Dell Technologies announced a new workstation platform called the Dell Deskside Agentic AI Series. This system moves heavy AI work from the cloud directly onto local hardware. This solves a major problem with high computing costs.
Businesses often spend too much money running AI programs in the cloud. Jon Siegal is a senior vice president at Dell. He noted that one developer recently burned through one billion cloud tokens in just 24 hours.
That single day of work resulted in a cloud bill of 3,400 US dollars. Running AI directly on the device can save companies up to 87 percent over two years. Dell notes that companies can make up the cost of the hardware in just three months.
Sam Grocott is another senior vice president at Dell. He said many companies struggle to put AI into action because of strict data rules. The new deskside systems let teams run AI safely on their own machines without sending data outside the building.
These computers use special software to manage AI tasks smoothly. This software includes the Agent Toolkit for building long running programs. A key part of this toolkit is the NemoClaw reference stack.
NemoClaw is built on the OpenClaw framework. It actively combines high performance open models for reasoning and coding. This setup works together with OpenShell to create a secure space for business data.
Dell knows that hardware and software alone do not solve every problem. The company is also offering targeted Dell Services to help teams. This service guides companies through the entire process of setting up their new AI tools.
These new workstations are designed as a starting point. Companies can build and test AI tools locally before moving them to bigger servers. The Dell Deskside Agentic AI solutions are available right now.
Dell Deskside Agentic AI Series Specifications
| System Model | Best Use Case | AI Model Size Supported |
| Dell Pro Max with GB10 | Small scale agent testing | 30 billion to 200 billion parameters |
| Dell Pro Precision 9 | Heavy duty scalable workloads | 30 billion to 500 billion parameters |
| Dell Pro Max with GB300 | Very large AI model tasks | 120 billion to 1 trillion parameters |
Source: Dell
