“AI Factories Drive the World” — NVIDIA’s Jensen Huang on the Next Industrial Revolution
NVIDIA founder and CEO Jensen Huang delivered a keynote speech lasting approximately an hour and a half at GTC Taiwan, where he spoke extensively about the company's journey to date and its future vision. This presentation concretely illustrated the process by which NVIDIA has transformed from a “mere technology company” into an “AI infrastructure company,” becoming an indispensable presence for every industry, region, and enterprise across the globe. This article organizes the content of the lecture by key points and provides explanations for each section, incorporating specific examples and quotes.
1. Redefining the Computing Ecosystem
1-1. Data Center = Computing Unit
At the beginning, Mr. Huang declared, “Modern computers are not just PCs or servers; the data center itself is the unit of computing.” The background to NVIDIA’s leap from a company founded in 1993 targeting a mere $300 million chip market to a $1 trillion cloud data center market, and now toward a “multi-trillion dollar” industry of AI factories, lies in the paradigm shift of viewing the data center as “a single massive computing unit.”
1-2. East-West Traffic and the Mellanox Acquisition
Traditional networking was divided into “North-South” (for storage and external connections) and “East-West” (for server-to-server communication). Mr. Huang emphasized that “the most important thing is the network where servers communicate with each other in the East-West direction,” and introduced the background of the 2019 acquisition of high-speed networking company Mellanox to strengthen this.
2. NVIDIA as AI Infrastructure
2-1. The Third Infrastructure Revolution—AI
As the third infrastructure following “electricity” in the Industrial Revolution and “information” in the Internet age, Mr. Huang proposed an “infrastructure of intelligence.” He stated, “In ten years, you will realize that AI has permeated every industry and region just like electricity and the Internet,” and explained why AI infrastructure is needed worldwide.
2-2. AI Factories and Tokens
While traditional data centers were “information factories” supporting storage and ERP systems, new AI data centers are positioned as “AI factories.” The output generated here is called “tokens,” and just as manufacturing industries talk about production volume, “token production volume per quarter, per month, or even per hour” becomes a performance metric.
3. The Ecosystem Created by Software Libraries
3-1. The Core of CUDA X Libraries
Mr. Huang repeatedly stated, “Libraries are the core of NVIDIA.” A group of specialized libraries for each application—including CUDA, DNN, Megatron, TensorRT, and the latest data frame operating system “Dynamo” and machine learning framework “Warp”—accelerates the creativity of developers and partner companies, forming a massive ecosystem.
3-2. AI-Transformed Graphics—The Innovation of DLSS
DLSS (Deep Learning Super Sampling), which applies AI to real-time ray tracing, is a technology that calculates only one-tenth of each pixel and has the AI “guess” and fill in the remaining 90 percent. “AI has revolutionized graphics and supported GeForce for 30 years,” Mr. Huang said, citing the success of the GeForce RTX 50 series as an example.
4. Introducing AI to Enterprise IT
4-1. DGX Spark and DGX Station—For Individuals and Small/Medium Developers
The new product “DGX Spark” is a 1-petaflop class machine for developers who want their own “personal AI cloud” at their deskside. Furthermore, the “DGX Station,” which runs on a standard household outlet, boasts the performance to run trillion-parameter class AI models.
4-2. RTX Pro Servers and MVLink Fusion
For the enterprise, we announced the RTX Pro server, which is x86-compatible and capable of hosting AI agents while running traditional virtualized environments as they are. In addition, we introduced “MVLink Fusion,” which integrates semi-custom ASICs and third-party CPUs via “MVLink,” supporting the construction of flexible AI infrastructure.
5. The Future of Agents and Robotics
5-1. Agentic AI—The “Understand, Think, Act” Loop
Huang proposed “Agentic AI,” which digitizes the “understand → think → act” cycle that we humans perform. He stated that AI agents, acting as digital workers, will support corporate activities across a wide range of fields, including research, development, and customer support.
5-2. Applications to the Physical World—Newton and Isaac
For physical robot development, we utilize “Newton,” a physics simulation engine developed in collaboration with Google DeepMind and Disney Research. By combining it with the “Omniverse” simulator, we introduced the “Isaac” platform, which provides end-to-end support from learning in virtual space to implementation on robots.
6. Digital Twins and the Practice of Industrial AI
6-1. Large-Scale Use Cases by Taiwanese Companies
Major Taiwanese electronics companies such as TSMC, Foxconn, Pegatron, and Delta Electronics have built digital twins of their factories and production lines on NVIDIA’s Omniverse, which were praised as being “so beautiful they could be mistaken for photographs, yet everything is a simulation.” This has enabled significant reductions in design and construction costs and improvements in productivity.
7. Partnerships and the AI Ecosystem in Taiwan
At the end of the keynote, we announced “NVIDIA Constellation,” a plan to build an AI supercomputer in Taiwan in cooperation with the Taiwan government, TSMC, and Foxconn. By positioning Taiwan as the “hub of cutting-edge industry,” a vision was presented to strengthen the entire regional AI ecosystem.
Jensen Huang’s GTC Taiwan keynote demonstrated a grand vision in which NVIDIA evolves from a mere GPU company into an “AI infrastructure company,” aiming to redefine the global industrial structure through the three pillars of hardware, software, and partnerships. The creation of AI factories, which will support an “intelligence infrastructure” equivalent to electricity or the internet, will fundamentally change how we live and do business over the next decade. Right now, companies and developers are participating in this new infrastructure construction race and taking on the role of shaping the future.
Related Articles