Microsoft unveiled the Surface Laptop Ultra starting at $2,599 alongside Nvidia RTX Spark hardware during an October 7 event in San Francisco, bringing desktop-class unified memory and local AI agent capabilities to a 15-inch portable workstation scheduled to begin shipping on October 16.
Microsoft Prices Surface Laptop Ultra At $2,599
The base configuration costs $2,599 and includes an Nvidia RTX Spark N1X chip with a 5,120-core GPU, an 18-core CPU, 24GB of unified memory, and a 512GB SSD. Preorders for the Surface Laptop Ultra are open now, with customer shipments scheduled to begin on October 16. Buyers looking for maximum performance can configure the device up to $5,899.99 for higher specifications, while a trade-in promotion offers up to $1,000 cash back for qualifying used MacBook Pro devices through November 23.
The 15-inch portable workstation weighs 4.4 pounds and is built around an aluminum unibody chassis. It features a PixelSense Ultra display with a 3,270×2,180 resolution, a 120Hz refresh rate, and mini-LED image quality. Connectivity options include three USB4 ports, a USB-A port, an HDMI port, an SD card reader, and a headphone jack. Microsoft also introduced Magnetic Connect, described as a new magnetic charging design. Pavan Davuluri, Microsoft’s executive vice president of Windows and devices, stated that Microsoft has simplified product naming to lead with the Surface brand while continuing to deliver premium hardware and hybrid AI experiences. Alongside the Surface Laptop Ultra, new devices designed around the RTX Spark platform are coming from ASUS, Dell, HP, Lenovo, MSI, Acer, and Gigabyte.
Nvidia RTX Spark Architecture Powers Local AI
The Nvidia RTX Spark platform pairs a Blackwell RTX GPU boasting up to 6,144 cores with an up to 20-core Grace CPU connected at 600 GB/s. The architecture supplies up to 1 petaflop of AI performance and supports up to 128GB of unified memory. Pavan Davuluri noted that the configuration lets users run models on this laptop that simply don’t fit on a traditional machine. Jensen Huang highlighted that the Blackwell RTX Spark GPU inside the computer matches the processing power of Nvidia’s DGX-1 data center server from 2016, which cost $250 million. NVIDIA founder and CEO Jensen Huang appeared at the Windows AI and Surface event in San Francisco alongside Microsoft CEO Satya Nadella for a fireside chat moderated by Sriram Krishnan, a former senior White House policy advisor on AI. Huang remarked during the discussion that if not for Windows, there would be no GeForce. The presentation also previewed the NVIDIA DGX Station for Windows, described as a deskside supercomputer bringing GB300 Grace Blackwell-class AI infrastructure directly into the Windows ecosystem, delivering 20 petaflops of FP4 AI compute and supporting large AI models up to one trillion parameters.
“Microsoft might get me back to Windows”
Max Weinbach, analyst at Creative Strategies, via CNBC
Developers who need the workstation performance without a built-in display or keyboard can purchase the Surface RTX Spark Dev Box.
Microsoft Introduces Execution Containers For AI Agents
Microsoft introduced Microsoft Execution Containers (MXC) to address security concerns regarding autonomous AI agents operating across local files. Available to developers today, MXC provides a sandboxing policy layer that directs compute traffic to a process container, session container, WSL container, or hardware-backed MicroVM, allowing administrators to set strict guardrails on agent access. Microsoft frames these hardware releases around a software strategy called Hybrid Intelligence, designed to move away from cloud-backed AI solutions and toward local ones.
Windows 11 has been re-engineered for the new silicon, featuring task scheduler modifications that distribute work across all 20 Arm CPU cores of the top RTX Spark chip. Satya Nadella and Jensen Huang outlined the joint vision during their San Francisco keynote, emphasizing that the desktop must be the most secure place for agents to execute. Pavan Davuluri noted that whether an agent works under user direction or acts on its own, every action taken should be attributed directly to the agent rather than the user.
