Microsoft will hold an event in San Francisco on October 7 to lay out its next moves in Windows and Surface, with local AI computing expected to take center stage. The presentation marks the company’s first major Windows-focused event in more than two years, since May 2024, when it unveiled Copilot+ PCs and built a strategy around AI-ready hardware. Microsoft CEO Satya Nadella and Pavan Davuluri, the executive leading Windows and Surface, are set to attend, and Nvidia CEO Jensen Huang is also scheduled to appear.
The Nvidia presence is not decorative. The event is expected to cover the RTX Spark platform, the chip Nvidia unveiled earlier this year as the hardware foundation for what the two companies call the age of personal AI agents. RTX Spark pairs a Blackwell RTX GPU with up to 6,144 cores and a 20-core Grace CPU over Nvidia’s NVLink-C2C interconnect, delivering up to 1 petaflop of FP4 AI performance and configurable with up to 128 GB of unified memory. Laptops built on it can be as slim as 14 millimeters and as light as three pounds, with Asus, Dell, HP, Lenovo, Microsoft Surface and MSI launching models this autumn, and Acer and Gigabyte models to follow. Compact desktop versions are also planned.
The Surface Laptop Ultra takes the stage
Attention will focus on the Surface Laptop Ultra, a system Microsoft introduced earlier this year and still lists as a pre-release product. The laptop runs on the RTX Spark platform with up to 128GB of unified memory and a petaflop of AI compute, specifications that point at users who need serious local compute for demanding AI tasks rather than the lightweight NPUs of the 2024 Copilot+ generation. A leaked benchmark of the machine circulated earlier this month ahead of the event, and Microsoft’s own Surface account teased the appearance of both Nadella and Huang at the gathering.
That could make the Surface Laptop Ultra a different kind of Windows machine from the Copilot+ PCs Microsoft introduced two years ago. Those machines leaned on 40-plus TOPS NPUs tuned for background features like Recall and Studio Effects. The Ultra is built to run 120-billion-parameter language models locally, render 90 GB 3D scenes and edit 12K video, workloads that until now required a cloud connection or a desktop workstation. Nvidia claims the chip can also play AAA games at 1440p resolution at frame rates above 100 frames per second, putting it in genuine gaming territory as well as creator and developer work.
Microsoft has not said whether the event will include other new Surface hardware or Windows features, but the company has framed the gathering as a look at how local AI will shape the next chapter of the PC. New Windows security primitives and Nvidia’s OpenShell runtime are expected to feature, giving users control over what on-device agents can access, a response to growing unease about what AI agents do on machines that hold personal and corporate data. OpenClaw and Nous Research are among the agent developers planning dedicated Windows apps for the platform, which suggests Microsoft wants an agent ecosystem ready at launch rather than a promise of one later.
Why local AI, why now
The timing reflects a shift in how AI work is distributed. Cloud inference costs money per token and raises privacy questions for enterprise data, while on-device inference is free at the margin once the hardware is bought. Nvidia’s pitch for RTX Spark leans on exactly that argument. Adobe is rearchitecting Photoshop and Premiere from the ground up for the chip with promised performance gains, alongside Blackmagic Design, Blender, ComfyUI and OTOY on the creative side. On gaming, KRAFTON, NetEase, Remedy Entertainment and Xbox are among studios pledging support, with the chip also expected to drive DLSS 4.5 Ray Reconstruction.
The competitive backdrop is crowded. Apple’s silicon has led on local AI performance per watt for years, and Qualcomm’s Snapdragon X series pushed Windows on Arm into the mainstream over the past two years. RTX Spark, built in collaboration with MediaTek on a custom Arm-based design, is Nvidia’s answer to both, and its CUDA software stack remains the default for developers shipping AI features on Windows. The full software stack, CUDA, DLSS, OptiX and Reflex, carries over, which lowers the porting cost for existing Windows applications compared with a move to a new instruction set.
There is also a server angle worth watching. Reports have circulated that Apple could return to the server market with M8 Ultra hardware and Nvidia networking, a sign that local AI silicon is pushing up the stack from laptops into data center edge deployments. Microsoft’s event will not cover that directly, but the same logic drives both moves: the cost of moving every AI token to a distant GPU rack is becoming the bottleneck for the next wave of applications, and the economics favor running what you can on the device in front of you.
The broader PC market could use the jolt. Windows shipments have been flat as the Copilot+ pitch failed to move mainstream buyers, and the industry has been waiting for a use case that justifies premium hardware again. If agentic AI on the desktop turns out to be that use case, the October 7 event will be remembered as the moment the pitch finally changed. If it does not, the Surface Laptop Ultra will join a long line of powerful machines looking for a reason to exist.
The event starts at 10am Pacific time, 1pm Eastern. Expect pricing and availability for the Surface Laptop Ultra to be the headline question, since a machine with 128GB of unified memory and a petaflop of AI compute will not come cheap, and Microsoft has yet to say what it will charge. The second question is how the promised agent security model works in practice, which will determine whether enterprises treat local AI PCs as an upgrade or a liability.