The ASUS Ascent GX10 is built from the ground up as a compact “agent computer” for modern AI teams who need serious performance without a full rack of servers.

Key capabilities for agentic AI workloads include:

• Extreme agentic AI performance: It’s powered by the NVIDIA GB10 Grace Blackwell Superchip, delivering about 1 PFLOP of performance. That’s enough to support fine-tuning models up to ~200B parameters and running multiple, high-demand personal or enterprise agents at the same time.
• 128 GB coherent unified memory: The system uses 128 GB LPDDR5x unified memory, so CPU and GPU share one memory space. This helps with large-context training, in-context learning, and fast iteration on big models without constant data shuffling.
• Developer-optimized ecosystem: The GX10 ships with an NVIDIA AI software stack and an optimized NemoClaw setup, streamlining workflows for AI researchers, developers, and data scientists who are building and orchestrating agents.
• Ready for real research environments: With a compact footprint (150 x 150 x 51 mm) and weighing about 1.48 kg, it fits on a desk, in a lab, or at the edge while still behaving like a dedicated AI workstation.

This combination makes the GX10 a good fit for AI model prototyping, large-model fine-tuning, inference workload processing, and building agentic AI systems in industries like healthcare, fintech, industrial automation, and retail.

The ASUS Ascent GX10 is designed to help teams start small and scale as their AI projects grow.

Scalability and connectivity highlights:

• Scalable interconnected architecture: The GX10 uses NVLink-C2C and NVIDIA ConnectX-7 networking so you can build dual-system stacks. This allows you to scale up with multiple GX10 units and keep ultra-fast communication between them for distributed or multi-agent workloads.
• High-speed networking: On the networking side, you get:
  • 1 x 10 GbE RJ-45 LAN for reliable wired connectivity.
  • 1 x NVIDIA ConnectX-7 NIC (up to 200 Gbps) for high-bandwidth, low-latency connections to other accelerated systems or storage.
  • Wi‑Fi 7 (2x2) and Bluetooth 5.4 with LE for flexible wireless setups, labs, or edge deployments.
• Compact but stackable form factor: The small chassis makes it easier to colocate multiple GX10 units in offices, labs, or edge locations, effectively creating a dedicated agent infrastructure without a full data center build-out.

For teams expecting rapid workload growth—more models, more agents, more concurrent users—the GX10 provides a path to scale horizontally by adding units and interconnecting them with high-speed fabrics.

The ASUS Ascent GX10 is built to help AI teams run private, governed AI workloads, including long-running agents, in a range of environments.

Key security, governance, and deployment features:

• Secure and governed autonomy:
  • OpenShell secured containers and NemoClaw sandboxes provide controlled environments for your models and agents.
  • They support private, on-device inference with governed access control, which is useful for sensitive data and regulated industries like healthcare and financial services.
• Enterprise-ready software stack:
  • The system comes with NVIDIA DGX OS and the NVIDIA AI software stack for a consistent, enterprise-style environment.
  • NVIDIA AI Enterprise is available separately for teams that need additional enterprise-grade tools and support (licensing via ASUS representatives).
• Designed for varied environments:
  • An advanced thermal design keeps this 1 PFLOP system running reliably in labs, offices, and edge locations, despite its minimal footprint.
  • 240W power supply and a compact chassis (150 x 150 x 51 mm) make it practical in space- and power-constrained settings.
• Regulatory and physical security:
  • The GX10 carries multiple certifications (including BSMI, CB, CE, FCC, UL, CCC, C-Tick, WiFi RF, VCCI), helping with compliance in different regions.
  • A Kensington lock slot supports basic physical security in shared or office environments.

Overall, the Ascent GX10 is positioned as a compact, governed AI platform for workloads like AI model prototyping, inference, data science, and edge application development where privacy, control, and deployment flexibility matter.