NVIDIA and Microsoft Bring RTX Spark and On-Device AI Agents to Windows PCs

On Wednesday, at a Microsoft event in San Francisco, NVIDIA CEO Jensen Huang and Microsoft CEO Satya Nadella took the stage together to explain how the two companies are co-engineering hardware and software so that AI agents can run on Windows PCs. Huang said NVIDIA was founded because of Windows, and now AI agents are coming to Windows.

According to NVIDIA's blog, the centerpiece is RTX Spark, paired with a full stack meant to let agents run locally on Windows machines. The message is not a single chip specification but the idea of tying silicon and software together so that AI agents become part of the Windows experience.

That work is matched by an expansion of local inference on the Windows side. Reports say Windows announced that DeepSeek V4 Flash and NVIDIA's Nemotron can run locally, and that the popular open-source inference engine llama.cpp is coming to Windows ML. Together these move stronger models onto the device instead of the cloud.

Running models locally matters for privacy, latency and cost. Cloud calls require uploading data, waiting on a network round trip and paying by usage, while on-device inference keeps sensitive data on the machine, responds faster and suits always-on agent workflows.

For Microsoft, this is another step in pushing Windows from an application platform toward an AI platform. The company has spent the past year filling in the fundamentals of on-device AI, and this announcement binds local model support and inference tooling to NVIDIA's hardware story, pointing at one goal: making the PC the home turf for agents.

For NVIDIA, it is a chance to stretch the value of its GPUs from the data center to personal devices. Data-center demand is strong, but the PC is a vast installed base of its own, and if local AI becomes a reason to upgrade, the hardware refresh cycle could restart.

What to watch next is the pace of rollout: which Windows devices and price tiers can actually run these local models, how the developer experience changes once llama.cpp joins Windows ML, and how permissions and safety boundaries for agents on Windows will be defined.