Last Updated on October 7, 2026 by Spencer Lee
Microsoft’s Surface Laptop Ultra is designed around NVIDIA’s RTX Spark platform, combining a Blackwell RTX GPU with up to 128GB of unified memory and full CUDA support. Microsoft says the laptop can deliver up to 1 petaflop of AI compute and run models with up to 120 billion parameters locally. A Windows and Surface event is scheduled for October 7.
Why is the Surface Laptop Ultra different?
The Surface Laptop Ultra is Microsoft’s first Surface laptop built around NVIDIA silicon designed specifically for high-performance local AI.
Microsoft says the system combines an NVIDIA Blackwell RTX GPU with an efficient CPU, up to 128GB of unified memory and full CUDA support.
That gives the laptop a very different AI profile from conventional Windows laptops that rely on a discrete GPU with a smaller dedicated VRAM pool.
Why does 128GB of unified memory matter for AI?
The large memory pool is primarily important for running larger AI models locally.
Microsoft says the unified memory can be dynamically allocated across the CPU and GPU, allowing compatible workloads to use more of the system’s memory when needed. The company says the Surface Laptop Ultra can run AI models with up to 120 billion parameters locally.
That does not mean every 120B model will run at the same speed or with every configuration. Model quantization, context size, software support and workload requirements can all affect actual performance.
How much AI performance does it offer?
Microsoft rates the Surface Laptop Ultra at up to 1 petaflop of AI compute.
The company also highlights a thermal system designed for sustained high-performance workloads, with up to 2.5 times the thermal capacity of the Surface Laptop 15-inch 7th Edition.
The laptop is therefore being positioned for developers, creators and AI users who need more than the typical productivity laptop can provide.
What can the Surface Laptop Ultra do locally?
Microsoft highlights several local AI applications, including AI-assisted code completion, video upscaling, intelligent masking and AI-powered noise reduction in supported applications.
The larger unified-memory architecture is also intended for local AI agents and larger models, while Microsoft says users can still move to cloud-based models when they need frontier-scale AI capabilities.
This hybrid approach could make the laptop useful for users who want local processing for privacy, latency or experimentation while retaining access to cloud AI when necessary.
Is the Surface Laptop Ultra available yet?
Microsoft has announced the Surface Laptop Ultra, but it remains a pre-release product and Microsoft has not yet published final availability information on its product page.
Microsoft is holding a Windows and Surface event on October 7, where the Surface Laptop Ultra and RTX Spark are expected to be major parts of the discussion.
Until Microsoft confirms final pricing, availability and configurations, those details should not be treated as official.
What does this mean for AI laptops?
The Surface Laptop Ultra shows where high-end Windows AI laptops could be heading: large unified-memory pools combined with powerful GPU acceleration.
That approach could be particularly useful for local LLMs and other AI workloads where memory capacity becomes a bigger limitation than conventional laptop performance.
The important test will come when independent reviewers can benchmark the final hardware with real AI models and workloads.
What happens next?
Microsoft’s October 7 event should provide more information about the Surface Laptop Ultra and Microsoft’s broader Windows strategy for local AI.
For LaptopHub readers, the most important details to watch are final pricing, configurations, availability and independent AI performance testing.
Also read: NVIDIA RTX Spark Brings 128GB Memory to AI Laptops
RTX Spark is the underlying platform that makes Microsoft’s Surface Laptop Ultra particularly interesting for local AI. Our earlier article explains the 128GB unified-memory architecture and why NVIDIA believes it can support much larger local AI models.
