Microsoft has opened preorders for the Surface Laptop Ultra, a 15-inch Windows machine starting at $2,599.99. The company calls it its most powerful Surface laptop and is pitching it at people who want to run demanding AI and graphics work directly on a PC.
The laptop is built around Nvidia’s RTX Spark platform. Microsoft says the package combines Blackwell RTX graphics, a Grace CPU with up to 20 cores and as much as 128 GB of unified memory. The computer weighs less than 4.5 pounds, measures 18 millimeters thick and includes a touchscreen, a magnetic USB-C charging connector and a mix of USB-C and USB-A ports.
The local-AI pitch
Microsoft says the Surface Laptop Ultra can run local AI agents and models with as many as 120 billion parameters. That is a manufacturer claim, not an independent performance test. The company’s product page also makes clear that the cloud remains part of the plan, suggesting customers reserve it for larger workloads rather than abandon it entirely.
Reuters reports that Microsoft’s broader strategy is to move some AI work from its costly Azure infrastructure onto higher-powered Windows PCs. Copilot would still use cloud systems for its hardest tasks, producing a hybrid setup rather than a fully offline one.
At the launch, Microsoft also unveiled an AI coding model designed to run on PCs and Microsoft Execution Containers, a security feature intended to restrict what an AI agent can access or do. A Microsoft executive told Reuters that Anthropic, OpenAI and Nvidia plan to use the container technology.
Premium hardware, premium price
The entry configuration comes with 24 GB of memory, according to Reuters. At the other end of the range, a $5,899 model pairs the 20-core processor with 128 GB of unified memory and 1 TB of storage.
That pricing makes the Surface Laptop Ultra a specialist machine rather than a mainstream default. Reuters noted that high costs remain a barrier as rising memory prices make powerful AI PCs harder to sell. Microsoft is betting that developers and other demanding users will pay to keep more work local, while still leaning on the cloud when a task outgrows the laptop.
