Microsoft and NVIDIA have announced a new class of Windows PCs capable of running advanced AI models entirely on-device, without a cloud connection. This development could fundamentally alter how users interact with artificial intelligence on personal computers.
The announcement was made recently at NVIDIA GTC in Taipei, centred on a new chip called RTX Spark and a new Surface laptop built around it.
120 billion parameters, no server required
The platform is capable of running up to 120-billion-parameter AI models locally — on the device itself, with no data leaving the machine. For context, models of that scale are comparable to some of the most capable AI systems available commercially today, which currently require data centre infrastructure to run.
RTX Spark delivers 1 petaflop of AI performance, with up to 128GB of unified memory and up to 6,144 Blackwell RTX cores.
Privacy by default
Microsoft and NVIDIA have worked to unlock the power of the GPU for local AI workloads through Windows ML, enabling AI developers to leverage TensorRT natively in Windows. Because processing happens on the device, user data — documents, prompts, files — does not pass through any external server.
That directly addresses one of the most consistent concerns around AI adoption in professional environments, where sending sensitive information to a third-party cloud service has remained a significant barrier.
Works offline. No subscription needed.
Microsoft also implemented Workload Profile Scheduling tuned specifically for RTX Spark, enabling the Windows scheduler to more efficiently scale workloads across all 20 cores — whether for email or running a local AI agent to debug code. 
The system functions without an internet connection and carries no recurring access cost for on-device inference, unlike cloud-based AI tools that operate on subscription models.
The agent framing
Both companies positioned the announcement around a broader shift in personal computing. Jeff Fisher, Senior Vice President of Personal Computing at NVIDIA, said: “NVIDIA and Microsoft share a vision that agents are the future of personal computing. RTX Spark combines NVIDIA’s full technology stack with Microsoft Windows and is purpose-built for creators, gamers and AI developers in the personal AI era.”
The announcement also marks a milestone in what Microsoft described as a multi-year, full-stack collaboration with NVIDIA spanning gaming, AI and cloud — from DirectX and RTX to NVIDIA-accelerated AI workloads on Azure.
The Surface Laptop Ultra, the first device built on RTX Spark, is expected to be available later in 2026. Pricing has not been disclosed.
