Surface Laptop Ultra: Microsoft Prices Its Nvidia AI PCs
In June, Nvidia announced that Microsoft and several other PC makers had agreed to build Windows machines around its RTX Spark chip. The pitch was "AI- and agent-ready" PCs. Almost no hardware details came with it. That changed on Wednesday, when Microsoft took the stage at an event in San Francisco during the city's Tech Week and gave specs and prices for its first RTX Spark devices.
There are two new Surface products, an updated Windows 11 with a new agent sandboxing feature, and a trade-in offer aimed directly at Mac-using developers.
Two machines, one target audience
The headline device is the Surface Laptop Ultra. It comes in two base configurations:
- A standard model starting at $2,600
- A model with a more powerful chip starting at $3,700
Adding memory and storage pushes the price up to $5,900. Microsoft says that top configuration has already sold out.
The second product is a workstation, the Surface RTX Spark Dev Box, which starts at $6,000. It ships with Microsoft's developer tooling already installed: VS Code, GitHub Copilot CLI, WSL (the Windows Subsystem for Linux, which lets developers run Linux environments inside Windows) and PowerShell 7.
Microsoft is selling both machines on the same idea. They are built to run AI models locally, on the device, without paying for cloud inference. The CPU, GPU, unified memory and upgraded cooling are all meant to support that kind of sustained workload. For developers already experimenting with local coding agents, this is a direct pitch: keep the model on your desk instead of in someone else's data center.
Microsoft also says the graphics hardware is strong enough for content creation, video processing and gaming. The marketing focus, though, is clearly on developers.
Windows 11 gets "Execution Containers"
The more interesting announcement may be the software. Both devices run a revamped version of Windows 11 that includes a feature called Execution Containers. According to Microsoft, it makes it easier to sandbox AI agents, which means isolating them so they can only reach what they are supposed to reach.
Microsoft CEO Satya Nadella said at the event that the feature will be available to all Windows 11 users, not only buyers of the new hardware.
Nadella also explained how the company now thinks about agents. "One of the things we realized in the last 3-4 years is that just having a model doesn't do much for anything," he said. "You really do need to orchestrate, and you need to have memory outside of the model. You need to have this harnessed layer that is able to take multiple models plus context, plus memory, and the action space."
He said Microsoft is building that layer "not just for our apps but apps for anyone's agent," and added: "That's what the Windows platform is all about going forward."
Going after the MacBook crowd
Microsoft is not hiding who it wants to win over. Developers who trade in a MacBook Pro can get up to $1,000 off. Macs have become a common choice among developers and AI agent enthusiasts, especially in Silicon Valley, and this offer is an attempt to pull some of them back to Windows.
Dell joins in
Microsoft was not the only PC maker with news on Wednesday. Dell gave details on its previously announced Dell XPS 16 Creator Edition, which also uses Nvidia's RTX Spark chip. It is available for pre-order now at Best Buy for $3,800, with delivery expected later in October.
The Bigger Picture
The hardware is expensive and clearly aimed at a niche. The more important part of this launch is the operating system. Nadella's comments suggest Microsoft sees the model itself as only one piece. The value sits in the layer around it: orchestration, memory and controlled access to actions. Execution Containers is the first concrete piece of that strategy, and shipping it to every Windows 11 user, not just Surface buyers, shows Microsoft wants Windows to be the default place where agents from any vendor run.
This fits a wider pattern. Operating system vendors are starting to treat agents as untrusted software that needs firm boundaries. Apple, for example, recently tightened macOS Full Disk Access because of agent risks. Isolation at the OS level does not depend on how well a model behaves, so it is a sensible place to put a security boundary. The open question is how strict these containers will actually be, and whether developers will accept the friction or look for ways around it.
On the hardware side, local inference at these prices is still a choice for professionals, not a mass-market feature. It is worth watching whether the MacBook trade-in offer moves real numbers of developers, and whether more PC makers follow Dell with RTX Spark machines at lower prices. Nvidia's growing role in shaping how agents run, including its open agent safety platform, also makes it worth tracking how closely Microsoft's sandboxing work ties into Nvidia's own stack. Independent tests of Execution Containers will be the first real check on whether the security claims hold up.