Huang emphasized that energy constraints are the core physical boundary of current AI development, and the upper limit of all calculations is ultimately constrained by bit flipping and the energy required for information transmission.

"We are far from reaching those fundamental bottlenecks that truly limit development," he said. At the same time, our task is to build a more energy-efficient computing platform.

Meanwhile, Huang mentioned that improving the energy efficiency of AI computing is NVIDIA's current priority. He emphasized that since 2016, the energy efficiency of AI computing has increased by 10,000 times, and this progress is comparable to the technological singularity in the automotive and lighting industries in terms of energy density improvement. To build smarter systems must be based on energy efficiency.