia.com) CNBC’s clip from the May 5 interview centers that exact point about software being additive, not destructive, for enterprise vendors. ### Why does this help software companies instead of hurting them? That has been the big investor fear. If agents do the work, maybe SaaS apps get squeezed. Huang has been pushing the opposite line since at least February — that markets “got it wrong” and agentic AI is “fully accretive” for software companies. His logic is simple: agents still need systems of record, workflow engines, permissions, governance, and interfaces to act safely inside enterprises. (cnbc.com) In other words, the app layer does not disappear. It becomes the control layer. ### Why is ServiceNow central to this pitch? Because ServiceNow already sits where enterprise work gets routed, approved, tracked, and audited. That makes it a natural place to insert agents. Nvidia’s new announcements with ServiceNow are about governed autonomous agents, using Nvidia models, toolkits, and AI-Q blueprints inside ServiceNow’s workflow stack. Huang is not randomly flattering a partner here — he is pointing to the kind of software company that benefits if agents become real coworkers. (cnbc.com) ### Where does Nvidia make money in this setup? Everywhere it can. More capable agents mean more inference, more orchestration, more model serving, and more enterprise infrastructure. Huang has also been saying newer reasoning and agentic systems need far more compute than older one-shot models — in one February CNBC appearance, he said next-generation AI can require 100x more compute. So the bet is two-sided: software vendors get a new platform cycle, and Nvidia sells the picks, shovels, and increasingly the workflow plumbing too. (blogs.nvidia.com) ### What is the catch? The hard part is not making an agent demo look smart. It is making an agent reliable inside a company. Enterprises need permissions, audit trails, policy controls, handoffs, and ways to stop an agent from confidently doing the wrong thing at scale. That is why the current pitch has shifted from “cool model” to “governed system.” The control layer is the product. (cnbc.com) ### So what changed this week? The message got sharper. At Knowledge 2026, Huang tied together three ideas that had been floating around separately — agents as workers, software as the beneficiary, and industrial-scale AI as the next market. That makes this less a throwaway conference soundbite than a blueprint for where Nvidia wants the next leg of AI spending to go. ### Bottom line? (blogs.nvidia.com) Huang is telling enterprises and investors to stop thinking of agentic AI as a nicer chatbot. He wants them to see a much bigger shift — services turning into software, software turning into control systems, and compute demand rising with every step. (cnbc.com) (nvidia.com)