Nvidia launches Blackwell Ultra architecture and AI personal supercomputer, and releases first customizable open robot model
NVIDIA has set new records in the field of AI and brought revolutionary technological breakthroughs.
Core content:
1. Blackwell Ultra architecture: powerful infrastructure in the era of AI reasoning
2. AI personal supercomputer: democratizing AI computing and protecting data privacy
3. The world's first customizable open robot model GR00T N1: promoting the development of robotics technology
NVIDIA released a new generation of AI technology at the 2025 GTC conference, including an enhanced Blackwell Ultra architecture, an AI supercomputer for individual users, and the world's first customizable open robot model, further consolidating its leadership in the field of artificial intelligence.
Blackwell Ultra: Infrastructure for the AI Reasoning Era
NVIDIA has launched an upgraded version of the Blackwell architecture, Blackwell Ultra, laying the foundation for the AI reasoning era. This new architecture has achieved major breakthroughs based on the original Blackwell:
The GB300 NVL72 system connects 72 Blackwell Ultra GPUs and 36 Arm Neoverse-based NVIDIA Grace CPUs in a rack-scale design, delivering 1.5 times the AI performance of the GB200 NVL72
- NVIDIA HGX B300 NVL16 system is 11 times faster than the previous generation Hopper architecture for large language model inference, with 7 times more compute power and 4 times more memory capacity
Blackwell Ultra brings 50 times the revenue opportunity to AI factories compared to Hopper architecture
“AI has made a huge leap forward – requiring orders of magnitude more computing performance for inference and agent AI,” said Jen-Hsun Huang, founder and CEO of NVIDIA. “We designed Blackwell Ultra for this moment – a versatile platform that makes it easy and efficient to perform pre-training, post-training and inference AI inference.”
"Personal Supercomputer in the AI Era": Putting Powerful AI at Your Fingertips
At this year's GTC conference, NVIDIA launched two "AI-era personal supercomputers" for personal and professional users, marking an important step in the democratization of AI computing:
-These devices use the latest B300 series GPUs to efficiently run large AI models locally
-Individual users can run advanced models including Llama Nemotron Reason on their devices
- Supported by the NVIDIA AI Enterprise software platform, users can use NVIDIA NIM microservices and a complete AI development tool chain
These personal AI supercomputers will enable developers, creators, and researchers to develop and apply AI directly on their own devices without relying on cloud services while maintaining data privacy.
The world's first customizable open robot model
Another big news at GTC 2025 is that NVIDIA launched the world's first customizable open robot function model GR00T N1:
- GR00T N1 is a basic robot model designed for physical AI and robotics applications
- Developers can perform custom training and optimization based on this model to accelerate the development of robotics technology
-The model supports multimodal input and can understand visual, language, and physical interaction data
- Through open release, the threshold for developing robotics technology will be greatly reduced
Nvidia also announced a collaboration with Google's DeepMind and Disney to develop Newton, an open source physics engine for robot simulation, which is expected to be completed later this year. Huang Renxun said: "The era of general-purpose robots has arrived. With the help of Nvidia Isaac GR00T N1 and new data generation and robot learning frameworks, robot developers around the world will open up the next frontier of the AI era."
Enhanced AI Inference Software and Network Infrastructure
To support these hardware innovations, NVIDIA is also introducing new software and networking technologies:
- NVIDIA Dynamo open source inference framework, which can scale inference-based AI services, improve throughput, reduce response time and model serving costs
-Supports Spectrum-X Ethernet and Quantum-X800 InfiniBand platforms, providing 800 Gb/s data throughput per GPU system
- BlueField-3 DPU supports multi-tenant networking, GPU computing elasticity, accelerated data access and real-time network security threat detection
The full range of product portfolio released by NVIDIA this time will accelerate the popularization of AI from the cloud to the edge, while creating new possibilities for reasoning AI, intelligent AI and physical AI, and pushing AI technology into the next stage of development.
