Many universities are expanding AI and machine learning programs, but their infrastructure hasn’t kept pace. A 2025 study shows that the number of Master’s programs in AI in U.S. universities has grown by almost 170% in just three years, and undergraduate programs have more than doubled in the past year. This growth puts real pressure on compute resources.

Traditional approaches create several pain points:

• Cloud-only access: Renting GPUs in the cloud often leads to unpredictable costs, competition for access, higher latency, and added compliance and data sovereignty risks.
• Legacy on-prem servers: Older servers can be power-hungry, hard to scale, and may not support modern AI workloads with models that have billions of parameters.
• Laptops and small workstations: These typically can’t handle large models or intensive multimodal, robotics, or agent-based workloads.

The ASUS Ascent GX10 is designed to close this gap by providing:

• Petaflop-class performance in a mini PC–sized system, powered by NVIDIA’s GB10 superchip.
• 128GB of unified system memory, enabling fine-tuning of models up to 200B parameters on a single unit.
• The option to connect two GX10s to work with even larger models, such as Llama 3.1 with 405B parameters.
• Local, predictable compute that reduces reliance on cloud rentals and helps maintain data sovereignty and regulatory compliance.

In practice, this means faculty and students can run advanced AI workloads—such as predictive models, multimodal workflows, and robotics intelligence—on campus, without being constrained by shared cluster queues or fluctuating cloud budgets.

The ASUS Ascent GX10 is built to help universities rethink the economics of AI infrastructure.

For many faculty-led AI programs, cloud access to capable GPUs can easily exceed $1,000 per week based on publicly available pricing. These costs can rise without notice due to demand, and they often come with additional infrastructure and management overhead.

At the same time, many programs run 250–400 lab hours per week. When you map those hours to cloud GPU usage, the spend can escalate quickly.

By contrast, a deployment of GX10 units can:

• Stabilize costs: Move from variable, usage-based cloud fees to a more predictable on-prem investment.
• Potentially pay for itself in a single academic term: For programs with heavy lab usage, the avoided cloud spend over one term can be comparable to the cost of the hardware.
• Lower energy consumption: The GX10 can reduce energy use by up to 60% compared to many legacy servers still in use in existing programs.
• Reduce compliance overhead: Keeping sensitive training data on local hardware simplifies data sovereignty and regulatory requirements.

In short, the GX10 helps institutions reimagine AI infrastructure as a predictable, on-campus asset rather than a fluctuating operational expense tied to cloud usage.

Every ASUS Ascent GX10 ships with NVIDIA DGX OS and full access to the NVIDIA AI software stack that’s used in larger, datacenter-grade environments. This gives your students and researchers hands-on experience with the same tools they’ll encounter in industry.

Out of the box, your team can work with:

• NVIDIA NeMo for model customization and large language model workflows.
• NVIDIA Cosmos for physical AI applications.
• NVIDIA Metropolis for vision AI and video analytics.
• NVIDIA Holoscan for AI sensor processing.
• NVIDIA Isaac for robotics development.

This stack enables coursework and research that mirrors real-world AI development, including:

• Large-scale model fine-tuning (up to 200B parameters on a single GX10, and larger with dual units).
• Multimodal workflows that combine text, vision, and sensor data.
• Robotics intelligence and agent-based systems.

Because the GX10 uses the same NVIDIA AI stack as larger systems, your program can scale seamlessly to more powerful hardware—such as the ASUS ExpertCenter Pro or the ASUS AI Pod—without changing your software environment or curriculum.

To help you evaluate fit before purchase, ASUS also offers a Virtual Lab program: a remote, on-demand environment where you can access a single or dual GX10 instance for up to 72 hours. This lets you test your actual AI workloads and teaching scenarios on real hardware, using the same NVIDIA software stack you’d have with a physical deployment.