Microsoft and NVIDIA want your next laptop to run AI agents on the device, not in the cloud. RTX Spark and a new Windows sandbox called MXC are the pitch.
RTX Spark Brings Local AI Agents to Windows PCs
NVIDIA and Microsoft used an October 7 event in San Francisco to announce RTX Spark, a new superchip that powers a line of Windows laptops built to run AI agents locally. The flagship is the 15-inch Surface Laptop Ultra, which is up for preorder now and starts shipping October 16 at $2,599.
The pitch is that these machines move AI work from the cloud onto the PC. Instead of paying per query to a remote service, the laptop runs the model itself. So who is this actually for?
Microsoft framed the shift as moving the PC from a tool to a teammate. Whether that lands depends on what these laptops can actually do, so let's go through it.
What RTX Spark Can Run on a Windows Laptop
The headline number Microsoft keeps repeating: RTX Spark machines can run models with more than 120 billion parameters locally. That's the class of model that normally lives in a data center.
The catch is that "can run" comes with conditions. Big local models need enough unified memory, careful quantization, and a manageable context window. Miss any one and you hit a wall. NVIDIA's spec sheet points to up to 128 GB of unified memory and up to a petaflop of AI compute on the Surface Laptop Ultra, which is what makes a 120B model plausible on a laptop. Run the same model with a huge context and you'll feel the ceiling.
NVIDIA lists creative and gaming workloads alongside the AI angle: rendering 90 GB+ 3D scenes, editing 12K video, generating 4K AI video, and playing AAA games at 1440p past 100 frames per second. This is a premium machine aimed at people who push their hardware.
Specs aside, does it change how you'd work?
That depends on you.
Inside Microsoft MXC for Agent Containment
The more consequential Windows change may be MXC, short for Microsoft Execution Containers. Microsoft moved it out of beta and into general availability at the event.
MXC is an operating-system-level feature that gives an AI agent its own controlled environment. It sets boundaries and policies on what the agent can touch: which files, which network resources, which parts of Windows. It's a sandbox. The agent does its job without roaming your whole machine.
That matters because agents are being handed real tasks that involve real data. This past summer saw several high-profile AI lab safety incidents, and agent boundaries moved from a niche worry to a mainstream one. MXC is Microsoft's answer, and it's the kind of plumbing that decides whether people trust an agent with anything sensitive.
Pricing, Availability, and the Dev Box
The Surface Laptop Ultra is the consumer face of the launch at $2,599. A Surface RTX Spark Dev Box, aimed at developers, starts at $5,999. Preorders are open, with shipping from October 16.
That's a steep price for a laptop, and it's the honest trade in this story. You're paying for local compute that replaces a cloud subscription, plus the privacy of keeping data on your own machine. For a developer or a creator who runs models constantly, the math can work. Not so for everyone.
Microsoft also said its GitHub HydraFusion routing technology is extending to local Windows models, which is how the system decides whether a request should run on the device or go out to a cloud model. Sensible design.
RTX Spark laptops are up for preorder now, with the first units shipping October 16 at $2,599. That's the date to remember. If you're weighing one against a cloud subscription, start with the agent tasks you'd actually run daily, then check whether the local model handles your document sizes and context needs. The sandboxing story is the part to watch as MXC reaches more machines.
