ASUS bundles AI agent management, software and servers for enterprises

ASUS introduced its Enterprise Agentic AI Solution at NVIDIA GTC Berlin on October 21, 2026, combining infrastructure, software and services in one enterprise offering. The company is pitching a path from developing AI agents to running and overseeing them in production, with management tools alongside the machines that support their multistep work.

Managing agents versus running their workloads

The package separates agent oversight from workflow coordination and computing capacity. ASUS describes three layers, each with a different responsibility: keeping track of agents, connecting their work to enterprise workflows, and supplying the underlying servers. The distinction puts governance inside the offering rather than presenting it solely as a hardware purchase.

• Agent oversight: ASUS AI HUB handles registration, governance and monitoring across an organization’s fleet of agents.
• Workflow coordination: NVIDIA AI Enterprise and the ASUS AI Software Stack provide orchestration—the coordination of work—and integration with enterprise workflows.
• Computing: ASUS servers and infrastructure supply the hardware layer, including an AI factory platform powered by NVIDIA Vera Rubin NVL72.

ASUS says customers configure deployments around model size, performance targets and operational requirements. It is targeting finance, healthcare, manufacturing, the public sector and other regulated industries. That configuration process is the company’s proposed bridge between the integrated package and the requirements of a particular organization, rather than a single fixed server specification.

Rack capacity and local development serve different jobs

The supporting server showcase includes two contrasting designs. The XA P2N-E2 is a 2U server powered by NVIDIA Vera CPUs. The larger XA P8A-E14AXL occupies 6U of rack space and supports eight liquid-cooled NVIDIA RTX PRO 6000 Blackwell Server Edition graphics processors. These are distinct hardware options within the enterprise portfolio, not one uniform machine.

On the desktop, ASUS positions the ExpertCenter Pro ET900N G3 for developing, training and deploying models on an organization’s own premises. It uses NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip, 748GB of coherent memory and the NVIDIA AI software stack. ASUS presents local data control as part of that workstation’s appeal.

The Ascent GX10 offers another desktop route, based on NVIDIA DGX Spark. ASUS lists 64GB and 128GB unified-memory configurations and says the system supports models ranging from approximately 30–35 billion parameters to as many as 200 billion. It also says up to four systems can be clustered using NVIDIA Sync Cluster Assistant. Those are ASUS’s stated capabilities, not independent workload measurements.

For work beyond the desk or data center, ASUS showed the PE3000N, powered by an NVIDIA Jetson T5000 module built on Jetson Thor. The company specifies 128GB of memory and up to 2,070 FP4 TFLOPS, a measure of low-precision computing performance. Its edge showcase targets robotics, autonomous machines and intelligent video analysis.

Server availability is not a package price

ASUS’s distribution release says its server products are available worldwide and directs buyers to local representatives. It does not disclose pricing for the integrated solution or the showcased products. The worldwide statement applies to server products; it is not a dated rollout commitment for every software and service component in the new offering.

For attendees, ASUS lists booth 3039 at Station Berlin as the main exhibition location. Its separate VIP Room at W3 Hub is open October 20–22, with a closer look at its edge-to-cloud portfolio, including enterprise infrastructure based on NVIDIA HGX and RTX PRO.