Nvidia Posts Record $82B Quarter as Agentic AI Arrives

Nvidia Posts Record B Quarter as Agentic AI Arrives

Nvidia Posts Record $82B Quarter as Agentic AI Arrives

https://www.pymnts.com/artificial-intelligence-2/2026/nvidia-posts-record-82b-quarter-as-agentic-ai-arrives/

Publish Date: 2026-05-20 22:09:00

Source Domain: www.pymnts.com

For years, artificial intelligence in business meant a feature. A smarter search bar. A fraud score. A recommendation engine tucked inside a larger product.

What’s arriving now is different. AI agents take instructions, break them into steps and finish the job. A finance team’s AI agent doesn’t flag an anomaly and wait for a human to investigate. It investigates, pulls the relevant records, drafts the memo and routes it for review. The compute infrastructure required to do that is nothing like what powered the last generation of AI.

“Agentic AI has arrived,” Nvidia CEO Jensen Huang said on the company’s fiscal first-quarter earnings call. “AI can now do productive and valuable work. Tokens are now profitable.” Infrastructure that is powering agents completing real work is no longer discretionary spending. It carries a return on every task it finishes.

Two Layers Behind AI Workflow

An AI agent doing real work inside a business operates in two layers. One handles reasoning. The other handles execution.

“All of the thinking happens on GPUs,” Huang said. “All of the orchestration essentially runs on CPUs. If the AI were to do a search, use a browser, that would run on the CPU.”

An agent processing a chargeback reasons through the evidence on one layer of hardware. It then pulls the transaction records, files the response and updates the case log on a second layer. The reasoning layer has existed for several years. The execution layer is what’s new. No chip was built for it until now.

Advertisement: Scroll to Continue

Nvidia’s Vera processor is designed specifically for that execution layer. Prior chips were built to rent processing capacity to many users at once, because that was how cloud economics worked. Agents don’t rent capacity. They need a task completed as fast as possible at the lowest cost per completed action.

“The economics of AI, of the future, is tokens per dollar,” Huang said. Nvidia told investors it expects…

Source