Nvidia Makes Big AI Enterprise Push With New RTX Pro Servers, OEM Partnerships

The AI infrastructure giant is hoping to convince enterprises to move their workloads from CPU-based systems to GPU-accelerated infrastructure with the new RTX Pro servers that will be delivered by a wide range of vendors, including Dell Technologies, Lenovo and Cisco.

Nvidia is teaming up with a wide range of server vendors to convince enterprises to move their workloads from CPU-based systems to GPU-accelerated infrastructure with the new RTX Pro servers that are powered by its popular Blackwell architecture.

At the Computex 2025 event Monday, the AI infrastructure giant said that the RTX Pro servers, powered by the new Nvidia RTX Pro 6000 Blackwell Server Edition GPUs, can “run virtually every enterprise workload,” including those for agentic AI, physical AI, simulation, design and digital twins, without the need for liquid cooling or massive amounts of power.

[Related: Nvidia Reveals Offering To Build Semi-Custom AI Systems With Hyperscalers]

With availability starting this summer, RTX Pro servers are expected from Dell Technologies, Hewlett Packard Enterprise, Lenovo and Cisco Systems. Other participating vendors include Advantech, Asus, Foxconn, Gigabyte and Supermicro.

“With the ability to accelerate a wide range of enterprise workloads. RTX Pro Blackwell servers are ideal for enterprise data centers that require air-cooled, power-efficient platforms,” said Dion Harris, senior director of high-performance computing and AI factory solutions go-to-market at Nvidia, in a briefing last week.

These servers, according to Harris, “deliver multifold leaps in performance for enterprise AI, scientific simulation and visual computing applications.”

“These powerful servers are optimized for Nvidia AI software stacks to power AI agents and generative AI applications built with Nvidia Blueprints,” he added.

Nvidia Wants To Make Adoption Easier With Validated Designs

To make it easier for enterprises to use and manage the RTX Pro servers, Nvidia has developed two validated designs that provide recommendations for hardware configurations and the necessary software to run AI workloads.

The first design, the Enterprise AI Factory, is “for building and managing these complex systems in on-premises data centers or private clouds,” Harris said. The second, the AI Data Platform, is “for integrating enterprise storage with Nvidia accelerated computing to provide AI agents with near-real-time business insight,” he added.

The Enterprise AI Factory design is supported by a broad ecosystem that includes server vendors like Dell and HPE, orchestration software providers like Nutanix and VMware by Broadcom, AI operations software providers like Datadog and Grafana as well as IT consulting firms developing AI agents such as Accenture and Deloitte.

The AI Data Platform design, on the other hand, is supported by several storage providers, including DDN, Hitachi Vantara, IBM, Nutanix, NetApp, Pure Storage, Weka and Vast.

“This new class of storage system enables enterprises to use AI to tap into 90 percent of their data that's unstructured: the numerous reports, manuals, presentations and videos that are the knowledge base of the enterprise,” Harris said. “These AI data platforms are turnkey, fully integrated solutions that include Nvidia compute, networking and AI software.”

RTX Pro Server, RTX Pro 6000 Blackwell GPU Details

The RTX Pro servers come in numerous configurations, featuring up to eight RTX Pro 6000 Blackwell GPUs as well as Nvidia’s BlueField-3 DPUs and ConnectX-8 SuperNICs with built-in PCIe Gen 6 switches.

Called a “universal data center GPU” by Nvidia, the RTX Pro 6000 comes in a dual-slot PCIe card form factor and is the successor to Nvidia’s L40S GPU, which debuted in 2023 using the company’s previous-generation Ada Lovelace architecture.

Packing 96 GB of GDDR7 memory, more than 24,000 CUDA cores, over 750 fifth-generation tensor cores and just under 200 fourth-generation ray-tracing cores, the RTX Pro 6000 can achieve 120 teraflops of 32-bit floating-point math, 4 petaflops of peak 4-bit floating-point math (FP4) and 355 teraflops for ray-tracing calculations.

Compared with the L40S, the RTX Pro 6000’s tensor core performance is up to five times faster for “improved AI model processing times and reduced memory usage,” according to its product sheet. It also doubles the ray-triangle intersection rate “to create photoreal, physically accurate scenes and immersive 3-D designs” with Nvidia’s RTX Mega Geometry technology, which increases the number of ray-traced triangles by over 100 times.

The RTX Pro 6000 doubles the 48-GB GPU memory capacity and nearly doubles the 865-GBps memory bandwidth of the L40S. The GPU’s PCIe Gen 5 support also allows it to provide double the chip-to-chip bandwidth of the L40S’ PCIe Gen 4 connectivity.

The new GPU also comes with Nvidia’s universal multi-instance GPU feature, which allows users to virtually divide it into “four fully isolated instances, each with dedicated resources, allowing for concurrent execution of graphics and AI workloads, optimized GPU utilization, and secure isolation of different applications or users,” according to the product sheet.