Per-instance Ethernet bandwidth is bounded by the NIC of the underlying instance type. No separate per-public-IP throttle occurs on top of that. Network speed for each instance type is listed in its specification table. See the GPU instances and CPU instances pages.
Frontend bandwidth is provided on a shared, best-effort basis over high-bandwidth data-center and backbone networks, with Ethernet and DPU offload designed for near line-rate performance between Nodes and the edge.
For high-bandwidth distributed workloads such as multi-Node training, use the dedicated HPC interconnect (InfiniBand or equivalent) rather than the standard Ethernet path.
Contact support for help sizing a workload that needs sustained high throughput.
Administrator
What are the bandwidth limits per instance or public IP?
Last modified on June 18, 2026