Microsoft's Surface Laptop Ultra brings Nvidia's Arm-based RTX Spark chip to Windows, from $2,599

Microsoft and Nvidia used a Windows and Surface event in San Francisco on Wednesday, 7 October, to launch the first PCs built on RTX Spark, Nvidia's Arm-based system-on-chip, also known as N1X. Microsoft's own model, the Surface Laptop Ultra, is available to preorder now and is due on 16 October, the date Nvidia gives for RTX Spark laptops in general; compact desktops are to follow in November. The Register summed up the launch in its headline as Microsoft dropping Intel in favour of Nvidia's chips for this Surface, while The Verge describes RTX Spark as Nvidia's entry into Arm-based PC processors. On stage, Microsoft CEO Satya Nadella and Nvidia CEO Jensen Huang talked about how the two companies are designing hardware and software together so that AI agents can run on Windows PCs.

The Verge and The Register give a starting price of $2,599 for a configuration with an 18-core CPU, 24GB of memory and a 512GB SSD; The Register puts the top model, with the 20-core chip, 128GB of memory and 1TB of storage, at $5,899. The 15-inch touchscreen uses a mini-LED panel, keeps Microsoft's usual 3:2 aspect ratio and peaks at 2,000 nits. Ports include three USB-C, full-size HDMI, an SD card slot and a single USB-A connector, according to The Register. One of the USB-C ports takes a charging cable that attaches magnetically, which Microsoft calls Magnetic Connect; The Verge notes that the port still handles data and video like any USB-C port. Microsoft says the cooling system has 2.5 times more capacity than in earlier Surface notebooks, and the SSD can be replaced by the user, The Register reports.

Pavan Davuluri, Microsoft's executive vice president for Windows and Devices, said the laptop was built around RTX Spark and that "you can run models on this laptop that simply don't fit on a traditional machine". Nvidia's own specifications list a Blackwell RTX GPU with as many as 6,144 cores, a Grace CPU with up to 20 cores, a 600 GB/s connection between the two, up to 128GB of unified memory and one petaflop of FP4 AI performance. Heise explains that the chip comes in two versions: a full one with 20 CPU and 6,144 GPU cores, and a reduced one with 18 CPU and 5,120 GPU cores that can address at most 32GB of memory, so any configuration with 64GB or more uses the full chip. Heise and The Register both link the design closely to GB10, the chip in Nvidia's DGX Spark mini PC, and The Register says the CPU was designed by MediaTek and Arm. It also questions the headline figure: the petaflop relies on 2:4 sparsity, which hardly any current inference workloads use, so the real peak is closer to 500 teraFLOPS, still ahead of AMD's Strix and Gorgon Halo chips.

Nvidia lists Acer, Asus, Dell, HP, Lenovo, MSI and Gigabyte alongside Microsoft as makers of RTX Spark systems, which were first shown at Computex in the middle of the year, heise notes. At launch, heise counts two Asus ProArt laptops with 14- and 16-inch screens and one large-screen model from each of the others: the Dell XPS 16, HP OmniBook Ultra 16, Lenovo Yoga Pro 9 and MSI Prestige N16 Flip. In the first shop listings heise checked, the Surface Laptop Ultra was the cheapest option at EUR 2,900 with 24GB and 512GB; Asus, HP, Lenovo and MSI models with 32GB and 1TB all appeared at an identical EUR 3,300, 64GB versions sat around EUR 4,300, and an Asus ProArt 16 with 128GB and 2TB reached EUR 6,700. Heise treats 16 October as the earliest possible date, as some retailers currently show shipping only in the second half of November.

For developers, Microsoft opened preorders for the Surface RTX Spark Dev Box, a mini PC priced at $5,999 that is due to ship in November. The Verge reports it has 128GB of unified memory and a 100-watt thermal envelope and is meant for local AI work such as testing models and prototyping agents. The Register notes that it costs less than the similarly specified DGX Spark, now $6,950, although that machine has 4TB of storage rather than 2TB. The event also gave a preview of DGX Station for Windows, a deskside system built on the GB300 Grace Blackwell Ultra Desktop Superchip, offering 748GB of coherent memory and as much as 20 petaFLOPS of FP4 compute. Until now DGX Station ran on Linux; Nvidia says Linux AI tools remain reachable through WSL. The sources give no price or release date for it.

On the software side, Microsoft announced general availability of Microsoft Execution Containers (MXC), which Nvidia describes as infrastructure at the operating-system level that lets agents keep running in the background under the control of Windows. "We are building these primitives directly into Windows, so agents can be secured, observed and governed," Davuluri said. The Verge reports that Copilot is getting access to local files and the ability to carry out actions across the operating system, part of a concept Microsoft calls Hybrid Intelligence, in which tasks are split between local and cloud AI models; a new search experience also offers quick actions from the search bar. The Register says the hybrid features pick an on-device or a cloud model depending on connectivity, privacy, cost and complexity, and will reach Windows 11 in the coming months. Davuluri also said Meta's AI agent Muse will get a native Windows app, without giving a date.