ASUS · Articles

A Desktop Supercomputer Built Specifically for Running AI Locally

Many AI workflows depend entirely on cloud infrastructure, sending prompts, files, and context to remote servers with every interaction. ASUS's Ascent GX10 takes a deliberately different approach, built as a compact desktop system specifically designed to run advanced AI agents entirely on local hardware.

A genuine 1 petaFLOP desktop AI supercomputer

The Ascent GX10 delivers up to 1 petaFLOP of AI performance, powered by the NVIDIA GB10 Grace Blackwell Superchip — a specific, substantial compute figure for a desktop-class device rather than a vague "powerful" claim.

Built specifically on the NVIDIA DGX Spark platform

The device runs on the NVIDIA DGX Spark platform — the same foundational architecture behind the first Windows PCs purpose-built for AI agents, giving it genuine enterprise-grade AI infrastructure in a desktop form factor.

Genuine scalability through multi-system clustering

Using ConnectX-7 200GbE networking, two 64GB units can pool together to reach 128GB and 2 petaflops of combined compute, while four 128GB units scale to 512GB and 4 petaflops — real infrastructure scaling rather than a single fixed-capacity device.

Support for frontier-scale local AI models

The system specifically supports running large AI models — including 70B+ parameter models — locally, alongside fine-tuning and retrieval-augmented generation (RAG) workflows, without requiring cloud infrastructure for these demanding tasks.

Genuine data privacy through local processing

By keeping prompts, files, and agent context on-device, the Ascent GX10 removes the need for a cloud round-trip — a concrete privacy and latency advantage for anyone working with sensitive or proprietary data.

Compatibility with established agentic AI frameworks

The device specifically supports agentic AI frameworks including OpenClaw and Hermes Agent — letting users deploy established, recognised agent architectures rather than being locked into a single proprietary system.

Predictable costs without ongoing API fees

Running AI models locally on the Ascent GX10 removes dependency on paid cloud APIs — a genuine cost consideration for anyone running AI workflows regularly enough that cloud API costs would otherwise accumulate significantly.

Enterprise-grade stability for serious development work

ASUS specifically positions this device for developers and researchers needing enterprise-grade stability and secure data privacy — a genuine professional-tier offering rather than a consumer novelty product.

At a glance

Product: ** ASUS Ascent GX10

Chip: ** NVIDIA GB10 Grace Blackwell Superchip

Performance: ** Up to 1 petaFLOP (single unit)

Memory: ** 64GB or 128GB (scalable via clustering)

Networking: ** ConnectX-7 200GbE

Supported frameworks: ** OpenClaw, Hermes Agent, and others

Who it suits

The Ascent GX10 suits developers, researchers, and organisations wanting to run substantial AI models and agents entirely on local hardware, prioritising data privacy and predictable costs over cloud-dependent infrastructure. Anyone with lighter, everyday AI needs — such as writing assistance or document summarisation — should consider ASUS's broader AI PC lineup instead, since the Ascent GX10 is specifically built for demanding, large-scale local AI workloads. Explore ASUS Agent Computers — **

Visit store