TL;DR
- Surface Device: Microsoft has introduced Surface RTX Spark Dev Box for developers building local AI agents.
- Hardware Range: DGX Station for Windows and RTX Spark PCs extend local AI testing toward larger model workloads.
- Security Runtime: OpenShell adds sandboxing and policy checks before agent actions reach files, networks or host processes.
- Availability Test: RTX Spark systems are slated for 2026 and fall 2026 rollouts across Microsoft and OEM channels.
Microsoft introduced Surface RTX Spark Dev Box at Build 2026 as a compact developer PC for local AI agents built with NVIDIA RTX Spark hardware.
The dev box is optimized for software that can plan steps, call services and act on data.
Surface and DGX Hardware Bridge Local and Cloud AI
Microsoft introduces the Surface RTX Spark Dev Box as the local Windows endpoint for agent routes. Designed as a compact developer PC, the device combines an Arm-based CPU, a Blackwell-generation RTX GPU and 128GB of shared memory.
Microsoft’s device spec lists up to 1 petaflop of AI compute. Unified memory helps larger models fit on one device, while Windows 11 Pro arrives preconfigured with Developer Mode, GPU-passthrough WSL 2, CUDA support, VS Code, GitHub Copilot, Git, Python and Node.js. Sales are planned in 2026 in the United States through Microsoft.com.
NVIDIA is placing its DGX Station for Windows in the same product family. For enterprise desks, the system uses the GB300 Grace Blackwell Ultra Desktop Superchip, with up to 748GB of coherent memory and 20 petaflops of FP4 performance. Back in 2025, NVIDIA still positioned the DGX Spark and DGX Station family mainly as desktop systems for large-model training and deployment but the focus now has shifted towards running local models and running AI agents. Its new RTX Spark Windows PC hardware adds a newer client-device layer to the mix.
Cloud infrastructure gives those local devices a scale-up counterpart. Across the stack, deployment turns location into a configuration choice across local and cloud infrastructure.
OpenShell Makes Agent Control the Security Test
Autonomous agents create a security problem because useful work often requires code execution, file access or network calls. NVIDIA OpenShell is the runtime layer meant to control that boundary, with integration into GitHub Copilot with sandboxing and policy enforcement for agent actions.
OpenShell creates isolated sandboxes for each agent and sub-agent, verifies permissions before execution and applies policy rules across filesystem, network and process layers. Its gateway evaluates what an agent can read, write, execute, reach over the network or delegate to external tools before actions touch the host environment.
Windows adds another control layer around local workloads. RTX Spark agent features are designed around OS-enforced identity, containment and manageability, with workload scheduling and memory changes for Arm-based RTX Spark PCs. Recent OpenShell work around secure runtime controls shows why the mechanism matters: controls that make agents manageable also deepen reliance on NVIDIA and Microsoft defaults.
RTX Spark Availability Sets the Competitive Test
RTX Spark laptops and compact desktops are slated for fall 2026 availability from ASUS, Dell, HP, Lenovo, Microsoft Surface systems and MSI, with Acer and GIGABYTE models to follow. OEM breadth keeps Surface from being the whole Windows story and makes the platform pitch larger than one Microsoft device.
Competition remains narrower than a general enterprise-agent race. RTX Spark systems extend beyond Surface to ASUS, Dell, HP, Lenovo and MSI, while Apple’s Mac mini and Apple Silicon systems remain local-compute comparators. Microsoft and NVIDIA are emphasizing model work, agent controls and CUDA-compatible workflows rather than only a compact workstation with unified memory.
