Last Updated on October 3, 2026 by Spencer Lee
HP has unveiled the ZBook Ultra G3a 16, a mobile workstation designed for running large AI models locally. It offers up to 192GB of unified memory, with up to 160GB available to the GPU, and HP says it can run AI models with up to 300 billion parameters without relying entirely on the cloud.
HP is pushing local AI beyond typical AI laptops
The ZBook Ultra G3a 16 is aimed at professionals who need more local AI performance than conventional Copilot+ PCs can provide.
HP says the workstation can run AI models and agents locally while handling demanding professional applications such as Autodesk Revit, 3ds Max and Chaos Enscape. The system uses AMD’s Ryzen AI Max+ PRO 495 processor and up to 192GB of unified memory.
The biggest difference is memory. Up to 160GB of the unified memory can be allocated to the integrated Radeon GPU, giving large AI models significantly more memory to work with than most laptop GPUs.
How large are the AI models it can run?
HP says the ZBook Ultra G3a can run LLMs with up to 300 billion parameters locally.
That doesn’t mean every 300B model will run quickly. Model quantization, context size, software and workload all affect actual performance.
Independent testing has not yet established the final performance of shipping systems. StorageReview currently has a pre-production unit and notes that its full benchmarks are still pending.
Why does 192GB matter for AI?
Memory capacity can determine whether a large AI model can fit on a laptop in the first place.
Traditional laptops often separate system RAM from dedicated GPU VRAM. The ZBook Ultra G3a’s unified-memory design gives the GPU access to a much larger pool, allowing models that would exceed the VRAM of many conventional laptops to be loaded locally.
HP also includes a 55-TOPS NPU for lower-power AI workloads, but the 192GB memory capacity is the headline feature for large local models.
What does this mean for laptop AI?
The ZBook Ultra G3a shows that AI laptops are starting to move beyond simple NPU performance numbers.
For developers, researchers and professionals working with local LLMs, AI agents and private data, having enough memory to run larger models can be more important than simply having a higher TOPS rating.
HP expects the ZBook Ultra G3a to be available in October 2026. HP’s current reference configuration lists the 192GB model at $7,449, making it a professional workstation rather than a mainstream AI laptop.
What happens next?
The key question is performance. Once retail systems are available, independent testing should show how quickly the ZBook Ultra G3a can run large models, how it handles sustained AI workloads, and whether its huge memory pool provides a practical advantage over high-end RTX laptops.
Related article:
Is 16GB RAM Enough for Machine Learning in 2026?
The ZBook Ultra G3a also highlights why RAM requirements are changing as more AI workloads move onto laptops. Our guide looks at how much memory different machine-learning workloads actually need.
