Retail pricing for Nvidia's high-end enterprise AI workstation has officially surfaced online through hardware vendor listings. The desktop supercomputer, equipped with the Grace Blackwell Ultra GB300 Superchip architecture, carries a starting cost of $94,930.
The system brings enterprise level artificial intelligence compute capabilities directly to deskside form factors. By making raw performance accessible outside traditional server racks, the workstation targets research institutions and corporate labs requiring high compute density.
First Retail Pricing Confirmed for GB300 Superchip System
Early industry expectations estimated that Nvidia's latest DGX Station desktop tower would settle around the six figure mark, though official public price tags were absent during its initial showcase. Recent online listings from system integration partner Exxact confirmed these projections, displaying the Valence VWS-158270643 configuration available for order at $94,930.
While Nvidia typically works through system integrators and enterprise quotes for its hardware deployments, this public listing offers a concrete reference point for buyers. Customers looking to secure local AI infrastructure can now evaluate direct acquisition costs without waiting for custom enterprise bidding processes.
Specifications of the Grace Blackwell Tower PC
At the center of the workstation sits the Nvidia GB300 Grace Blackwell Ultra Desktop Superchip. The silicon design integrates a 72-core Arm-based Grace CPU alongside a Blackwell Ultra GPU. The components communicate through an NVLink-C2C interconnect, offering up to 900 GB/s of bidirectional bandwidth between CPU and GPU memory pools.
The workstation features a total of 748GB of unified memory. This consists of 252GB of high-speed HBM3e VRAM attached directly to the GPU, paired with 496GB of LPDDR5x system memory connected to the CPU. Together, this unified memory structure allows developers to run complex generative models locally without experiencing bandwidth throttling across standard PCIe buses.
Nvidia GB300 DGX Station Price Listing Highlights Deskside AI Demand
The availability of the system highlights growing demand among enterprises seeking localized compute solutions. By integrating data center architecture into a deskside chassis, the station provides high processing headroom for massive large language models, vision models, and simulation tasks.
Industry analysts note that while the purchase price is significant, local execution offers long-term advantages. Operating high-throughput workloads on dedicated local hardware reduces recurring cloud API service fees and eliminates bandwidth bottlenecks during model fine-tuning. Furthermore, local deployments appeal strongly to organizations managing strict data privacy standards or proprietary information that cannot be transmitted over third-party cloud networks.
Enterprise Workstation Capabilities and Memory Overhead
Running frontier artificial intelligence models requires high memory capacity to store model weights alongside runtime parameters. Traditional workstation configurations using discrete GPUs often face limitations caused by isolated VRAM pools.
The GB300 architecture overcomes these constraints through its unified memory pool. Developers can load models containing hundreds of billions of parameters into memory directly on a single deskside computer. The system also utilizes fifth-generation Tensor Cores supporting NVFP4 precision modes, which further optimizes memory efficiency and compute throughput during active inference.
Market Context and Availability for Local AI Hardware
Nvidia has increasingly diversified its workstation ecosystem to cover various enterprise requirements. Alongside the full-sized DGX Station tower, smaller systems like the DGX Spark offer lower entry costs for lightweight development environments. However, the GB300 DGX Station remains the flagship configuration for organizations requiring top tier computational performance without building out dedicated server rooms.
Partner OEMs including Dell, HP, ASUS, Gigabyte, and Supermicro are expected to roll out customized variants based on the same reference architecture. As orders open across international distribution channels, supply availability will depend on manufacturing yields for Blackwell silicon and high-bandwidth memory supplies.
The online listing of the GB300 DGX Station marks an important milestone in bringing server-class Grace Blackwell hardware into desktop deployment, serving research institutions and enterprise developers looking for standalone AI computing power.
