Story
October 9, 2026

Microsoft’s $2,599 AI Laptop Tries to Make Windows the Place AI Work Happens

Microsoft and Nvidia see local AI as a practical answer to cloud cost, latency and enterprise workflow friction; Microsoft’s harder challenge is proving that this premium hardware-and-agent strategy can succeed where its earlier AI-PC push stumbled.

Microsoft’s latest AI-PC gamble began well before Wednesday’s San Francisco launch. In 2024, its Copilot+ PC initiative got off to a rocky start after Recall, its signature feature, was delayed—an awkward backdrop as rivals raced ahead in frontier models and personal AI agents.

Microsoft first previewed the Surface Laptop Ultra in May, then filled in the crucial details this week: the Nvidia RTX Spark-powered machine starts at $2,599, is available to preorder and begins shipping on Oct. 16. It brings up to 128GB of unified memory and up to one petaflop of AI compute—specifications aimed at developers, creators and other users who want demanding models close at hand rather than permanently rented from the cloud.

The argument from Microsoft and Nvidia is straightforward: local computing can cut lag, avoid recurring cloud bills and let people run models that conventional laptops cannot accommodate. Windows chief Pavan Davuluri said the machine can run models that “simply don’t fit on a traditional machine.” Nvidia’s Jensen Huang called the RTX Spark-equipped Ultra “a beast,” while framing the partnership as the next phase of a long Windows-Nvidia relationship.

But this is not an anti-cloud pitch. Microsoft calls the approach “Hybrid Intelligence”: Copilot and other tools should choose between local and cloud models without forcing users to make that decision themselves. In a demonstration, an Autopilot agent searched folders for tax documents, renamed and zipped them, then drafted an email—activity Microsoft says can be performed locally when heavy cloud compute is unnecessary.

That aspiration carries a familiar risk. Windows users previously complained about Copilot’s constant presence, and some of the newly shown agent functions are not yet available. Microsoft is now pairing its promise of more autonomous software with execution containers intended to limit agents’ system access. Nadella nevertheless cast the moment in sweeping terms, saying Windows would bring “unmetered intelligence to every desk and every home” while letting agents work securely for users.

The hardware bet is expensive—Microsoft’s developer-focused RTX Spark Dev Box will cost $5,999—but the company is wagering that enterprises will pay to keep AI workloads, data and workflows nearer to the Windows machines they already use.

Story coverage