At Computex 2025, Nvidia opened its keynote with NVLink Fusion, a game-changing initiative that lets customers mix and match rival CPUs and custom AI chips with Nvidia’s Blackwell GPUs, dramatically broadening its AI ecosystem. Alongside, Jensen Huang unveiled the Grace Blackwell chip architecture—built on an advanced 4 nm process with chiplet design—promising up to 800 GB/s bandwidth and trillion-parameter LLM support for next-gen AI systems. To bring supercomputing power to desktops and offices, Nvidia introduced DGX Spark and DGX Station personal AI supercomputers, co-developed with OEMs like Acer, Dell, HP, and MSI, enabling seamless desktop-to-cloud AI workflows.
NVLink Fusion: An Open AI Fabric for Hybrid Systems
Nvidia’s NVLink Fusion initiative enables integration of non-Nvidia CPUs—such as Fujitsu’s and Qualcomm’s—with Blackwell GPUs, and connects Grace CPUs to partner AI accelerators from Alchip, Marvell, and MediaTek. By leveraging the high-speed NVLink interconnect, Fusion creates a unified AI computing fabric, reducing data-center silos and accelerating machine-learning training and inference. Jensen Huang emphasized that opening the AI infrastructure to competitors will drive broader ecosystem growth and help capture more of the $1 trillion data-center market.
Grace Blackwell: Powering the Next Generation of AI
The Grace Blackwell NVL72 chip fuses GPU and CPU cores in one package, offering a trillion-parameter LLM engine alongside Hopper-derived GPU acceleration for demanding scientific simulations. Built on TSMC’s cutting-edge 4 nm process, Grace Blackwell delivers up to 800 GB/s memory bandwidth and significant power-efficiency gains over prior architectures. Previewed within the GB300 AI system, coming Q3 2025, Grace Blackwell is positioned to become the backbone of hyperscale AI deployments in cloud and enterprise data centers.
DGX Spark & DGX Station: Desktop to Data-Center AI
Nvidia’s new DGX Spark and DGX Station personal supercomputers bring AI-factory capabilities to workstations, preloaded with the DGX OS and NVIDIA AI software stack. Partners include Acer, ASUS, Dell, GIGABYTE, HP, Lenovo, and MSI, ensuring global availability of compact systems tailored for researchers and developers. These systems support popular tools—PyTorch, Jupyter, and Ollama—for rapid prototyping and seamless scaling from desktop to DGX Cloud environments.
Democratizing AI: Marketplace & Partnerships
To further lower barriers, Nvidia launched an AI GPU compute marketplace, connecting developers to on-demand GPU resources from global cloud and OEM partners. Meanwhile, strategic collaborations with Foxconn, TSMC, and the Taiwanese government will yield a sovereign AI supercomputer—built with 10,000 Blackwell GPUs—to fuel national-level AI research and industrial applications. The open marketplace and sovereign supercomputer efforts signal Nvidia’s commitment to making powerful AI infrastructure accessible across industries and geographies.
Pioneering AI Physical & Robotics
Jensen Huang introduced “physical AI” capable of real-world interaction, showcased in new robotics engines—Newton for agile manipulation (launching July) and Groot for humanoid locomotion. He demonstrated AI-driven robotics prototypes that blend perception, simulation, and autonomous control, underscoring Nvidia’s vision for AI beyond screens—into factories, labs, and homes. These robotics platforms, powered by Blackwell and Grace technologies, aim to accelerate research in autonomous vehicles, industrial automation, and service robots.
Expert Insights & Market Reaction
Industry analysts view NVLink Fusion as a clever strategy to extend Nvidia’s AI leadership while appeasing customers seeking heterogeneous computing stacks. Despite a 2.6% dip in Nvidia’s stock on May 19, 2025—amid broader market headwinds—Jensen Huang’s announcements were hailed for their long-term potential to capture software and services revenue beyond hardware sales. Business Insider noted the enthusiasm around Nvidia Constellation, a new Taipei campus set to become a hub for AI collaboration and innovation.