Microsoft event debuts new AI-friendly hardware and Windows changes

Microsoft previewed new hardware and Windows 11 changes aimed at local AI development and agent-driven workflows at its first live Surface event in two years. The Surface Laptop Ultra, previously teased in May 2026, is built around Nvidia's RTX Spark SoC and starts at $2,599. Microsoft highlighted configurations that include either a 5,120-core or 6,144-core Blackwell GPU and up to 128GB of LPDDR5x unified memory, and said the laptop will ship beginning October 16 and is available to pre-order. On the higher end, Microsoft introduced the Surface RTX Spark Dev Box, a $5,999 anodized aluminum unit described as a purpose-built machine for frontier developers. The Dev Box ships with a custom Windows 11 build tuned for AI development, includes 128GB of unified memory and an onboard Nvidia RTX GPU, and is billed as providing one petaflop (1,000 teraflops) of AI compute performance. Microsoft’s materials emphasize the Dev Box’s compact, console-like design and note that its chassis venting is a visual nod to that advertised compute level. Alongside hardware, Microsoft detailed Windows 11 changes that expand on-device AI capabilities. Copilot will be reorganized into three tabs — Home (for interacting with LLMs), Code (for prompting-driven creation of widgets and apps), and Autopilot (for overseeing AI agents). Windows Search will gain the ability to execute system-level tasks directly from the search bar, and certain OS builds on the new hardware will include software support for local model testing out of the box. To address containment and governance for agent activity, Microsoft announced Microsoft Execution Containers (MXC), a framework available to developers that confines agents to sandboxed environments and provides policy controls for admins. MXC can route workloads to different isolation backends, including process containers, session containers, WSL containers, or an experimental hardware-backed MicroVM. Microsoft described the broader strategy as "Hybrid Intelligence," promoting local compute and support for frameworks such as llama.cpp and models from Deepseek and Nvidia Nemotron families.