Dell Technologies has released two new desktop workstations specifically engineered to run massive artificial intelligence models entirely on-device, without relying on cloud services. The flagship "Dell Pro Max with GB300" can process models with up to 1 trillion parameters using 748 gigabytes of combined memory, while the compact "Dell Pro Max with GB10" fits 128 gigabytes of unified memory into a machine smaller than a shoebox. Both systems represent a significant shift toward local AI development for research institutions and enterprise teams.

What Makes These Machines Different From Regular Computers?

The key difference lies in raw processing power and memory architecture. The GB300 is built around NVIDIA's Grace Blackwell Ultra Superchip, which integrates 72 CPU cores and specialized GPU cores capable of delivering 20 petaflops of AI performance in FP4 compute, a measurement of how many quadrillions of operations the system can perform per second. To put that in perspective, this is roughly equivalent to the computing power of thousands of standard laptops working in parallel.

The GB10 is more modest but still powerful, delivering up to 1 petaflop of performance while weighing as little as 1.31 kilograms. What makes it remarkable is that the CPU and GPU share the same 128-gigabyte memory pool, meaning developers can run a large language model (LLM), an AI system trained on vast amounts of text to generate human-like responses, and image generation simultaneously on a single machine without running out of memory.

How Do These Workstations Perform in Real-World Use?

Testing revealed practical performance metrics that matter to developers. A 120-billion-parameter mixture-of-experts model, which uses multiple specialized AI networks to process information, achieved generation speeds of 36 to 47 tokens per second when running LM Studio, a tool for managing and running local AI models. A larger 284-billion-parameter model called DeepSeek V4 Flash ran at approximately 15 tokens per second using a technique that offloads memory to solid-state drives.

In simultaneous workload testing, the GB10 handled combined LLM and image generation tasks using approximately 100 gigabytes of memory. A more complex workflow combining language model processing, image generation, and video generation completed in 9 minutes and 25 seconds while using 92 gigabytes of memory. These results demonstrate that memory-intensive workloads previously impossible on standard desktop GPUs can now run on a single machine.

Why Would Organizations Choose Local AI Over Cloud Services?

Several practical advantages emerge from running AI models locally rather than sending data to cloud providers. The GB10 consumes minimal power and generates little heat, allowing long batch processes like model fine-tuning to run overnight without noise or thermal concerns. For development workflows involving long codebases, the system's strong compute performance during the input-processing phase minimizes perceived slowness. Additionally, keeping sensitive data on-premises rather than transmitting it to external servers addresses privacy and security requirements that many enterprises face.

Dell positions these systems within its broader "Dell AI Factory with NVIDIA" ecosystem, which allows code, models, and workflows built locally to transfer seamlessly to data centers and cloud environments. This hybrid approach lets organizations prototype AI applications on-premises and scale them to production without rewriting code.

Steps to Evaluate Local AI Workstations for Your Organization

• Memory Requirements: Calculate the total parameters of models you plan to run and ensure the workstation has sufficient memory; the GB300 supports up to 1 trillion parameters while the GB10 handles 200 billion standalone or 400 billion when two units connect via high-speed networking.
• Workload Patterns: Assess whether you need simultaneous multi-model processing, such as running language generation and image synthesis at the same time, which influences whether the GB10's shared memory pool or GB300's larger capacity better suits your needs.
• Thermal and Power Constraints: Evaluate your physical environment; the GB10's compact form factor and low heat output make it suitable for office settings, while the GB300 requires Dell's proprietary MaxCool cooling technology to manage sustained peak performance.
• Integration with Existing Infrastructure: Confirm compatibility with your current software stack and cloud environments, since both systems ship with Ubuntu Linux and NVIDIA's AI frameworks pre-installed to minimize setup time.

The Dell Pro Max with GB300 sold out immediately upon launch on August 20, 2026, suggesting strong demand from research institutions and enterprise AI teams. The GB10-powered system, while not exclusive to Dell, represents a growing market segment of compact, high-memory AI development machines designed for teams that need local processing without sacrificing capability.

These workstations signal a broader industry trend toward decentralized AI development. As models grow larger and organizations prioritize data privacy and operational independence from cloud providers, the ability to run trillion-parameter models on-premises becomes increasingly valuable for competitive advantage and regulatory compliance.