For years, enterprise artificial intelligence has lived in the cloud.
Employees typed prompts into browser windows, queries traveled to data centers and responses came back. That model suited hyperscalers. It suited corporate IT and security teams less.
Nvidia and Microsoft moved to change that at GTC Tapei, unveiling what they called in a Sunday (May 31) press release a new class of Windows PC built for agentic computing. Laptops and compact desktops will be shipping this fall from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI, powered by Nvidia’s new superchip, RTX Spark. They are designed to run AI agents directly on the device.
The pitch to enterprise buyers is straightforward. Keep the data, the decisions and the agents inside the building.
Keeping Data Where It Starts
For enterprise buyers, the case centers on what stops leaving the building. Agents running locally can work across files, calendars and internal applications without sending data to a third-party server, cutting the latency, privacy exposure and usage costs that come with constant cloud inference.
Nvidia said earlier this year that traditional processors were “becoming the bottleneck” as agentic workloads grew, CNBC reported Monday (June 1).
On-device processing removes that constraint and keeps sensitive business data inside the firewall.
To govern what agents can see and do, Nvidia said in the release that it introduced OpenShell, a tool that lets users define access permissions and masks personal information before anything leaves the device. Microsoft built new security controls directly into Windows to manage agent identity and containment at the OS level. Hermes Agent and OpenClaw, two open-source agent projects that have reached record download numbers on GitHub, will build their Windows apps on the joint security stack.
Adobe committed to rebuilding Photoshop and Premiere for the platform, with tools like Generative Fill and Generative Extend among the accelerated features, targeting performance gains of up to 2x across AI and creative workflows, the release said.
A Stalled Market Gets a New Pitch
The launch comes as PC makers navigate a difficult year. IDC projected in March that global PC shipments will decline by 11.3% in 2026, a revision from its forecast just months earlier, driven by rising memory costs and supply constraints.
The boom year before the slide offered little comfort. Most of 2025’s shipment growth reflected Windows 10 end-of-life deadlines and tariff timing rather than genuine AI-driven demand. Copilot+ devices sold on the promise of local AI features that rarely changed how people worked.
Agent-ready devices are Nvidia’s answer to that gap. On-device agents that handle real tasks like drafting, researching and editing across applications make a more direct case for replacing hardware.
Nvidia sees the broader CPU market reaching $200 billion, with Arm-based chips gaining ground over the x86 processors that Intel and AMD have long dominated, according to the CNBC report.
Nvidia CEO Jensen Huang predicted in March that AI tokens would become a standard job perk, with companies issuing token budgets alongside base salaries so employees could run agents at scale.
The Sunday announcement extends an agent push Nvidia began earlier in March, when it debuted NemoClaw, an enterprise-hardened version of OpenClaw that adds security controls, policy enforcement and audit capabilities for companies in regulated industries.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
