Nvidia's Computex 2025 Keynote: Blackwell, AI, and Robotics Innovations

Advancements in AI and Computing

• 💡 Nvidia is refining a software-defined, AI-enhanced Radio Access Network (RAN) stack, achieving performance on par with state-of-the-art ASICs for 5G and 6G.
• 🚀 The company is developing a quantum-classical computing platform called CUDA Q, integrating GPUs for pre- and post-processing in quantum computing.
• 🧠 Nvidia is introducing Grace Blackwell, a new computer system designed for advanced AI inference, enabling "thinking AI" and "reasoning AI" with significantly more computation.
• ⚙️ The GB300 upgrade, set for Q3, will enhance Grace Blackwell systems with 1.5x inference performance, 1.5x HBM memory, and 2x networking, all within the same physical footprint.

NVIDIA NVLink Fusion and Infrastructure

• 🔗 NVLink Fusion is announced to enable the creation of semi-custom AI infrastructure, allowing integration of custom ASICs or TPUs alongside Nvidia's GPUs and CPUs.
• 🛠️ This technology allows for the creation of large-scale, powerful computers by connecting semi-custom systems, with Nvidia providing chiplets and IP for integration.
• 🌐 The platform supports integration with custom CPUs and accelerators, enabling a mix-and-match approach for diverse AI hardware needs.

New AI Systems and Servers

• 💻 DGX Spark is introduced as a personal AI supercomputer for developers and researchers, offering one petaflop of performance and 128 GB of HBM memory.
• 🏠 DGX Station is a powerful desk-side supercomputer capable of running a one trillion parameter AI model, designed for high performance from a standard wall socket.
• 🖥️ The Nvidia RTX Pro Server is designed for enterprise AI agents, supporting text, graphics, and video modalities, and capable of running all existing applications, including games.

Robotics and Digital Twins

• 🤖 The development of humanoid robotics is highlighted as a critical area requiring massive compute for AI training, synthetic data generation, and skill learning.
• 🏭 Omniverse is presented as a platform for creating digital twins of robots, equipment, and entire factories, enabling AI to learn and operate collaboratively.
• 📈 The creation of digital twins is crucial for the $5 trillion in new plants being planned globally, ensuring efficient and cost-effective construction and operation in a robotic future.
• ✨ Nvidia is fostering a new industry around AI factories, enterprise agents, and robots, built on a unified architecture and a strong partner ecosystem.