News
Microsoft and NVIDIA Unveil Surface Laptop Ultra Tonight: Running AI Agents Locally on Your Laptop
Microsoft will hold an event in San Francisco at 1 AM Beijing time on October 8, with CEO Satya Nadella and NVIDIA CEO Jensen Huang sharing the stage to unveil the Surface Laptop Ultra powered by the RTX Spark chip — up to 128GB of unified memory, capable of running 120B-parameter models locally. Microsoft wants to move AI agents from the cloud back onto local devices.

At 1 AM Beijing time on October 8 (10 AM on October 7 in San Francisco), Microsoft is holding an event in San Francisco. CEO Satya Nadella and NVIDIA CEO Jensen Huang will share the stage to unveil a new laptop, the Surface Laptop Ultra. The event will be livestreamed in full, with the theme of “local AI.”
Microsoft’s positioning for this product is blunt: let AI agents run directly on the laptop, no cloud calls needed. Agents that can autonomously handle complex, long-running tasks — writing code, organizing projects — will soon be able to do it all on a personal computer.
RTX Spark: A Chip Designed for Local AI
At the heart of the Surface Laptop Ultra is NVIDIA’s RTX Spark, a chip that was already unveiled back in June.
A single RTX Spark packages a 20-core Grace Arm CPU and a Blackwell-architecture RTX GPU on the same substrate, linked by high-speed NVLink C2C interconnect, sharing one pool of LPDDR5X unified memory. Unified memory lets the CPU and GPU read and write the same memory addresses directly — no shuttling data back and forth — which means lower latency when running large models.
The specs: 6,144 CUDA cores, up to 128GB of unified memory, an 80-watt power envelope, and full CUDA ecosystem support. According to Microsoft, this configuration can run 120-billion-parameter models locally, fully offline.
The RTX Spark platform launches with six laptop designs: the ASUS ProArt P16, Dell XPS 16, HP OmniBook Ultra 16, Lenovo Yoga Pro 9n, MSI Prestige N16 Flip AI+, and Microsoft’s own Surface Laptop Ultra. There’s also a developer-targeted Surface RTX Spark Dev Box claimed to deliver 1 petaflop of AI compute.
Microsoft’s Bet: Moving Cloud Costs onto Your Device
Microsoft floated the idea of running AI workloads on Windows PCs two years ago. Back then, Copilot+ PCs were priced under $2,000 and the goal was to save the cloud money. Now the ambition has leveled up: what’s supposed to run locally are AI agents that work autonomously for extended stretches.
It’s a good deal for Microsoft. The cost of running inference in Azure data centers sits on Microsoft’s own books; moved onto users’ laptops, the hardware bill becomes the user’s. For NVIDIA, this is its first real break into the Windows laptop market — territory long held by Intel and AMD.
Apple is on the same road. Reuters reports that Apple, while pushing new Macs, is tightening the approval process for granting AI agents full-disk access on the Mac.
Soaring Memory Prices: Local AI’s Biggest Roadblock
Great idea. Brutal pricing.
Memory chip prices have surged in recent weeks, pushing up whole-machine costs directly. NVIDIA raised the price of its own DGX Spark AI desktop by about 75% to $6,950, driven by the soaring cost of 128GB of memory.
Moor Insights & Strategy analyst Anshel Sag told Reuters that two years ago “the software wasn’t ready, but the hardware was; now the software is ready, but the hardware is too expensive to run locally.” His conclusion was blunt: local AI is becoming something only the well-funded can afford.
The Surface Laptop Ultra’s price and ship date will be announced at tonight’s event. That’s the question analysts care about most: with memory prices where they are, what price will the market accept for a laptop that runs 120-billion-parameter models?
Security: Is an Agent Locked in Your Laptop Actually Safe?
Reuters notes that Microsoft and NVIDIA must answer one question tonight: how do you safely contain an agent that runs autonomously for long stretches on a personal computer?
Agents aren’t chatbots — they call tools on their own, read and write files, and execute multi-step tasks. Keeping them local does reduce the risk of data being uploaded to the cloud, but once an agent “goes rogue” locally, it’s the user’s own machine and files at stake. NVIDIA is already working to avoid a repeat of the Hugging Face breach. On Microsoft’s side, Windows has been adapted for RTX Spark’s shared-memory architecture so AI workloads and regular desktop apps can run side by side.
This is Microsoft’s first major Windows-centered event since the Copilot+ PC launch in May 2024. NVIDIA has confirmed there will be no Windows 12 at the event — the theme is local AI and RTX Spark.
Tonight’s watch list: the Surface Laptop Ultra’s final price and availability, whether the first six RTX Spark laptops get prices too, and what Microsoft means by the “new experiences for developers and builders” teased in the announcement.