NVIDIA DGX Station for Windows brings Blackwell Ultra AI power to enterprise desktops
NVIDIA is expanding its DGX Station lineup with a Windows version, giving enterprise developers a powerful local AI workstation built around the GB300 “Blackwell Ultra” chip. The move is designed to bring high-end AI computing into the Windows ecosystem, where many large companies already run their daily productivity tools, development platforms, and business applications.
Until now, the upgraded DGX Station was mainly focused on Linux-based environments. That created a familiar challenge for enterprise AI teams: use Linux for demanding AI workloads, but rely on Windows for everyday workflows. For developers working on large language models, mixture-of-experts models, and multi-agent AI systems, switching between environments can slow down projects and add unnecessary complexity.
With DGX Station for Windows, NVIDIA is aiming to close that gap. The system brings serious AI acceleration to a familiar desktop platform, making it easier for companies standardized on Windows to build, test, and deploy advanced AI models locally.
At the heart of the new DGX Station is NVIDIA’s GB300 Blackwell Ultra GPU. This accelerator is built for demanding AI workloads and delivers up to 20 PFLOPs of AI compute, giving developers the performance needed for next-generation model development, inference, fine-tuning, and agent-based AI applications.
The GB300 GPU includes 160 streaming multiprocessors, with 20,480 CUDA cores and 640 fifth-generation Tensor cores. These Tensor cores support newer low-precision formats, including FP8, FP6, and NVFP4, helping improve AI throughput while reducing memory usage. NVIDIA says the NVFP4 format can deliver near-FP8 accuracy, with differences often below 1%, while cutting memory footprint significantly compared with FP8 and FP16.
Memory is one of the biggest upgrades in the Blackwell Ultra platform. The DGX Station features 252 GB of HBM3e memory with 7.1 TB/s of bandwidth. That gives large AI models more room to operate and helps reduce bottlenecks when working with massive datasets, large context windows, and complex model architectures.
The workstation also includes a Grace CPU with 72 cores based on Arm Neoverse V2 architecture. Alongside the GPU’s HBM3e memory, the system carries 496 GB of LPDDR5X memory, bringing the total coherent CPU-GPU memory pool to 748 GB. The Grace CPU and GB300 GPU are connected through NVIDIA’s NVLink-C2C interconnect, which provides up to 900 GB/s of bandwidth.
For networking, the DGX Station includes a ConnectX-8 SuperNIC capable of up to 800 Gb/s. This makes the system better suited for AI teams that need fast data movement between workstations, storage systems, and data center infrastructure.
The platform is built as a serious high-performance workstation. It has a 1600W power envelope, four M.2 Gen5 slots, multiple Ethernet connections, and PCIe Gen5 expansion support. It can also support NVIDIA’s latest RTX PRO Blackwell graphics cards, giving developers more flexibility depending on workload requirements.
The arrival of DGX Station for Windows also highlights the growing competition in premium AI workstations. AMD has introduced its Threadripper Halo Station concept, combining Threadripper Pro CPUs with Radeon PRO and Instinct-class GPU options. That system is expected to focus on high CPU core counts, large DDR5 memory capacity, and AMD’s ROCm software ecosystem.
NVIDIA’s approach is different. DGX Station is centered on a tightly integrated Grace and Blackwell Ultra design, with CUDA, TensorRT, NVLink-C2C, high-bandwidth HBM3e memory, and a unified memory architecture built for AI acceleration. For developers already invested in NVIDIA’s AI software stack, this could make the Windows version especially appealing.
Pricing for the DGX Station is expected to be around the $100,000 range, placing it firmly in the professional and enterprise workstation category. NVIDIA has not confirmed an exact launch date for the Windows version, but it is planned for release in the fourth quarter.
The bigger story is clear: local AI computing is becoming increasingly important. As businesses build more advanced AI tools, they need systems that can handle large models without depending entirely on cloud infrastructure. Workstations like the NVIDIA DGX Station for Windows are designed to bring data center-class AI performance closer to developers, researchers, and enterprise teams.
From compact AI systems to full-scale desktop workstations, hardware companies are racing to meet demand for local AI compute. NVIDIA’s Windows-based DGX Station shows that the next wave of AI development will not only happen in cloud servers, but also on powerful machines sitting directly inside enterprise offices and research labs.
