ASUS unveils Enterprise Agentic AI Solution and full-stack agent computers at NVIDIA GTC Berlin
ASUS unveiled its Enterprise Agentic AI Solution at NVIDIA GTC Berlin on October 21, 2026, an integrated full-stack offering of infrastructure, software and services designed to help enterprises deploy, govern and scale autonomous AI agents from the data center to the developer desk and the edge.
The architecture has three layers: ASUS AI HUB, a centralized platform for fleet-wide agent registration, telemetry and compliance monitoring; a software orchestration stack combining NVIDIA AI Enterprise with the ASUS AI Software Stack; and scalable server hardware underneath.
At the rack level, ASUS introduced AI factory infrastructure powered by NVIDIA Vera Rubin NVL72, which the companies co-designed to deliver up to 30-times higher throughput per megawatt and 45-times lower token generation cost for agentic AI. For private data centers, ASUS revealed NVIDIA MGX modular servers, including the 2U XA P2N-E2 powered by Vera CPUs and the 6U liquid-cooled XA P8A-E14AXL supporting eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
Beyond the rack, ASUS launched local agent computers: the ExpertCenter Pro ET900N G3 workstation with 748GB of coherent memory on Grace Blackwell Ultra, the Ascent GX10 desktop AI supercomputer now available with 64GB or 128GB of unified memory, and ruggedized edge systems such as the PE3000N running on NVIDIA Jetson Thor.
For enterprises in finance, healthcare and the public sector, the pitch is an on-premises pathway to operationalize multi-agent workflows without sending proprietary or regulated data to public clouds, with ASUS AI HUB providing the local governance layer for agent permissions and observability.
Buyers should note the caveats: the 30x and 45x figures are vendor-projected design targets from Rubin architecture specifications, not third-party benchmarks; ASUS has published no list pricing, delivery lead times or AI HUB subscription fees, so procurement runs through custom quotes; and liquid-cooled Blackwell and Rubin clusters demand power distribution and cooling retrofits many facilities cannot support without significant investment.