The ASUS Ascent GX10 is a compact desktop AI supercomputer that delivers 1 petaflop of computing power, enough to run large AI models and autonomous agents entirely on-premises without relying on cloud services. Powered by NVIDIA's GB10 Grace Blackwell Superchip and featuring 128 gigabytes of unified memory, the system is designed to democratize access to enterprise-grade AI development for researchers, data scientists, and developers who want to prototype, fine-tune, and deploy AI agents locally.

What Makes Local AI Execution a Game-Changer for Enterprises?

Running AI agents on local hardware instead of cloud platforms addresses three persistent pain points that have frustrated organizations deploying autonomous systems. First, there's the cost problem. Cloud-based AI services charge per API call or token processed, which adds up quickly when agents run continuously. Second, there's data sovereignty; sending sensitive business information to external servers creates compliance and security risks. Third, there's reliability; depending on third-party APIs means your agents stop working if the service goes down or changes its pricing model.

The ASUS Ascent GX10 tackles these challenges by providing enough local computing horsepower to run sophisticated open-source AI models without perpetual API dependency. The system can handle models like Llama 3.1 with up to 405 billion parameters when two units are linked together, and it supports DeepSeek R1 inference optimized for up to 70 billion parameters on a single machine.

How to Deploy Secure, Autonomous AI Agents Locally?

• Enable NemoClaw Sandboxing: The ASUS Ascent GX10 comes with NVIDIA NemoClaw, an open-source stack that adds privacy and security controls to agent frameworks. It installs NVIDIA OpenShell to enforce policy-based guardrails, giving teams control over how agents behave and handle data without exposing systems to external threats.
• Leverage Unified Memory Architecture: The system's 128 gigabytes of coherent unified memory, connected via NVLink-C2C with five times the bandwidth of PCIe Gen 5, enables seamless CPU and GPU collaboration. This architecture supports concurrent, high-demand personal agents and eliminates memory bottlenecks that slow down complex reasoning tasks.
• Scale with Dual-System Stacking: Two ASUS Ascent GX10 units can be linked via ConnectX-7 networking for ultra-fast communication, allowing teams to handle larger models and distribute workloads across dedicated agent infrastructure without sacrificing performance.
• Run Open Models for Cost Predictability: By executing open-source models like Llama and Qwen locally, organizations achieve greater cost predictability, enterprise-grade stability, and reduced risk of unintended data exposure compared to cloud-dependent alternatives.

Why Is Desktop AI Supercomputing Arriving Now?

The convergence of three factors makes this moment significant. NVIDIA's latest Blackwell GPU architecture delivers unprecedented performance density in a compact form factor. Second, the AI agent ecosystem has matured enough that frameworks like OpenClaw and Hermes Agent can run reliably on local hardware. Third, enterprises are increasingly concerned about vendor lock-in, data privacy, and the hidden costs of cloud-dependent AI systems.

The ASUS Ascent GX10 integrates a complete AI software stack out of the box, including NVIDIA DGX OS (an Ubuntu-based environment), CUDA, PyTorch, TensorFlow, Jupyter, NVIDIA TensorRT for high-performance inference, and NVIDIA NIMs and Blueprints for prebuilt AI workflows. This means developers can start building and deploying agents immediately without spending weeks on infrastructure setup.

What About Security and Governance for Autonomous Systems?

One of the system's standout features is its focus on secure, governed autonomy. The ASUS Ascent GX10 merges enterprise readiness with OpenShell secured containers and NemoClaw sandboxes to provide private, on-device inference and governed access control for long-running autonomous agents. This is critical because autonomous systems that operate without proper guardrails can make costly or harmful decisions.

The hardware also includes built-in security acceleration for TLS, IPsec, and MACsec, ensuring encrypted data transmission without consuming CPU resources. For time-sensitive applications, IEEE 1588v2 PTP support enables microsecond-level time synchronization, which is essential for coordinated multi-agent systems and edge computing deployments.

The desktop AI supercomputer represents a shift in how organizations think about AI infrastructure. Rather than treating AI as a cloud service consumed on demand, teams can now own and operate their own petaflop-scale compute resources, running autonomous agents 24/7 with full control over data, costs, and behavior. For enterprises tired of API dependencies and concerned about data privacy, the ASUS Ascent GX10 offers a compelling alternative to the cloud-first model that has dominated AI deployment for the past three years.