Nvidia’s Jensen Huang Says AI Agents Have Sent Token, GPU Demand ‘Through the Roof’: Gene Munster Sees 15 Million Bot Users Soaring to Billions

Nvidia Corp. (NASDAQ:NVDA) CEO Jensen Huang stated that token consumption and graphics processing unit demand have gone “through the roof” due to artificial intelligence agents. Deepwater Asset Management Managing Partner Gene Munster supported this statement, predicting that daily bot users could reach the billions in the coming years and drive compute demand.

Gene Munster Backs Bullish Outlook

Munster echoed Huang’s comments, stating that the rise of personalized agents is “one of the clearer reasons compute demand should keep ramping for the next few years”. Munster estimated that agents can consume “10-15x the tokens of a chatbot.”

Currently, about 15 million consumers use bots daily, Munster noted. However, he predicted that “in a few years that could be billions,” which would fuel sustained growth for Nvidia.

Jensen highlights that personalized agents (Muse, Grok Bot) will push token usage "through the roof."

My take: One of the clearer reasons compute demand should keep ramping for the next few years. Agents can consume 10-15x the tokens of a chatbot.
Today roughly 15m consumers use…

— Gene Munster (@munster_gene) October 21, 2026

Chief Market Strategist Shay Boloor similarly noted that agents already consume approximately five times more tokens than human users on the OpenRouter platform, stating that “AI compute demand will keep growing for years”.

$NVDA CEO Jensen Huang says personalized AI agents like $META Muse and $SPCX Grok Bot will send token usage "through the roof" as agents already consume ~5x more tokens than human users on OpenRouter.

AI compute demand will keep growing for years. pic.twitter.com/1HzV4gUwWC

— Shay Boloor (@StockSavvyShay) October 21, 2026

Agents Drive Token Usage

During his keynote address in Berlin on Oct. 21, Huang discussed the emergence of personalized AI agents, specifically highlighting “the incredible Meta Muse, Grok bot, and of course Gemini.” Huang explained that agents “think, reason, iterate, use tools, and as a result consume an enormous amount of tokens.”

Because agents iterate to solve complex problems, they become more capable and are adopted by a wider user base. As agents must “use a lot of tokens to think, the amount of token consumption has gone through the roof,” Huang said. He added that “GPU demand has really gone through the roof” as a direct result.

The Architecture of AI Agents

Huang outlined five elements of an AI agent: a model, an exoskeleton that orchestrates memory and tools, a sandbox that provides isolation, knowledge and tools. He characterized the harness as an “exoskeleton” that orchestrates work and allows the agent to engage working memory and long-term memory.

This architecture requires dedicated computing power for processing the large language model, managing the agent’s knowledge data, and executing the tools it needs to perform tasks autonomously. He noted that the industry is shifting from simply consuming AI to manufacturing it within massive data centers, which he referred to as AI factories.

How Has NVDA Performed in 2026?

The stock was 2.41% lower in premarket trading on Thursday. It advanced by 22.53% over the last year, up 19.02% year-to-date, and down 2.38% over the last month. The stock closed up 2.54% at $221.97 on Wednesday.

Benzinga’s Edge Stock Rankings indicate that NVDA maintains a weak price trend in the short term but a strong trend in the long and medium terms, with a poor value score.

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.

Photo courtesy: Shutterstock