Artificial intelligence is becoming a new infrastructure layer comparable to electricity and the internet, Nvidia CEO Jensen Huang told the Dreamforce audience in San Francisco this week.
Appearing alongside Salesforce CEO Marc Benioff, Huang framed the current wave of AI development as the beginning of a new industrial revolution—one that will turn companies and countries into producers and consumers of machine intelligence.
“With electricity, we can power everything. With the internet, we can find anything,” Huang said. “And now with artificial intelligence as an infrastructure layer across the planet, we can know everything and do anything.”
That vision places data centres at the heart of the transformation. The models and agents being introduced into corporate workflows depend on facilities capable of supplying the computing power, storage, networking and energy required to train and operate them.
Huang said Nvidia has evolved from a graphics processor manufacturer into a “full-stack AI infrastructure company” building large systems that function as AI factories.
Those factories, he said, turn electricity and the data stored inside enterprises into intelligence.
The description reflects Nvidia’s expanding position across the AI infrastructure stack. Its technology now supports not only the processors used to train and run models, but also networking, systems, software and models designed for industries ranging from life sciences to autonomous vehicles and robotics.
Huang predicted that every enterprise and country will eventually become an AI organization.
“You’re an AI company. I’m an AI company,” he told Benioff.
Salesforce supplied a concrete example at Dreamforce with the unveiling of Koa, a new CRM reasoning model built on Nvidia’s Nemotron technology.
Koa is designed to support enterprise tasks using the customer data, business logic and workflows held within Salesforce. It forms part of the company’s broader AIforce strategy, which connects AI interfaces with Salesforce information and applications.
Huang expects enterprise adoption to include a combination of closed models supplied by leading AI laboratories and private models customized by individual companies.
That mixed environment could increase infrastructure requirements as organizations fine-tune, deploy and operate specialized models closer to their own data. It also gives enterprises more control over how their information is used and how AI systems behave within specific workflows.
The expansion of AI infrastructure comes amid an intensifying debate over whether the industry is advancing too quickly. Anthropic CEO Dario Amodei has called for coordinated efforts to pace frontier development, while OpenAI CEO Sam Altman has backed stronger evaluations and unconditional responsibility from model developers.
Huang rejected the idea that speed and safety are inherently in conflict.
“Safety is paramount. In a lot of ways, it’s job one,” he said. “However, safety is an engineering problem.”
The Nvidia CEO said developers need controlled environments for testing complex systems and should decline to release products when they are not confident in their safety. He argued that secure infrastructure, rigorous testing and market incentives can allow the industry to keep advancing without new laws or a coordinated slowdown.
For data-centre operators, the larger message was one of continued demand. Huang said the number of tokens being generated by AI systems is rising rapidly as enterprises adopt both proprietary services and customized models.
The next phase will also extend beyond training. AI agents designed to work continuously across corporate systems will require persistent access to computing infrastructure as they monitor information, generate software and take action on behalf of employees.
Salesforce’s vision of the “agentic enterprise” is one example of that shift. Its new tools are intended to connect agents with governed business data and allow them to operate across interfaces such as Claude, Slack and Salesforce itself.
If Huang’s prediction is correct, AI infrastructure will not remain concentrated among a handful of technology companies. It will become a foundational layer beneath nearly every enterprise—and the data centres powering it will increasingly function as factories for intelligence.