Microsoft Build 2026: Aion 1.0 On-Device AI, Cross-Silicon Windows APIs, and the New Surface RTX Spark Dev Box

Windows AI APIs: Breaking Free of the NPU

Until now, many of Windows' advanced AI features were locked to Copilot+ PCs with dedicated NPUs. Microsoft shattered that limitation at Build 2026 by expanding its Windows AI APIs to run on CPUs and GPUs, opening up on-device AI to a massive new install base across hardware from AMD, Intel, Qualcomm, and Nvidia .

The expansion is immediate and tangible. Phi Silica, Microsoft's on-device language model, is now available on GPUs. Video Super Resolution (VSR) and live captions can now run on CPUs . A new Speech Recognition API, available in preview, delivers real-time, offline speech-to-text on the NPU or CPU, enabling dictation, captioning, and accessibility tools without an internet connection . This API is initially launching in English .

Underpinning this cross-platform strategy is Windows ML, the built-in AI inferencing runtime that handles model execution across any available processor. At Build 2026, Microsoft integrated Windows ML with Microsoft Foundry on Windows, creating a full lifecycle platform for model selection, fine-tuning, optimization, and deployment . To reinforce the hardware ecosystem, AMD announced updated NPU and GPU Execution Providers for Windows ML, delivering up to a 1.5x improvement in time-to-first-token on NPUs and a 3.5x increase in sustained token generation .

Surface RTX Spark Dev Box: A Petaflop on Your Desk

The hardware hero of Build 2026 was the Surface RTX Spark Dev Box, a compact desktop workstation designed for serious local AI workloads . Its design draws comparisons to the Xbox Series X—compact, aluminum-bodied, and fanless, with the chassis itself acting as a 100W heatsink .

The core of the machine is Nvidia's new RTX Spark ARM-based superchip, which delivers up to 1 petaflop of AI compute paired with 128 GB of unified memory shared between CPU and GPU . This gives developers enough horsepower to run large language models with up to 120 billion parameters locally, with support for a 1-million-token context window .

It ships with a purpose-built, developer-optimized Windows 11 Pro configuration. Visual Studio Code, GitHub Copilot in Windows Terminal, Windows Subsystem for Linux (WSL), and PowerShell 7 all come preconfigured and ready out of the box . Microsoft says the machine is "designed for sustained workloads: long-running training jobs, agentic AI pipelines and local model fine-tuning" . The Surface RTX Spark Dev Box is expected to ship in the US later this year; pricing has not yet been announced .