Explain It in 30 Seconds
Rubin is best understood as a co-designed AI supercomputing platform, not just a new GPU.
NVIDIA describes Rubin as combining six major chips:
- Vera CPU;
- Rubin GPU;
- NVLink 6 Switch;
- ConnectX-9 SuperNIC;
- BlueField-4 DPU;
- Spectrum-6 Ethernet Switch. [S7]
That list shows the direction NVIDIA is taking.
The product is increasingly the whole system, not the accelerator by itself.
Where Rubin Sits in the NVIDIA Stack
AI workloads
↓
CUDA + NVIDIA AI software
↓
DGX / rack-scale systems
↓
Vera CPU + Rubin GPU
↓
NVLink 6
↓
ConnectX + BlueField + Spectrum networking
↓
Physical data-center infrastructureWhy Rubin Exists
AI workloads continue to demand more compute, memory and communication.
But scaling by adding more GPUs creates new problems:
- processors must communicate;
- memory must feed them fast enough;
- networks must connect racks;
- software must schedule work;
- power and cooling must remain practical.
Rubin represents NVIDIA's attempt to design more of these pieces together from the beginning.
Blackwell vs. Rubin
The most useful difference is not a benchmark table.
It is the product philosophy.
Blackwell
Blackwell pushed NVIDIA further into rack-scale, highly integrated AI systems.
Rubin
Rubin takes the same idea further by explicitly co-designing six major chip categories as one platform. [S7]
The transition is:
faster GPU
→ integrated accelerator system
→ co-designed AI supercomputer platformWhy the Vera CPU Matters
Rubin pairs with NVIDIA's Vera CPU.
That matters because NVIDIA is moving beyond a model in which its GPU is simply inserted next to a processor designed by another company.
The more of the compute stack NVIDIA designs, the more tightly it can coordinate:
- CPU;
- GPU;
- interconnect;
- networking;
- software.
Why Networking Is Part of Rubin
The inclusion of ConnectX, BlueField and Spectrum components in the platform is a strong signal.
NVIDIA's networking capabilities, strengthened through Mellanox, are not peripheral accessories.
They are part of the system architecture.
Rubin and DGX
NVIDIA's current DGX SuperPOD page lists Rubin and Blackwell-powered compute options. [S10]
That makes the relationship:
Rubin = platform generation
DGX = system / infrastructure familyWho Uses Rubin?
Rubin is designed for organizations operating large AI infrastructure:
- cloud providers;
- AI labs;
- hyperscalers;
- enterprises;
- research computing;
- advanced AI factories.
Most end users will access Rubin indirectly through cloud services or applications.
What Rubin Depends On
Rubin depends on an even broader ecosystem than a standalone processor:
- advanced semiconductor manufacturing;
- high-bandwidth memory;
- networking;
- server manufacturing;
- rack integration;
- power delivery;
- cooling;
- CUDA software;
- cloud and system deployment.
NVIDIA said in May 2026 that hundreds of ecosystem partners across many factories and countries were involved in the Rubin ramp. [S8]
Why Rubin Matters to NVIDIA
Rubin makes NVIDIA's strategic direction unusually clear.
The company is trying to control the architecture of the entire AI-computing system while still relying on a broad manufacturing and deployment ecosystem.
That combination is the modern NVIDIA model:
Own more of the design
while coordinating more external manufacturing
and distribution partners.What Most People Misunderstand
"Rubin is just the next GPU."
Too narrow.
The Rubin platform explicitly includes CPU, GPU, switches, NICs, DPU and Ethernet components. [S7]
"Rubin means Blackwell is obsolete."
No.
Platform transitions take time, and Blackwell systems remain important while the next generation ramps.
Common Questions
Is Rubin shipping?
NVIDIA said Vera Rubin was ramping into full production in May 2026. [S8]
What comes before Rubin?
Blackwell.
Does Rubin use CUDA?
Rubin is part of NVIDIA's broader accelerated-computing platform and CUDA software ecosystem.