In This Article:
If you missed Nvidia’s (NVDA) keynote at Computex 2025, here’s the short version: AI is no longer just a tool — it’s the infrastructure of the future. Robots, desktops, smart cities, physics engines, and even national computing grids — all of it, Nvidia says, will be built on artificial intelligence.
Speaking in Taipei at Asia’s largest electronics expo, CEO Jensen Huang laid out a sweeping vision of the company’s future — one where Nvidia isn’t just enabling AI, it’s helping build a world run by it.
Leading the announcements was the DGX Spark, a compact AI supercomputer built for individual developers, researchers, and students. Think of it as a desktop-sized version of Nvidia’s massive Blackwell-based systems, with enough power to train and fine-tune large AI models locally — no cloud access required.
It’s the plug-and-play engine for what Huang called “AI-native” users — those who want high-performance AI development at their fingertips. Major OEMs such as Dell (DELL), ASUS, Lenovo, MSI, and Gigabyte are already lined up to offer Spark systems tailored for different segments of the market.
That’s just one node in Nvidia’s broader push toward what Huang calls “AI factories” — purpose-built computing environments where massive amounts of data are processed and AI-based systems are created, refined, and deployed.
To support this expansive vision, Nvidia launched something Huang said was “very special” and “one incredible ingredient” that has been missing: NVLink Fusion, an initiative that allows its GPUs to work seamlessly with third-party CPUs and custom AI chips. That means companies such as Qualcomm (QCOM), Fujitsu, Marvell (MRVL), and Alchip Technologies can plug into Nvidia’s ecosystem — a notable shift for a company long known for its tightly integrated stacks.
“I’ll let all of our partners price it for themselves, but one thing’s for sure: Everybody can have one for Christmas,” Huang said.
But this wasn’t just a cloud-computing conversation. It was one about bringing AI into the physical world.
Nvidia doubled down on “physical AI” — the idea that machines should understand and interact with the laws of the real world. Huang talked about Isaac Groot N1 (announced in March), which is a foundational model for robotics that gives humanoid robots the ability to reason, plan, and perceive their surroundings.
“The age of generalist robotics has arrived with breakthroughs in mechatronics, physical AI, and embedded computing — just in time as labor shortages limit worldwide industrial growth,” Huang said.
One challenge in robotics today is data scarcity, especially high-quality, physical interaction data. Nvidia hopes to answer this problem with Cosmos, a photorealistic simulator that creates synthetic data by observing human behavior in virtual environments. Isaac Groot N1, combined with Cosmos, means developers could now create massive amounts of training data in virtual worlds before rolling out systems in the real one.