HP ZBook Ultra G3a Will Bring Perplexity’s Local-First AI to Mobile Workstations

HP has announced that Perplexity Computer will come preloaded on the HP ZBook Ultra G3a, putting an agentic AI workspace on a high-memory Windows mobile workstation. Reporting on the partnership says HP plans to place the Perplexity Windows app on the taskbar, beginning with the G3a and potentially expanding to other HP systems; HP’s own launch materials confirm that Perplexity Computer is included on the workstation. 710

The significance is less about another desktop shortcut than the hardware behind it. HP is positioning the ZBook Ultra G3a as a machine that can keep more AI work—and potentially more sensitive project context—on the device rather than sending every task to cloud inference.

What HP announced

The 16-inch ZBook Ultra G3a uses AMD Ryzen AI Max PRO 400-series processors and can be configured with up to 192GB of unified memory, with up to 160GB allocated to graphics workloads. HP says this configuration supports local execution of language models with up to 300 billion parameters and concurrent professional and multi-agent workflows. 510

HP expects the workstation to be available in October 2026. Pricing had not been disclosed at announcement. 3

There is a rollout nuance worth noting: an industry report says the machine initially includes a pre-installed Perplexity link, with the full desktop application to be pre-installed later in the year and broadened to G3 mobile workstations. Buyers should therefore confirm the software state of the specific configuration and region they are purchasing. 4

How Portable Computer keeps work local

Perplexity Computer is intended for multi-step work spanning files, apps, tools, and the web. Its Portable Computer mode takes a local-first approach: the agent harness, orchestrator, planner, tool routing, and supported local models run on the PC. Perplexity lists PPLX 27B and Qwen 27B as local options, while saying Nemotron support is coming. 20

For a task that stays local, Perplexity says the model, conversation, and task trajectory remain on the user’s machine, and local inference does not consume cloud-inference credits. Users can download local models separately; HP notes that local LLMs are not pre-installed on the ZBook itself. 519

That does not mean every feature is offline. Web search, app connectors, or more capable frontier-model reasoning can require a cloud step. Perplexity says its orchestrator asks for approval before routing a task to cloud models, giving users a decision point before data leaves the device. 1920

Why the ZBook’s memory matters

Large-model AI work is often constrained by available memory. HP’s premise is that up to 192GB of unified memory—and as much as 160GB available for GPU workloads—creates room for substantially larger local models than many conventional laptops can accommodate. The company’s 300B-parameter claim is an advertised capability, not an independent performance benchmark, and real-world speed will depend on the model, quantization, task, and system configuration. 10

Still, the practical goal is clear: run an AI assistant beside demanding workstation applications, using local files and project data without making cloud inference the default path. HP explicitly frames the system around reducing token costs and helping keep data secure while working away from a fixed workstation. 5

Professional workflows: Revit and connected business data

HP says the preloaded Perplexity Computer includes supported Autodesk Revit integration. Coverage of the HP and Autodesk positioning describes a Revit 2027 MCP server that can enable tasks such as extracting project information, checking model quality, retrieving schedules, and drafting RFIs. 410

Beyond design and construction workflows, Perplexity says its Windows offering can work across local folders and desktop apps and connect to more than 400 applications, including Snowflake, Salesforce, and HubSpot. A Snowflake connector, for example, is designed to let organizations query data already used for analytics while respecting existing role-based and row- or column-level policies. 1822

Those capabilities bring an important governance consideration: local processing can limit unnecessary data transfer, but an app connector still operates with the permissions an organization grants it. Teams should review connector access, data policies, and the circumstances in which users approve cloud escalation.

A local-first option, not a blanket privacy guarantee

The HP-Perplexity deal reflects a wider product direction in AI PCs: add enough on-device compute and memory to handle useful work locally, then use cloud models selectively when a task needs the web or more advanced reasoning. The benefits Perplexity and HP emphasize are reduced cloud-token use, more control over sensitive data, and less dependence on a network connection for local tasks. 519

For prospective ZBook Ultra G3a buyers, the key distinction is simple: local processing is the default design, not a guarantee that every task stays local. The privacy and cost advantages apply most directly when users choose local models and avoid or carefully authorize cloud-connected steps. 1920