Nvidia CEO Jensen Huang outlined the company's roadmap through 2027 including a new GPU platform called Rubin, a new CPU in Vera and networking gear. Huang added that Nvidia will follow an annual cadence.
Speaking at Computex in Taipei on Sunday, Huang said, "our company has a one-year rhythm. Our basic philosophy is very simple: build the entire data center scale, disaggregate and sell to you parts on a one-year rhythm."
Huang's talk served as a sequel to his GTC keynote with some highlights for the years ahead. Nvidia’s revenue has surged amid the AI boom.
- Nvidia's 10Q filing sets off customer guessing game
- Nvidia acquires Run.ai for GPU workload orchestration
- Nvidia Huang lays out big picture: Blackwell GPU platform, NVLink Switch Chip, software, genAI, simulation, ecosystem
- Nvidia today all about bigger GPUs; tomorrow it's software, NIM, AI Enterprise
At a high-level:
- 2025: Nvidia will launch the Blackwell Ultra GPU and Spectrum Ultra X800 Ethernet Switch 512-Radix. Nvidia has been pushing hard into AI networking with the launch of NVLink 5 Switch, CX8 SuperNIC, Spectrum-X800 Ethernet Switch and Quantum-X800 Switch.
- 2026: The company will launch the Rubin GPU, Vera CPU, NVLink 6 Switch, CX9 SuperNIC and X1600 InfiniBand/Ethernet Switch.
- 2027: Rubin Ultra GPU will launch along with three other-yet-to-be-named items.
In a nutshell, Nvidia is launching new platforms on even-numbered years with updates and enhancements in odd numbered years.
The move into CPUs is notable, but not unexpected. Huang said that the combination of integrated GPUs and CPUs can boost speed by 100x while only increasing power consumption by a factor of three. If this Nvidia cadence sounds familiar that's because it rhymes with how smartphone makers operate. There's a big splash in one year, followed by an incremental update and then rinse and repeat.
Here's a look at other items from Nvidia at Computex:
- Nvidia said its Nvidia ACE generative AI microservices were available. These technologies are designed to make it easy to animate and operate digital humans that can be used in multiple industries. These digital humans could be deployed across 100 million RTX AI PCs.
- Nvidia NIM inference microservices are available for download by developers. NIM was outlined at GTC.
- ASRock Rack, ASUS, GIGABYTE, Ingrasys, Inventec, Pegatron, QCT, Supermicro, Wistron and Wiwynn will all build systems based on Nvidia's latest chips for everything from cloud, data center and edge systems.
- Delta Electronics, Foxconn, Pegatron and Wistron are using Nvidia reference designs to build, simulate and operate their robotics-enhanced facilities. Huang appeared on stage with a series of robots powered by Nvidia's platforms.
