DGX Spark and Spectrum-X: Breaking Down NVIDIA GTC 2025
Release date: 24 March 2025
A blog by Allan Kaye, CEO and co-founder at Vespertec
The eyes of the tech world were on Jensen Huang during his keynote speech at GTC 2025. Living up to the expectations was always going to be hard, but somehow, he managed to surpass them. NVIDIA has firmly set itself up as the foundation of tomorrow’s AI factories with a series of much-anticipated, enterprise-scale systems—but at the other end of the infrastructure scale are DGX Spark and Station.
Personal AI Supercomputers
NVIDIA Announces DGX Spark and DGX Station Personal AI Computers / Credit: NVIDIA
Think about what a university or corporate researcher could achieve with the world’s smallest personal AI supercomputer on their desk. Prototyping, fine-tuning, and running large models from the comfort of their workstation. This could cause the same explosion in adoption and experimentation that the Altair 8800 or Commodore PET did for home computing.
AI Factories and Networking
On the enterprise side, leaders are eyeing AI factories—specialist infrastructure that unites the data lifecycle and generates real-time insights—as the key to returning on their AI investments. Refining raw materials into intelligence. Businesses put in data and electricity and out comes information they can use to power AI agents and other revenue-driving applications. But one factor often gets overlooked here. Networking.
Huang has urged us to think about data centres as a single ‘unit of compute’. That means considering compute, memory, and storage as one system—and data centres as more than the sum of their parts. To conduct all these separate players effectively, you need a low-latency, high-performance network.
One of NVIDIA’s greatest innovations was the NVLink switch, allowing organisations to interconnect dozens of GPUs to act as a single GPU. Now, NVIDIA might have outdone itself. Its new Spectrum-X Silicon Photonics Ethernet switch will boost data transfer speed to 1.6 Tb/s per port, potentially bringing millions of GPUs together in harmony.
NVIDIA Spectrum-X™ and NVIDIA Quantum-X silicon photonics networking switches / Credit: NVIDIA
Server Innovations: Vera Rubin, Blackwell Ultra, and Olympus
The headlines will, of course, be full of the Blackwell Ultra NVL72 and Vera Rubin NVL144 unveilings – followed by the Rubin Ultra NVL576, which will feature 576 GPUs when released in 2027. It will be particularly interesting to see how the custom Olympus-based CPU architecture fares in comparison to the current, off-the-shelf Arm generation. These servers will make staggering breakthroughs in performance. But enterprises aren’t just buying one rack; some purchase hundreds of thousands of GPUs—and need a way for them to work together.
Even the less flashy announcements would, on any other day, have given us weeks’ worth of headlines. For example, the scale and power of systems like the GB300 means some businesses must make their first forays into liquid cooling. NVIDIA’s ecosystem now offers organisations a whole new world of possibilities. Experimentation is key. What really excites me isn’t just what was announced at GTC, but what customers do with it.
What did you think about GTC 2025? If you have any questions about the products announced at this year’s flagship event, please feel free to contact us or give us a ring at +44 161 947 4321.
