HP is taking the idea of an AI PC several steps further with its new ZBook Ultra G3a, a 16-inch mobile workstation built around AMD’s Ryzen AI Max PRO processors and designed to run large AI models locally. The machine also comes with Perplexity integrated into its professional workflow, giving users a way to combine on-device AI with Perplexity’s agentic tools without sending every task to the cloud.

Announced September 15, the ZBook Ultra G3a is aimed at professionals working with engineering, architecture, design, content creation, data science and AI development. HP says the workstation can run models with up to 300 billion parameters locally, depending on the configuration and workload, while supporting multiple AI agents alongside traditional professional applications.

At the center of the system is AMD’s Ryzen AI Max PRO 400 Series, paired with integrated Radeon graphics and up to 192GB of unified memory. HP says as much as 160GB of that memory can be allocated as VRAM, giving the system an unusually large pool of memory for local AI workloads without relying on a conventional discrete GPU. The processor also includes an NPU capable of up to 55 TOPS for supported AI workloads.

That memory configuration is a major part of HP’s local-AI pitch. Rather than requiring professionals to send large models and proprietary data to remote servers, the ZBook Ultra G3a is designed to perform inference directly on the workstation. HP says this can reduce cloud dependence and recurring token costs while giving organizations more control over where their data is processed.

But the Perplexity integration is what makes this particular ZBook more interesting.

Perplexity comes built into the workflow

HP and Perplexity are bringing Perplexity’s agentic capabilities directly into the workstation experience. The ZBook Ultra G3a comes with a pre-installed Perplexity link, along with a 30-day trial of Perplexity Enterprise Pro and 5,000 free credits, according to HP.

The partnership goes beyond simply installing the Perplexity app. HP says Perplexity connectors and HP-exclusive skills can help users automate tasks inside professional applications, including Autodesk Revit.

The companies have specifically enabled Perplexity Portable Computer for Windows to work with Autodesk Revit through an MCP Connector. That allows AI agents to interact with the application and project data while keeping the relevant information on the local device for supported workflows. HP says this approach is intended to minimize cloud token costs while reducing the amount of manual work involved in selecting tools, finding instructions and constructing prompts.

Perplexity describes the underlying approach as Portable Computer, a mode of its Personal Computer experience designed to run AI models, agents and tools on the device. The idea isn’t necessarily to eliminate cloud AI altogether. Instead, work can begin locally and a user can bring in a cloud model when a particular step requires frontier-level intelligence.

That hybrid approach is becoming increasingly important as AI moves from chatbots into software that can actually perform multi-step tasks.

For a workstation user, the difference is significant. A traditional AI chatbot might answer a question about a Revit project. An agentic system can potentially work with the application itself, interact with tools and carry out defined tasks. The ZBook Ultra G3a is being designed around that second model.

A workstation built around local AI

The ZBook Ultra G3a isn’t a conventional thin laptop with an AI accelerator added to the specification sheet. HP has redesigned its thermal system to allow the machine to sustain up to 100W of performance in a relatively thin 16-inch chassis.

HP says its new thermal design delivers up to 81% higher performance compared with the previous generation, with dual mid-mounted fans, a larger vapor chamber and BIOS-level power management that dynamically distributes power according to the workload.

The result is a machine that HP says can simultaneously handle large local AI models, multiple agents and demanding professional applications.

That matters because local AI is often constrained less by the theoretical capability of the processor than by available memory and sustained power. The ZBook’s 192GB unified memory configuration gives AMD’s integrated Radeon graphics access to a much larger pool than is typical on a laptop, with up to 160GB available for graphics and AI workloads.

HP is also targeting professional applications such as Autodesk Revit 2027, Autodesk 3ds Max and Chaos Enscape, making the ZBook Ultra G3a as much an engineering and design workstation as it is an AI machine.

Thunderbolt 4 on an AMD workstation

There’s another interesting detail in the hardware.

Despite being powered by AMD rather than Intel, the ZBook Ultra G3a includes two Thunderbolt 4 ports. HP specifies both ports with 40Gbps USB-C signaling, USB Power Delivery and DisplayPort 2.1 support.

That is notable because Thunderbolt has historically been closely associated with Intel platforms, but the technology isn’t restricted to Intel processors. The ZBook Ultra G3a demonstrates that an AMD-powered workstation can also carry certified Thunderbolt connectivity.

Beyond the two Thunderbolt 4 ports, the workstation includes another USB-C port with 10Gbps signaling, HDMI 2.1, USB-A, a microSD card reader and an optional nano-SIM slot for cellular connectivity.

HP is also offering up to 8TB of PCIe 5 NVMe storage and a choice of Windows 11 Pro or Ubuntu 26.04 LTS, giving the machine a more workstation-oriented software configuration than the typical consumer AI PC.

Local doesn’t mean completely offline

One of the more interesting aspects of HP and Perplexity’s approach is that they aren’t presenting local AI and cloud AI as mutually exclusive.

The ZBook Ultra G3a is designed to run local models and agents directly on the workstation, particularly when privacy, latency or recurring cloud costs matter. But Perplexity’s Portable Computer model can also use cloud intelligence when a task requires it.

That gives the machine a hybrid architecture: local computing where it makes sense, cloud AI where additional capability is needed.

For businesses handling proprietary engineering projects, designs or other sensitive information, that distinction could be more meaningful than simply having a faster AI accelerator. HP is explicitly positioning local inference as a way to give organizations more control over data and AI costs.

The company is even introducing a Local AI Value Calculator to help Z workstation customers estimate the potential savings from moving AI workloads away from cloud services.

The ZBook Ultra G3a therefore represents a slightly different vision of the AI PC. Instead of treating AI as a collection of small features running in the background, HP is building the machine around the assumption that professionals may want to run enormous models, autonomous agents and conventional workstation applications at the same time.

And with Perplexity providing the agentic software layer, HP is betting that the next generation of mobile workstations won’t just help professionals create things faster. They will increasingly be able to take on parts of the work themselves.

HP says the ZBook Ultra G3a 16 is expected to become available in October 2026, with pricing to be announced closer to launch.