SEATTLE — One hundred twenty billion parameters. That number is the headline inside Microsoft’s new Surface RTX Spark Dev Box, a desktop PC the company unveiled Tuesday. It is the ceiling for what a developer can run locally on this machine without phoning home to a cloud server. Microsoft says the device is built around Nvidia’s Arm-based RTX Spark chip. It is compact. It is aimed squarely at artificial intelligence engineers.
The 120-billion-parameter figure matters because it changes the workflow. Most large AI models today live in the cloud. Developers send data out, wait for a response, pay for compute time. The Spark Dev Box keeps everything on the desk. That means faster iteration. It also means sensitive model weights never leave the room. Microsoft is betting that security and speed together will pull developers away from remote servers.
The machine ships with Windows 11, Visual Studio Code, the Windows Subsystem for Linux and PowerShell 7. That is a standard developer stack, not a custom one. Microsoft is not asking engineers to learn a new environment. It is giving them the tools they already use, but on hardware designed for local AI. The company says the box can handle models up to that 120-billion-parameter threshold. That covers a wide range of current open-source models, including many that previously required a cluster or a rented GPU instance.
This is part of a broader push across the industry. Apple has been selling Macs with unified memory and neural engines for local AI. Qualcomm is pushing its Snapdragon X chips for on-device processing. Microsoft’s play is different. It is not putting AI in a laptop or a phone. It is building a dedicated desktop for one job. The Surface RTX Spark Dev Box is a tool, not a general-purpose computer. It is a workbench.
The implications for security are direct. A developer working on a proprietary model for a bank or a hospital does not have to trust a cloud provider with the data. The model trains and runs on the box. That reduces exposure. It also reduces latency. A developer can test changes instantly, without a round trip to a data center. Microsoft says the machine is designed for local AI development. That phrase — “local AI development” — is the whole pitch.
There are limits. One hundred twenty billion parameters is a lot, but frontier models now run into the hundreds of billions and beyond. The Spark Dev Box cannot run those. It is not trying to. It targets the engineer who needs to build, test and refine models that fit within that boundary. That is a large and growing audience. Open-source models at that scale are becoming standard tools for real-world applications.
The device also reflects a strategic bet by Nvidia. The RTX Spark chip is Arm-based. Nvidia has been pushing Arm into data centers and edge devices for years. This is the first time it appears in a Surface product. The partnership gives Microsoft a custom chip for AI work and gives Nvidia a high-profile consumer-facing showcase for its Arm design.
Microsoft did not announce pricing or a release date. The company said the machine is part of a broader industry push toward powerful local AI hardware that reduces dependence on cloud services. That is the trend. The Spark Dev Box is one version of it. The 120-billion-parameter limit is the number to watch. It defines what kind of work this machine can do. It also defines what kind of developer will buy one.
