NVIDIA CEO Jensen Huang has been maintaining a packed global itinerary. On January 21 local time, Huang discussed three major advancements in AI models over the past year at the World Economic Forum in Davos, Switzerland. Subsequently, he embarked on a visit to China, making appearances in Shanghai, Beijing, Shenzhen, and Taiwan, China. On February 3 local time, Huang appeared at an event hosted by industrial software company Dassault Systèmes in Houston, USA, where the theme of his speech was industrial AI.
"In the past, we spent one-third of our time on design and digitalization, and perhaps two-thirds on building the physical form. In the future, it is highly likely we will spend 100% of our time on digitalization. Even after completing design, simulation, and verification, software integration must still be done," Huang stated. He claimed that in the future, everything from design and depiction to simulation and operation will be software-defined, from a pair of tennis shoes to everything else; cars are software-defined, and the factories where robots operate are also software-defined.
On that day, NVIDIA announced a collaboration with Dassault Systèmes. The two companies will leverage virtual twin technology to jointly build an industrial AI platform. Dassault's virtual twin technology will be combined with NVIDIA's AI infrastructure and open-source accelerated software libraries to create scientifically validated industry world models for fields such as biology, materials science, engineering, and manufacturing. Beyond enabling software-defined production systems in engineering and manufacturing, this can also be used to advance the development of new molecules and next-generation materials.
Huang Renxun indicated that the fusion of their technologies will enable engineers to work on a scale 100,000 times larger than before. What they see while working will no longer be pre-rendered or offline simulations, but a virtually generated twin world in real-time. He suggested that engineers designing products, simulating in real-time within wind tunnels, and simulating real-time robot operations will bring about very significant changes over the next 5 to 10 years.
Discussing the integration of physical AI and simulation, Huang stated that AI can learn to predict physical behavior. When this process runs in real-time, it can predict on a scale over 10,000 times larger, and combining simulation and emulation in design will bring revolutionary changes. Furthermore, in factories, millions of factories can first complete production line arrangements and robot organization within a virtual twin world.
"Today, manufacturing and logistics systems are rigid, difficult to scale, and fragile," Florence Hu-Aubigny, Executive Vice President of R&D at Dassault Systèmes, told reporters during a media briefing. She stated that future factories will be defined by software-defined production systems, and the combination of physical AI and virtual twin technology will enable factories to test and reconfigure production in a virtual environment, reducing the time required for these processes from several months to just a few hours. The complexity of AI factories is even greater than that of ordinary factories; without pre-simulation, it is difficult to ensure the entire system functions correctly.
Huang Renxun also highlighted the necessity of applying these related technologies in the construction of infrastructure like AI factories. He mentioned that the world has now begun the largest-scale industrial infrastructure construction in history, with infrastructure worth trillions or even tens of trillions of dollars being built. To meet the demands of AI infrastructure construction, three industries are scaling up: chip factories, computer factories, and AI factories. These factories are extremely complex, and designing and simulating them in a virtual twin world before breaking ground can save significant amounts of time and money.
Over the past year, NVIDIA has been frequently making moves in the field of industrial AI.
Last June, NVIDIA announced the construction of the world's first industrial AI cloud in Germany, equipped with 10,000 Blackwell GPUs. In NVIDIA's envisioned scenario, cars can be designed in virtual environments, machines can be trained in virtual environments, and after optimization, deployed to operate in real-world factories. Jensen Huang proposed that every manufacturer will have two factories: one for manufacturing products, and the other for creating the intelligence that drives these products.
Comments