ASUS is expanding its Ascent GX10 AI supercomputer lineup with a new 64GB configuration designed to make local agentic AI development more accessible to developers, researchers, and data scientists.

The new model joins the existing 128GB ASUS Ascent GX10, giving buyers a lower-memory option built around the same NVIDIA DGX Spark platform and NVIDIA GB10 Grace Blackwell Superchip. ASUS says the 64GB version is designed to run today’s capable 30–35-billion-parameter open-source AI models and autonomous AI agents entirely on-device.

The announcement was made on October 2, 2026, in Taiwan, with ASUS publishing the press release on October 5.

ASUS Ascent GX10 64GB targets local agentic AI

The Ascent GX10 64GB is aimed at developers who want to build and run AI agents locally rather than relying entirely on cloud-based inference.

ASUS says keeping models, prompts, and inference on-device can help organizations keep sensitive data and intellectual property in-house. The company positions the 64GB model as a more accessible entry point to the same NVIDIA DGX Spark platform used by the 128GB configuration.

The system is powered by NVIDIA’s GB10 Grace Blackwell Superchip, which combines a Blackwell GPU with fifth-generation Tensor Cores and FP4 support with a 20-core NVIDIA Grace Arm CPU.

ASUS says the GB10 can deliver up to 1 petaflop of FP4 AI performance using sparsity while operating from a standard wall socket. Its NVLink-C2C interconnect provides a coherent CPU-and-GPU memory architecture with five times the bandwidth of PCIe Gen 5, according to the company.

64GB of unified memory

The new Ascent GX10 configuration comes with 64GB of LPDDR5X coherent unified memory and 273GB/s of memory bandwidth.

Rather than having separate pools of system RAM and GPU memory, the platform uses a unified memory architecture that allows the CPU and GPU to work from the same memory pool. ASUS says this configuration is sized for efficient open-source models, including models in the 30–35-billion-parameter class.

ASUS specifically highlights Qwen3.8-27B as an example of a model that can fit within the 64GB configuration.

The company continues to position the 128GB Ascent GX10 for larger local models and more demanding AI workloads, while the new 64GB model provides a lower-cost entry point for experimentation, model development, and inference.

Two Ascent GX10 systems can be clustered

The 64GB model can also be scaled beyond a single machine.

ASUS says two Ascent GX10 64GB systems can be connected using their built-in NVIDIA ConnectX-7 networking, creating a combined system with up to 128GB of unified memory, up to 2 petaflops of AI compute, and up to 546GB/s of combined memory bandwidth.

The company says the two-system configuration can deliver up to 1.7 times the performance of a single Ascent GX10 128GB system. NVIDIA’s Sync App and Cluster Assistant are designed to simplify the setup without requiring changes to the existing software environment or workflow.

That gives developers an upgrade path from a single system for prototyping and inference to a clustered setup as model sizes, context windows, and the number of concurrent AI agents increase.

Full NVIDIA AI software stack

ASUS Ascent GX10 64GB ships with NVIDIA DGX OS and the full NVIDIA AI software stack, including PyTorch, Jupyter, and Ollama.

The system also includes NVIDIA OpenShell and NVIDIA NemoClaw for policy-based agent security, along with SSH-based remote management for IT teams managing multiple systems.

ASUS also says the Ascent GX10 shares the thermal design of the 128GB model. Its cooling system includes seven-level fan control, five heat pipes, ultrawide fins, and two 140-by-80mm fans designed to support sustained AI workloads.

ASUS Ascent GX10 64GB vs. 128GB

The new 64GB configuration doesn’t replace the 128GB model. Instead, ASUS is positioning the two versions for different workloads.

The 64GB Ascent GX10 is intended for developers, researchers, and AI enthusiasts working with efficient open-source models and local agentic AI. The 128GB model remains the better fit for teams that need more memory for larger models and more demanding workloads.

For users who eventually need more capacity, two 64GB systems can also be clustered to create a 128GB pool of unified memory.

ASUS Ascent GX10 64GB availability

ASUS has announced the new 64GB configuration, but its press release does not provide a specific US price or retail availability date.

That makes the 64GB configuration particularly interesting as a potential entry point into NVIDIA’s DGX Spark ecosystem, but buyers will need to wait for ASUS or its regional partners to provide pricing and availability details.

For now, the key difference is straightforward: the Ascent GX10 family now offers 64GB and 128GB configurations, giving local AI developers more choice depending on the size of the models and workloads they need to run.