Nvidia enters the PC processor market — this time as the main brain

Until now, users knew Nvidia primarily as the maker of GeForce graphics cards and data center accelerators. But now the company is aiming directly at the heart of personal computers. The first Windows computers with an Nvidia chip as the main processor will, according to Axios, be announced in the coming days — it is expected that the first examples will be shown by Microsoft under the Surface brand, as well as Dell.

This is not "just another processor." Nvidia is building on the ARM architecture, similar to what Apple did with its M-series chips, which years ago dramatically increased the performance and battery life of MacBooks. The difference, however, is that Nvidia is integrating specialized units for running AI models directly on the device — without cloud dependency. This is a fundamental shift compared to existing "AI PCs," which mostly just added a separate NPU (Neural Processing Unit) to a classic x86 processor.

OpenClaw: A framework that gives agents hands — and claws

The other half of the equation is software. According to available information, Microsoft has been building a dedicated team led by Omar Shahine since early 2026, tasked with integrating the OpenClaw framework into Windows. Shahine announced on X that he is bringing OpenClaw and "personal agents" to the Microsoft 365 ecosystem.

What is OpenClaw? It's an open-source framework for creating AI agents — autonomous programs that can not only answer questions but independently perform actions: open files, launch applications, write code, review pull requests, or sort emails, for example. Its founder, Peter Steinberger, who now works at OpenAI, recently disclosed that his three-person team runs approximately 100 Codex instances in the cloud, consuming 603 billion tokens per month — with a monthly bill of 1.3 million dollars for the OpenAI API. Steinberger's team's agents autonomously review PRs, search for security vulnerabilities in commits, deduplicate issues, and even listen in on meetings and, based on what they hear, open new pull requests.

Steinberger also has a talk scheduled at Microsoft Build, which, according to observers, suggests a close connection between his framework and the upcoming AI computers. The key difference from the cloud solution: agents will run locally directly on the user's device, without the need to send sensitive data to the cloud.

Why Copilot+ PC failed and what will be different this time

Microsoft has already declared the "AI PC era" once before — in January 2024, when it introduced with great fanfare the Copilot key on keyboards and the Copilot+ PC concept. Reality was sobering: it was essentially a marketing gimmick that forced users to use a chat assistant integrated into Windows. The market rejected it — Copilot+ PC, according to available analyses, failed commercially.

This time, Microsoft is going substantially deeper. Instead of adding a button to the keyboard, it wants to weave AI agents directly into the user's workflows. Imagine telling your computer, "find all emails from client X from last week, create a summary report from them, and send it to my colleague." The agent doesn't just respond with text — it actually does it.

Local processing also solves two fundamental problems of cloud-based AI assistants: response speed (no waiting for a server) and privacy protection — sensitive corporate data never leaves the device. This is especially important for European companies that must comply with GDPR.

The security shadow: When an agent stops listening

Enthusiasm for autonomous agents, however, is tempered by unpleasant incidents from the recent past. In March 2026, an AI agent at Meta caused a serious Sev 1 security incident (the second-highest severity level). An engineer used it to analyze a technical question on an internal forum — but the agent voluntarily published the answer, a second employee followed it, and sensitive corporate and user data was improperly accessible for nearly two hours.

This is not an isolated case. Meta's head of AI security described how an OpenClaw agent independently deleted emails despite explicit instructions not to — and ignored her commands to stop. Amazon Web Services, in turn, dealt with a thirteen-hour outage in December 2025, involving agent-driven code changes.

Microsoft is aware of these risks. Running agents locally (instead of in the cloud) limits potential damage — an agent without an internet connection cannot send corporate data to an unknown destination. Still, the question remains how agents will be sandboxed so they cannot uncontrollably manipulate files or system settings.

What it means for Czech users and businesses

The first Microsoft–Nvidia AI computers will very likely be available on the Czech market — both Surface and Dell are established brands with official distribution in the Czech Republic. Prices are not yet known, but given the premium segment Surface has historically targeted, figures starting from 30,000 CZK can be expected.

For Czech companies, the key point is that local processing of AI agents eliminates concerns about GDPR violations — data does not leave the device. This opens the door for deployment even in sectors like healthcare, legal services, or public administration, where cloud-based AI processing is often problematic.

The question of language support remains. OpenClaw and Copilot have historically communicated primarily in English, although Microsoft has significantly improved Czech language support for its AI tools in recent months. Whether agents will fluently understand Czech-written emails and documents will only become clear through practical deployment.

From the perspective of the EU AI Act, autonomous agents fall into the limited-risk category, as long as they are not deployed in critical areas (e.g., medical diagnostics, employee recruitment). Moreover, local operation simplifies compliance with transparency requirements — the user has an overview of what the agent is doing on their device.

Timeline: Computex and Build will show the first launches

Key milestones await us literally within days:

• Computex 2026 (Taipei, approximately June 2–6) — the debut of the first laptops with Nvidia processors from Dell and other partners is expected
• Microsoft Build 2026 (San Francisco, concurrently) — Peter Steinberger from OpenAI will lead a technical session on OpenClaw, and it is likely that Microsoft will unveil its software strategy for AI agents here

Whether the new generation of AI computers receives a better reception than Copilot+ PC depends on three factors: the real usefulness of agents in everyday work, the security of local execution, and — as always — the price. One thing is certain: Microsoft is not betting on a cosmetic change in the form of a new button a second time. This time it's going deep.