Huang was speaking at the company’s annual GTC (GPU Technology Conference) developer conference in San Jose, US, from March 16-19.

He cited Nvidia's newly introduced Vera Rubin supercomputer, which was built on a CPU called Vera and high-speed AI inference system developed by chip startup Groq.

According to Nvidia, the world is rapidly shifting from training models to inference, which is the process of running these models to generate results. These solutions will help optimize AI processing speed in real-time, especially in applications like agentic AI and automation.

"Finally, AI is able to do productive works. And therefore, the inflection point of inference has arrived. AI now has to think, and in order to think it has to inference. AI now has to do, and in order to do, it has to inference," Huang said. 

Nvidia is leading in developing AI chips, especially chips for data centers. However, many tech companies are developing their own AI chips. Analysts believe that Huang’s 1 trillion USD figure is not just a forecast, but a move to strengthen market confidence in long-term demand for AI infrastructure.