Nvidia founder and CEO Jensen Huang was at the Gartner Symposium this week to share his approach to leadership and encourage the audience to try AI applications and improve their infrastructure in accelerated computing.
“I’ve had the benefit of long experimentation,” Huang told Gartner Fellow Daryl Plummer, noting that he’s been doing his job for nearly 33 years, longer than any tech CEO.
He has been able to experiment with management techniques and has argued that a long-term perspective is essential. Fourteen years ago, for example, Nvidia first saw deep learning emerge as a workload that needed to be accelerated by its GPUs.
“Researchers came to us and asked how we could help, and we created perhaps one of the most important domain-specific libraries the world has ever seen called Cu-DNN, which accelerated neural networks,” Huang said. That runtime made it possible for each framework to be built on top of Nvidia’s CUDA architecture and the many libraries that run on top of it.
Importantly, Nvidia saw something that most others did not, which is that AlexNet was much more important and deeper than the initial applications of computer vision. Nvidia thought it was a highly scalable technique that enabled it to approximate almost any function. If he could discover a universal function that changed the way he developed and used software, the whole stack would be reinvented. That turned out to be true, Huang said.
“The leadership moment is when you see something impactful and something unexpected, and you have to ask yourself, what does this mean and what is the long-term impact?” Huang said.
Nvidia disrupted everything to the point where it realized that every aspect of computing would be changed, but only if the company took action.
“It’s surprisingly easier to live in the future than to live in the past,” Huang said, noting that once Nvidia was able to imagine what the future looked like, it could make it happen.
He noted that Nvidia has amazing computer scientists and the will to make this happen. There were things the company had to learn, but he said, “You can’t learn anything unless you lean forward and decide to do it.” The hardest part is deciding to do something, even though you know you will make mistakes and get hurt.
Gartner Fellow Daryl Plummer and Jensen Huang (Credit: Michael J. Miller)
Huang talked about how his firm moved from coding software to machine learning software to AI. He said the first thing Nvidia built were its software development tools, which led to AI systems that can learn how to do things by observing them. Now, he said, there are AI systems that are learning how to reason by taking a problem statement and breaking it down into tasks. This has created a multi-trillion dollar industry called AI.
He also talked about building systems that are really good at transforming the raw material – data – into this new invisible thing that is earned by millions of tokens per dollar. (He explained that arguments are floating-point numbers—often equivalent to about three-quarters of a word—that can be reconstructed into language, video, or images. And that will lead to physical things like robotic articulation or tagged versions of proteins. and chemicals.
“This is the beginning of a new Industrial Revolution,” he said, not unlike 300 years ago when someone created the Dynamo, which produced electricity.
The question for all of us is how this change manifests itself in all our companies. At Nvidia, he said, it was first designing tools to create AI, and then creating tools to help design chips, software and supply chain management. The company plans to have 50,000 employees with over 100 million AI assistants and suggests other companies do the same.
Recommended by our Editors
Huang currently uses multiple AIs such as ChatGPT, Perplexity and Notebook LM, and suggested putting key conference notes into the system and summarizing them.
He ended by talking about the concept of an “AI Factory”. The reason AI started in the cloud was that it required a reinvention of the compute stack, and that was easier to deliver as a service. But now, he said, all that capacity has been repurposed for on-premise use, and we need to turn every platform into an AI platform. The first step is to vectorize the database so that it can do regression augmented generation (RAG). The next step, he said, will be “agent AI,” and that too is being designed to run on-premises as well as in the cloud.
Huang predicted that we will soon have digital workers who will have to be onboarded and learn new skills, corporate values and culture taxonomy. Like humans, they will be valued and have guardrails.
First we need to create more AI jobs and then we can enable more human jobs, because more AI jobs will create more productive companies with higher profits, which will allow them to hire more many people. He noted that AIs may be able to do five, 10, 20, or even 80% of a given job, but not all.
Huang concluded by urging CIOs to build accelerated data centers. If you track the data, you’ll see that data processing continues to roughly double every year, but Moore’s Law—the doubling of CPU performance—started to slow down about 10 years ago. Therefore, he said, if your demand for computing continues to grow exponentially, but general-purpose computing does not, you should expect inflation in the cost of computing and rising energy costs. Therefore, you should use acceleration (by which he mostly meant GPUs) wherever you can, from video transcoding to SQL processing to weather forecasting.
This new paradigm shift is happening across all industries, Huang said, and “the most important thing is just to get started.”
Get our best stories!
Register for What’s new now? to get our top stories delivered to your inbox every morning.
This newsletter may contain advertisements, deals or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You can unsubscribe from newsletters at any time.
About Michael J. Miller
Former editor-in-chief
