News
NVLink 6 switch architecture supports 400G bidirectional bandwidth
Monday, December 22, 2025 at 11:20 AM
The NVLink 6 switch reportedly supports 200G bidirectional bandwidth, reaching a total of 400G for next-generation data center interconnects.
Context
NVIDIA is doubling its data interconnect performance with the NVLink 6 switch architecture, a critical component of the upcoming Rubin GPU platform. The new architecture supports 400G bidirectional bandwidth per link, effectively doubling the speeds provided by the current Blackwell generation’s NVLink 5. This technological leap enables a total per-GPU bandwidth of 3.6 TB/s, a crucial threshold for removing communication bottlenecks in the massive AI clusters used to train and run trillion-parameter models.
This advancement reinforces NVIDIA’s dominance in the AI supply chain by addressing the escalating demand for high-speed, low-latency GPU-to-GPU communication. The transition to NVLink 6 and the Rubin platform, expected to enter the market in 2026, follows the Blackwell Ultra refresh scheduled for late 2025. For investors, this maintains the company’s aggressive annual product cadence and secures its competitive advantage in the high-end data center and AI factory markets.
Sources (2)
Related Companies
Nvidia
NVDA