Microsoft and Nvidia open sales on first "agent-ready" Windows PCs, starting at $2,599

Microsoft and Nvidia used a joint event in San Francisco on Tuesday to open pre-orders on the first wave of computers built around Nvidia's RTX Spark chip, a part the companies describe as a scaled-down version of the silicon inside Nvidia's data-center AI systems. The centerpiece is the Surface Laptop Ultra, a $2,599 machine aimed at creative and technical professionals, alongside a $5,999 desktop box built for software developers. Both devices go on sale this month, and both are being pitched less as faster laptops than as the first consumer hardware designed around AI programs, or "agents," that run locally rather than in the cloud.

Nvidia chief executive Jensen Huang and Microsoft chief executive Satya Nadella appeared together on stage, a pairing Microsoft has used sparingly in recent years, to frame the launch as a reset of what a Windows computer is for. Pavan Davuluri, Microsoft's executive in charge of Windows and devices, and Brett Ostrum, the corporate vice president who leads the Surface hardware group, also spoke or published remarks tied to the announcement, which Microsoft and Nvidia each confirmed in posts published the same day.

The hardware and the price

The Surface Laptop Ultra starts at $2,599 for a configuration built around Nvidia's RTX Spark Superchip, which pairs a Blackwell-generation RTX graphics processor with up to 6,144 cores and an Arm-based Nvidia Grace CPU with up to 20 cores, linked by an interconnect Nvidia rates at 600 gigabytes per second. Microsoft's listing for the laptop specifies up to 128 gigabytes of memory shared between the CPU and GPU, up to 2 terabytes of user-replaceable storage, and a 15-inch mini-LED touchscreen rated at 2,000 nits of peak brightness, which Microsoft says is roughly 25 percent brighter than Apple's MacBook Pro with the M5 Pro chip. The chassis is under 18 millimeters thick and weighs less than 4.5 pounds, with a cooling system Microsoft says carries two and a half times the thermal capacity of the previous 15-inch Surface Laptop. Pre-orders opened Tuesday, with shipments beginning October 16; Microsoft is also offering up to $1,000 in trade-in credit through November 23 to buyers who turn in an eligible MacBook Pro.

The second device, the Surface RTX Spark Dev Box, is a compact desktop built for programmers rather than general consumers. It carries the same RTX Spark chip, the same 128 gigabytes of unified memory ceiling, and Nvidia's claimed one petaflop of AI performance at low ("FP4") numerical precision, inside a 100-watt, fan-cooled aluminum enclosure. It ships preloaded with a developer-oriented build of Windows 11 Pro, including Visual Studio Code, Git, GitHub Copilot and the Windows Subsystem for Linux, and Microsoft says it can run AI models with more than 120 billion parameters entirely on the machine rather than over the internet. It costs $5,999, sold only through Microsoft's own online store in the United States, with shipments beginning in November. By comparison, Nvidia's own equivalent machine, the DGX Spark, has sold since earlier this year for around $3,999, and AMD's rival "Ryzen AI Max" desktops occupy similar territory — a reminder that Microsoft's device is priced at the premium end of a category that barely existed a year ago.

How a data-center chip ended up on a desk

RTX Spark is not a new idea so much as a repackaging of one. Nvidia first showed the chip, built with help from MediaTek and manufactured on a 3-nanometer process at TSMC, at the Computex trade show in Taiwan in June, describing it as a cousin of the GB10 Grace Blackwell superchip that powers the company's DGX Spark "personal AI supercomputer." The pitch then, as now, was that the same architecture Nvidia sells to cloud providers and research labs for training large AI models could be shrunk enough to fit inside a laptop, letting individual developers and creators run sizeable AI models on their own machines instead of renting time on someone else's servers. Nvidia has said it expects more than 30 laptops and roughly a dozen desktop systems using RTX Spark to reach the market this year from manufacturers including Acer, Asus, Dell, HP, Lenovo, MSI and Gigabyte, in addition to Microsoft's own Surface line.

Tuesday's event folded that hardware rollout into a broader argument about software. Nadella described plans to treat the Windows desktop as a secure environment for autonomous AI agents, including new "execution containers" meant to wall off what an agent can touch and an extension of Microsoft's Entra identity system to track agent activity separately from a human user's. Huang, for his part, tied the announcement back to Nvidia's own history with Microsoft's operating system, according to Nvidia's own account of the event.

"What happens in the era of agents, when the agent is on your computer? It is now your personal assistant," Huang said, according to Nvidia's recap of the San Francisco appearance, published by the company the same day.

Who stands to be affected

The immediate audience is narrow: buyers willing to spend between $2,600 and $6,000 on a Windows machine, a group that skews toward video editors, engineers, independent software developers and AI researchers rather than everyday shoppers. Microsoft's trade-in offer and its explicit price comparisons to Apple's hardware make clear it is chasing MacBook Pro owners in particular, continuing a rivalry that has defined the high end of the laptop market for close to a decade. Software developers are a second, more specific audience for the Dev Box, which Microsoft places inside what it calls its "Project Zenith" family of developer-focused hardware and positions against both Nvidia's own DGX Spark and compact AI workstations from AMD.

The launch also matters to the dozen-odd PC makers that signed on to build RTX Spark machines of their own, since Microsoft's pricing sets an early public benchmark for how the rest of the category will likely be priced when Acer, Asus, Dell, HP, Lenovo, MSI and Gigabyte ship their own versions later this year. And it is a data point for Nvidia's broader strategy of moving beyond data-center chips into consumer and workstation hardware, an area where it has historically supplied components to others rather than entire branded systems.

Reaction and the skeptics

Public reaction so far has split along predictable lines. Early hands-on coverage has focused on build quality and raw specifications rather than independent performance testing, since retail units have not yet reached outside reviewers; an early hands-on account from Tom's Guide described the laptop as capable of running recent games at high settings during a controlled demonstration, while cautioning that nothing about its performance under sustained load has been independently verified. Nvidia's one-petaflop figure refers to a low-precision computing mode used mainly for AI inference, not a number comparable to conventional benchmark scores, a distinction the companies' own materials acknowledge in footnotes. Coverage of the companion Dev Box has been more measured so far, with early commentary noting that it enters a market already occupied by Nvidia's own DGX Spark at a lower price, leaving Microsoft to justify the premium largely on the strength of its Windows-integrated developer tools.

What happens next

The nearest test comes quickly: Surface Laptop Ultra units begin shipping October 16, which will put the machine in front of independent reviewers for the first time and settle questions about battery life, thermal throttling and real-world AI performance that neither company's own materials can answer. The Dev Box follows in November. Over the same stretch, rival RTX Spark machines from other PC makers are due to reach shelves, which will show whether Microsoft's pricing holds up as a market standard or proves to be an outlier at the top of the range. Further out, Nvidia used the same event to preview a larger desk-side machine, the DGX Station for Windows, built around a more powerful Grace Blackwell Ultra chip, suggesting the current launch is an opening step in a longer plan to move data-center-style AI hardware progressively closer to individual desks rather than a one-time product cycle.