Microsoft’s October 7 Windows Event: RTX Spark, Surface and the Local AI PC

Microsoft’s October 7 event in San Francisco is shaping up as a statement of direction for the Windows AI PC rather than a conventional operating-system launch. The company says the conversation will focus on “how local AI will shape the next chapter of the PC,” spanning Windows, NVIDIA RTX Spark, Surface and the wider PC ecosystem. CEO Satya Nadella will appear alongside NVIDIA CEO Jensen Huang and Microsoft Windows and Devices leader Pavan Davuluri. 12

What is confirmed—and what is still speculation

The confirmed agenda is broad: Windows, NVIDIA RTX Spark, Surface and the PC ecosystem. The speaker lineup makes RTX Spark a particularly credible focus, but Microsoft has not published a detailed announcement slate, prices or regional launch plans for the event. Reports that it will bring Surface Laptop Ultra availability, configurations or additional RTX Spark machines are expectations, not confirmations. 279

That distinction matters because Microsoft has already revealed its two headline RTX Spark Surface systems. October is therefore more likely to add practical details—such as shipping information, demonstrations and partner hardware—than to serve as their first unveiling. 1922

The Surface hardware behind the local-AI pitch

Surface Laptop Ultra

Microsoft describes Surface Laptop Ultra as its first laptop combining an NVIDIA Blackwell RTX GPU, full CUDA support and up to 128GB of unified memory. Unified memory lets the system dynamically allocate the same pool of memory across CPU and GPU workloads, which Microsoft positions for AI creation, 3D rendering and multi-model workflows. The company says the device can deliver up to 1 petaflop of AI compute and run models of up to 120 billion parameters locally; it is listed as coming later in 2026. 19

Surface RTX Spark Dev Box

The Surface RTX Spark Dev Box is a compact Windows developer PC built around NVIDIA’s RTX Spark superchip. Microsoft says it has up to 1 petaflop of AI compute and 128GB of unified memory, with a developer-optimized Windows 11 Pro setup for local AI development, training and fine-tuning. 2230

Microsoft’s developer materials specifically frame the Dev Box around local AI agents and NVIDIA’s software stack. Reporting on the launch says the machine is intended for workloads such as long-running training, agentic pipelines and local model fine-tuning, with support for models of up to roughly 120 billion parameters. 2022

Both the Dev Box and Laptop Ultra are expected later in 2026. Microsoft had not disclosed the Dev Box price in the supplied official and launch reporting, so any reported price should be treated as unconfirmed until the company provides ordering details. 1929

Why Windows 11 matters more than Windows 12 here

There is no public Windows 12 announcement, and the event invitation does not name a successor to Windows 11. The available evidence instead points to continued Windows 11 work aimed at RTX Spark systems. 335

Microsoft says it implemented Workload Profile Scheduling (WPS) for RTX Spark’s heterogeneous 20-core architecture. The feature is intended to help the Windows scheduler scale workloads more efficiently across those cores, whether the machine is handling everyday desktop work or running a local coding agent. 35

RTX Spark is also an Arm-based platform, making compatibility with existing x86 software important. Reporting based on a discussion with a Microsoft product manager says Microsoft has overhauled the Prism emulation layer’s thread assignment for RTX Spark, with the goal of making performance more stable for translated applications. That is a reported implementation detail, rather than a formal October-event commitment. 33

A more concrete test for Microsoft’s AI-PC strategy

Microsoft’s 2024 Copilot+ PC rollout was overshadowed by concern about Recall, the feature designed to build a searchable history of activity on a PC. Microsoft delayed its initial release following privacy and security concerns. 46

The current local-AI pitch is different in emphasis. RTX Spark hardware is designed to place much larger models and development workflows on the device, potentially reducing the need to send every inference request, file or prompt to a cloud service. Microsoft says local inference and experimentation can lower API and cloud-compute costs while enabling faster iteration. 30

But running AI locally is not, by itself, a privacy guarantee. Privacy still depends on what a feature collects, whether it is opt-in, how long data is retained, which permissions it receives and how securely it is protected. Microsoft’s revised Recall architecture says Recall is opt-in, keeps snapshots and associated data locally, and lets users delete, pause or disable snapshots. 61

For October’s event, the meaningful question is therefore not simply how large a model RTX Spark can run. It is whether Microsoft can show local AI agents that are useful while giving people clear controls and credible safeguards. That would make the event a substantive next step for Windows AI—not just another hardware-performance claim.