Huang Renxun's Latest Speech! Open Ecosystem NVLink Fusion Supports Semi-custom AI Infrastructure; Open Source Humanoid Robot Basic Model
Nvidia CEO Jensen Huang delivered a keynote speech at Computex 2025 at 11:00 am Beijing time on May 19, focusing on the company's latest progress in data centers, enterprise AI, and robotics. During the nearly 2-hour presentation, the "Leather Swordsman" released several blockbuster news, such as:
* Partnering with computer manufacturers such as Asus and Dell to launch AI-First DGX personal computing systems
* Introduced NVLink Fusion, allowing customers and partners to create custom rack designs using NVLink
* Released a series of AI tools for humanoid robots, such as NVIDIA Isaac GR00T and GR00T-Dreams Blueprint
* Accelerate the transformation of IT infrastructure into AI factory based on NVIDIA RTX PRO
* …
This article will introduce the key information⬇️
1 AI-First DGX Personal Computing Systems
Huang Renxun announced that NVIDIA is deepening its collaboration with manufacturers such as Acer, GIGABYTE, and MSI to bring DGX Spark and DGX Station personal AI supercomputers to a wider market, bringing unprecedented performance and efficiency to developers, data scientists, and researchers around the world.
“AI has completely reshaped every layer of the computing stack — from chips to software,” Huang said. “As direct descendants of the DGX-1 system that ignited the AI revolution, DGX Spark and DGX Station were built from the ground up to drive the next generation of AI research and development.”
DGX Spark is equipped with the NVIDIA GB10 Grace Blackwell superchip and fifth-generation Tensor Cores, providing up to 1 PetaFLOP of AI computing power and 128 GB of unified memory, and can seamlessly export models to NVIDIA DGX™ Cloud or any accelerated cloud platform and data center infrastructure.
DGX Station is equipped with the NVIDIA GB300 Grace Blackwell super desktop chip, which provides up to 20 PetaFLOPs of AI performance and 784 GB of unified system memory. The system is also equipped with NVIDIA ConnectX®-8 SuperNIC, which supports up to 800 Gb/s high-speed network connection and multi-device expansion capabilities. DGX Station can be used as a desktop system for a single person to run advanced AI models on local data, or as a centralized computing node shared by multiple people.
2 NVLink Fusion
Huang Renxun said, "We are facing a dramatic change in computing architecture: for the first time in decades, data centers are facing a fundamental reconstruction - AI is being integrated into every computing platform. NVLink Fusion opens the door to the NVIDIA AI platform, enabling partners to build dedicated AI infrastructure."
Specifically, NVLink is the world's most advanced and widely used computing interconnect architecture. Enterprises can use NVLink Fusion and the extensive NVIDIA NVLink™ collaboration ecosystem to build semi-custom AI infrastructure.
MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence are among the first partners to adopt NVLink Fusion to build custom chips that scale to meet the needs of high-load applications such as model training and agent-based AI reasoning. With NVLink Fusion, CPUs from Fujitsu and Qualcomm Technologies can also be integrated with NVIDIA GPUs to build a high-performance NVIDIA AI factory.
3 Comprehensively accelerate the development of humanoid robots
Huang Renxun believes that "Physical AI and robots will lead the next industrial revolution. From AI brains built for robots, to simulated worlds used for training, to AI supercomputers for basic model training, NVIDIA provides the building blocks needed for the entire cycle of robot development."
He introduced the open source, fully customizable humanoid robot base model NVIDIA Isaac GR00T N1.5. This version update significantly improves the robot's environmental adaptability, workspace configuration recognition ability, and object recognition capabilities, especially when performing common material handling and manufacturing tasks, such as sorting and storing objects.
Isaac GR00T-Dreams can generate new videos from a single image input, showing the robot performing new tasks in a new environment. Action tokens are then extracted from these videos - compressed data fragments that can be understood and learned by the robot. This technology solves the problem of insufficient data in the development of Physical AI, allowing robots to quickly learn new behavior patterns.
4 Accelerate the transformation of IT infrastructure into AI factory
Huang Renxun said, "AI is revolutionizing every industry - every company will build or rent an AI factory to run its business and empower its products with intelligence. With our global partner ecosystem, we are helping companies integrate AI into their workforce, automate factories, and build AI-native products."
He announced that through the new NVIDIA RTX PRO™ server and Enterprise AI Factory Validated Design, the transformation of trillion-dollar IT infrastructure into enterprise AI factories is being accelerated.
Among them, the RTX PRO server is equipped with the NVIDIA RTX PRO 6000 Blackwell server version GPU, which extends the leading performance and energy efficiency of the NVIDIA Blackwell architecture to the data center. It can run almost all enterprise-level workloads and promote the comprehensive transformation from traditional CPU systems to efficient GPU acceleration infrastructure.
References: