ASUS Adds 64GB Memory Option to Ascent GX10 AI System
ASUS has expanded its Ascent GX10 compact AI computer line with a 64GB memory variant. The new model keeps the NVIDIA GB10 Grace Blackwell Superchip and the same up to 1 PFLOP FP4 AI performance and 273GB/s memory bandwidth as the 128GB version. ASUS positions the 64GB configuration for smaller local models and autonomous agent workloads.
ASUS’s 64GB Ascent GX10 sits beside the existing 128GB model, sharing the same NVIDIA GB10 Grace Blackwell Superchip, up to 1 PFLOP FP4 AI throughput, and 273GB/s memory bandwidth. Announced on October 2, 2026, it is not a new hardware generation; the 2025-launched system simply gains a lower-memory choice. ASUS frames the 64GB unit for open models around 30–35B parameters and local agent tasks, while the 128GB version can handle larger frameworks up to 200B parameters.
The GX10 line is marketed as a compact agent workstation rather than only an AI supercomputer. Its messaging highlights NVIDIA NemoClaw and OpenShell, isolated execution, privacy controls, and clustering multiple systems. Running workflows locally lets developers test on their own hardware instead of relying on remote services. Still, local operation alone does not guarantee privacy or security; teams must manage permissions, separate workloads, protect stored data, and limit agent tool access.
Count expanded: first para ~65? Let's count. "ASUS’s 64GB Ascent GX10 sits beside the existing 128GB model, sharing the same NVIDIA GB10 Grace Blackwell Superchip, up to 1 PFLOP FP4 AI throughput, and 273GB/s memory bandwidth." 30? Announced on October 2, 2026, it is not a new hardware generation; the 2025-launched system simply gains a lower-memory choice. 22. ASUS frames the 64GB unit for open models around 30–35B parameters and local agent tasks, while the 128GB version can handle larger frameworks up to 200B param
" labels. No extra. Need maybe blank line between sections? Fine. The expanded has two paragraphs. Under 120 total. Let's count final expanded exactly. Final expanded: ASUS’s 64GB Ascent GX10 joins the 128GB version, both using NVIDIA’s GB10 Grace Blackwell Superchip, up to 1 PFLOP FP4 AI performance, and 273GB/s bandwidth. Announced October 2, 2026, it is not a new generation; the 2025 system gains a lower-memory option. ASUS targets open 30–35B-parameter models and local agent tasks, while 128GB suits larger frameworks up to 200B parameters. The line is pitched as a compact agent workstation, with NVIDIA NemoClaw and OpenShell, isolated execution, privacy settings, and clustering several systems. Local workflows let developers test on their own hardware rather than remote services. Yet local processing alone does not ensure privacy or security; permissions, workload separation, stored-data safeguards, and tool-access limits remain necessary. Count first para: ASUS’s(1) 64GB2 Ascen