Huang Renxun: Gambling on AI started five years ago, and everything is just beginning.

Behind every success, there is a right beginning.

NVIDIA is no exception.

On August 8th, local time, Jensen Huang, the founder and CEO of NVIDIA, delivered a speech at SIGGRAPH, a computer graphics conference held in Los Angeles. He stated that in 2018, the company made a life-or-death business decision, which few people realized would redefine its future and contribute to reshaping a constantly evolving industry.

That industry is artificial intelligence, a future driven by NVIDIA hardware.

Huang said that five years ago, the turning point was the choice to achieve AI-driven image processing through ray tracing and intelligent upgrades, namely RTX and DLSS technologies.

We realized that rasterization (the traditional 3D scene rendering method) had reached its limits.

2018 was the moment to "bet on the company's (future)". We had to redevelop hardware, software, and algorithms. While we were reshaping CG (Computer Graphics) with AI, we were also reshaping GPUs for AI.

Although ray tracing and DLSS are still widely used in GPUs and game development, the underlying architecture of these technologies can be perfectly applied to machine learning.

It is worth mentioning that traditional data centers cannot provide the massive amount of data required for training large generative AI models. However, systems like NVIDIA H100, designed from the beginning for executing necessary operations on a large scale, can achieve this. It can be said that the development of artificial intelligence is limited in some aspects by these computing resources.

Riding the wave of large model development, NVIDIA has sparked a frenzy of AI chip purchases. However, Huang emphasized that this is just the beginning. New models not only need to undergo training but also need to be run in real-time by millions or even billions of users.

In the future, LLM (Large Language Models) will be at the forefront of almost everything: "human" is the new programming language.

Huang also stated that from visual effects to rapidly digitized manufacturing markets, factory design, and heavy industries, everything will to some extent adopt natural language prompts.

The entire factory will be software-defined robots, and the cars they are about to manufacture will be robots themselves. Therefore, robots design robots that produce robots.