Microsoft and Nvidia used a joint event in San Francisco on October 7 to make their biggest push yet for AI on the personal computer. The centerpiece is the 15-inch Surface Laptop Ultra, priced at $2,599 and built around Nvidia's RTX Spark platform — a combination of a Grace CPU and Blackwell RTX GPU that Microsoft says delivers up to a petaflop of FP4 AI performance, with configurations offering up to 128GB of unified memory. Preorders opened the same day, with shipments starting October 16.
The specs matter because they change what a PC can do on its own. Microsoft says the machine can run AI models with more than 120 billion parameters locally, including DeepSeek V4 Flash, which the company describes as a 284-billion-parameter model, alongside its own MAI Code 1.1 Flash and Nvidia's upcoming Nemotron. Onstage, Microsoft showed the laptop running GitHub Copilot and Gears of War: E-Day at 60 frames per second.
This is also Nvidia's move into territory long dominated by Intel, Qualcomm, and AMD chips — its first flagship play for the Windows PC. RTX Spark is not Surface-only: Acer, ASUS, Dell, HP, Lenovo, MSI, and Gigabyte all plan laptops and compact desktops on the same platform. Microsoft also opened preorders for a $5,999 Surface RTX Spark Dev Box, a compact desktop aimed at developers, with shipments scheduled for November.
The hardware is only half the story. Microsoft's pitch, in CEO Satya Nadella's words, is "hybrid intelligence": software that decides task by task whether to run locally or call the cloud, so users get responsive on-device AI without giving up the reach of bigger cloud models. That matters because local compute is fast, private, and cheap per query — no network round trip, no per-token billing, and sensitive files never leave the machine.
The other major announcement is about trust. Microsoft said Execution Containers are now generally available on Windows 11. They isolate autonomous AI agents — software that acts on your behalf, opening files and using tools — and let organizations specify exactly which files and networks an agent may touch, enforcing those rules while it runs. "We needed to make the desktop the most secure place for agents to execute," Nadella said onstage, with Nvidia CEO Jensen Huang beside him. As agents graduate from chat windows to actually doing things on a computer, that guardrail layer is arguably the most consequential part of the announcement.
Some skepticism is in order. The $2,599 starting price makes this a premium professional proposition, not a mass-market AI PC. Broad adoption will depend on what the partner manufacturers do with the platform, and whether developers build agents people actually want running locally. And Microsoft's local-model claims are manufacturer benchmarks until independent tests land.
Still, the direction of travel is clear. Two years ago, the "AI PC" meant a neural processor that ran background blur. Now Microsoft and Nvidia are shipping a laptop that can hold a 100-billion-plus-parameter model in its own memory, an operating system that sandboxes the agents running on it, and tooling aimed at developers who want AI working in the background all day. The cloud is not going anywhere — but the most interesting computer in the house may soon be the one on the desk.