Microsoft teams up with Nvidia to release flagship Surface device and simultaneously launches the "Lobster" personal assistant Scout

Microsoft has teamed up with Nvidia to launch a new offensive in the AI PC sector.

On Wednesday in San Francisco, Microsoft unveiled its flagship Surface Laptop Ultra powered by Nvidia chips, claiming its AI performance surpasses Apple's MacBook Pro M5 in multiple benchmarks, and entering Apple’s core market with a significantly lower starting price.

During the launch event in San Francisco, Microsoft executives stated that the Surface Laptop Ultra offers twice the response speed of the MacBook Pro M5 when running AI models locally and generates videos six times faster. The starting price is $2,599, roughly $400 less than the entry-level 16-inch MacBook Pro, and shipping will begin soon.

This series of moves aims to reinforce Windows' dominant position in the AI era. Apple Mac computers have seen soaring demand in recent years due to their strong capability in running AI models locally, and Microsoft’s counterattack directly intensifies competition.

Meanwhile, at the concurrent Build Developer Conference, Microsoft announced the launch of its AI agent, Microsoft Scout. Scout is positioned as an "always-online personal agent."

The Surface Laptop Ultra is equipped with Nvidia’s brand-new RTX Spark chip designed for Windows laptops, handling both general-purpose and AI computing tasks.

Nvidia CEO Jensen Huang and Microsoft CEO Satya Nadella attended the event together to endorse the new PC chip. At the event, Huang remarked:

We have completely reinvented the personal computer.

Nvidia’s entry directly challenges the traditional dominance of Intel and AMD in the mainstream PC processor market. Major Windows-based PC makers like Dell and HP have also introduced devices powered by Nvidia’s new chip.

The launch event focused on establishing a business logic centered around local AI computing: by running computing tasks on local devices, developers can save on cloud service costs and enhance data security. Unlike the previous consumer-focused Copilot Plus PC series, the RTX Spark platform is clearly targeted at AI developers and creators.

Nadella noted that over time, both developers and users will expect AI tools to be powered by a combination of local hardware and cloud services. Microsoft also showcased new Windows security features for monitoring and managing AI agent activities.

However, the mass adoption of AI PCs may face a timing challenge. Industry research firm Gartner previously predicted that rapidly rising memory costs could delay AI PCs from becoming the majority of shipments until 2028.

Beyond hardware, Microsoft also announced the launch of its AI agent Microsoft Scout at the concurrent Build Developer Conference.

Scout is positioned as an “always-online personal agent,” belonging to Microsoft’s newly established “Autopilots” agent category, capable of carrying out tasks autonomously without the need for prompts each time.

Scout integrates deeply into the Microsoft 365 app ecosystem and can access Teams, Outlook, OneDrive, and SharePoint to automatically handle scheduling, resolve meeting conflicts, and draft emails, among other tasks.

This product is built on OpenClaw open-source technology. Each agent instance has an independent Microsoft Entra service account and is managed within a unified permissions governance system. Combined with Microsoft Purview data protection policies, this offers comprehensive compliance for enterprises.

Currently, Scout is in private preview for a limited customer group, with desktop apps open to subscription users who opt into “early access” features. Using the desktop app requires a valid GitHub Copilot subscription. Microsoft says its internal sales team has become one of the largest and fastest-growing user groups for Scout.

While rapidly introducing agent products, Microsoft is also facing security challenges. Tools like Scout face the risk of prompt injection attacks, where malicious actors may coax AI agents to perform unauthorized actions or leak sensitive information.

Microsoft’s countermeasures include phased rollouts at a limited scale, providing administrators with tools to track all agent activities, and requiring manual approval for sensitive operations.

This challenge is not unique to Microsoft. Back in the summer of 2024, when Microsoft cooperated with Qualcomm to promote AI-optimized PCs, privacy concerns related to flagship features sparked public debate and slowed the overall release schedule. How to achieve a balance between automation capability and user trust remains a core question Microsoft must continuously address in the AI agent arena.