The new 2U rack-mounted systems, offered by the likes of Cisco, HPE, Lenovo and Supermicro, are pitched as universal acceleration engines for workloads from agentic AI and scientific simulation to video processing and industrial modelling.

Nvidia says dual-GPU configurations can deliver up to 45 times the performance of CPU-only 2U servers while using 18 times less energy, trimming ownership costs for power- and cooling-strapped data centres.

The kit slots into the RTX PRO Server family unveiled in May at Computex, which already includes models housing two, four or eight RTX PRO 6000 Blackwell GPUs. The new “mainstream” additions are designed to make Blackwell silicon more accessible to enterprises refreshing millions of servers each year as AI demands grow.

Nvidia flagged Dell’s updated AI Data Platform, built on its own reference design and now bundled with Dell PowerEdge R7725 servers carrying a pair of RTX PRO 6000 GPUs, Nvidia AI Enterprise software and networking gear.

The move is part of a broader push to turn data centres into AI factories without blowing budgets on floor space or cooling bills.