ASUS’s new GB300 DGX workstation combines advanced Arm cores, high-bandwidth GPUs, and massive coherent memory into a single desktop unit, enabling extraordinary AI workloads at a price point near $120,000. This device challenges traditional assumptions around AI infrastructure by delivering data center-class capabilities in a compact chassis with strong reliability and expandability.
- High-throughput AI output with up to 128 concurrent streams
- Robust liquid cooling and 80 PLUS Titanium power supply improve reliability and efficiency
- Expandable architecture with PCIe slots and multi-unit memory pooling
Infrastructure signal
The ASUS ET900N G3 workstation integrates 72 Arm cores, the Nvidia Blackwell Ultra GPU, and 748GB of coherent memory, delivering a powerful AI-focused compute platform built for demanding workloads. Its enterprise-grade sealed liquid cooling system and a 1,600W 80 PLUS Titanium-certified power supply illustrate a focus on efficient thermal management and power utilization, critical for sustained high-load operation without throttling.
Networking capabilities include two 400Gbps ports that enable linking up to two GB300 stations, effectively creating a single shared memory pool close to 1.5TB. This design supports scaling AI model sizes and workload concurrency beyond typical desktop limits, underscoring a shift toward desktop devices that blur lines with data center-grade infrastructure.
Developer impact
For developers, the workstation’s ability to run dozens of AI agents in parallel without performance degradation marks a significant leap in local compute power. It supports concurrent workloads at a scale previously reserved for larger cluster setups, reducing reliance on remote cloud resources and potentially accelerating AI experiment iteration times.
The mirrored NVMe storage array enhances system reliability against drive failure, while multiple expansion slots allow customization for evolving workflows. High throughput observed during benchmarks suggests developers can efficiently handle large models and complex AI pipelines in a contained desktop setting, improving productivity and workflow simplification.
What teams should watch
Teams focusing on AI model development requiring extensive on-premises compute should consider the implications of this workstation for reducing cloud dependence. The high-cost entry ($100K–$120K) positions it for larger enterprises or research groups prioritizing performance over capex constraints, especially those managing sensitive data that limits cloud use.
Operations teams should monitor power consumption close to 1,000 watts during heavy loads and plan for robust cooling infrastructure. Software engineering teams need to plan for integration with the dual 400Gbps networking environment and take advantage of the shared-memory architecture for optimized data processing and throughput.