Tech

Microsoft and Nvidia want $2,599 to put AI on your lap

The Surface Laptop Ultra runs AI models on the device, starting at $2,599. It’s built for a market that’s still forming, and a Windows that wants to act for you.

Published

12 hours ago

October 21, 2026

Massih Shahbazi

A backlit laptop keyboard in the dark, in black and white, beside the headline AI you can carry

For years, the pitch for AI has been that it lives in the cloud: you type, a data center somewhere thinks, and an answer comes back. Microsoft and Nvidia now want to sell you the data center. It folds shut, it has a 15-inch screen, and it starts at $2,599.

The Surface Laptop Ultra was announced on Oct. 7, with preorders opening the same day and sales starting Oct. 16, Windows Central reported. It’s built around Nvidia’s new RTX Spark platform and designed to run sophisticated AI models directly on the device instead of relying entirely on remote servers, TechRepublic reported.

What $2,599 buys you

The entry model pairs an 18-core Nvidia Grace CPU with a 5,120-core Blackwell GPU, 24GB of unified memory and a 512GB SSD, at $2,599.99, according to Notebookcheck, citing Microsoft’s configuration page. Higher configurations go up to 20 CPU cores, 6,144 GPU cores and 128GB of memory, and one listing puts a 128GB model with a 1TB drive at $5,899.99, Notebookcheck reported.

The rest is premium-laptop standard: a 15-inch mini-LED display, a haptic trackpad and three USB-C ports, according to GSMArena.

That memory number is the tell. Large AI models need a lot of fast memory to run locally, and 128GB of unified memory is the kind of figure you’d normally see in a workstation, not something that fits in a backpack.

Not just a Microsoft story

The Surface isn’t alone. RTX Spark laptops from Asus, Dell, Lenovo, HP and MSI also went on sale Oct. 16, VideoCardz reported, and HP briefly listed upcoming OmniBook models starting at $2,999. This is Nvidia moving from selling chips to other companies’ computers to defining a whole new class of laptop.

The launch also lined up with major changes to Windows 11 aimed at what Microsoft calls agentic AI, software that takes actions on your behalf instead of just answering questions, Tom’s Hardware reported. The hardware and the operating system are being pitched as two halves of the same idea.

Nvidia’s benchmark, Nvidia’s numbers

Nvidia says RTX Spark systems can run AI workloads faster than a MacBook Pro with Apple’s M5 Pro chip, based on its own preliminary testing, according to Tech Insider. Those are company figures, not independent benchmarks. Every chipmaker’s launch-day chart shows its own product winning. Wait for reviewers with stopwatches before taking that one to the bank.

The comparison is still revealing. Apple’s laptops have owned the “run big AI models locally” conversation because of their unified memory design. Nvidia picking the MacBook Pro as its target tells you exactly which customers it wants: developers, researchers and creators who currently carry a Mac for this work.

Who actually needs this

Here’s the honest question nobody’s launch event answers: does a normal person need to run AI on their laptop?

For most people, no. Chatbots, writing tools and photo editors work fine through the cloud on a laptop that costs a third as much. The case for local AI is narrower but real. It’s for people who work with sensitive files they don’t want leaving the machine, people who build and test AI models themselves, and people who need AI tools to work on a plane or anywhere the connection is bad.

There’s also a privacy argument that the industry is only starting to make loudly. When the model runs on your device, your prompts and documents don’t have to travel to someone else’s server. Microsoft and Nvidia haven’t promised that every AI feature will stay local, so buyers should check which tools actually run on the machine and which still phone home.

The software half of the bet

The timing with Windows matters more than the spec sheet. Agentic AI means software that doesn’t just answer a question but goes off and does things: opens files, fills in forms, works across apps. That kind of assistant needs constant access to what’s on your computer. Running it locally, on a chip built for the job, is one way to make that feel less like handing your whole digital life to a server you’ll never see. Whether Microsoft keeps that promise is the part buyers should watch most closely.

The price of owning the brain

The Surface Laptop Ultra is impressive on paper, expensive in practice and aimed at a market that’s still forming. Its top configurations cost more than double its starting price.

That’s the trade the whole industry is now offering. You can rent your AI from a data center for a monthly fee, or you can buy it outright and carry it around. For the first time, a mainstream laptop brand is betting that enough people will pay to own the brain instead of renting it. The next few months of sales will show whether that bet was early or right.