Microsoft's Surface Laptop Ultra puts Nvidia's RTX Spark platform into a laptop marketed for local AI work. The company's product page lists a starting price of $2,599.99. TechCrunch also reports Microsoft's move toward Nvidia-based PCs for locally running models and agents.
A starting price is not the whole configuration
The Verge places the new RTX Spark laptops alongside premium MacBook Pros and names models from Microsoft, Dell, Lenovo, MSI, HP and Asus. Its report describes starting configurations with an 18-core CPU, 24GB of unified memory and 512GB of storage at roughly $2,599.
Those entry specifications should be read separately from Microsoft's advertised maximums. Microsoft lists configurations reaching a 20-core Grace CPU and 128GB of unified memory. A maximum specification does not mean the least expensive version includes it.
The Verge's comparison also identifies a storage difference: the similarly priced 14-inch M5 Pro MacBook Pro includes 1TB. Readers comparing these systems should compare the exact memory and storage options rather than assuming similar prices buy equivalent configurations.
What the local AI pitch establishes
Microsoft describes CUDA-accelerated agent workflows and support for models up to 120 billion parameters. That is the manufacturer's capability claim, not a Terminal Press benchmark. Microsoft also notes that performance depends on the workload and configuration.
The memory architecture matters to interpreting the headline number. Microsoft's footnotes say unified memory is shared between CPU and GPU, with GPU-addressable memory below total system capacity. The 128GB figure therefore should not be read as a promise that an application has that entire amount available for its model.
Questions beyond the headline
Microsoft lists a 15-inch touchscreen, HDMI, an SD card reader, USB-A and three USB-C ports. It also describes upgradeable storage and SSD options up to 2TB, with replacement components and service availability depending on the market.
Our comparison checklist is straightforward: identify the exact configuration, confirm the software you need, and look for measurements of your own workload. This report establishes announced specifications and reported pricing; it does not establish battery endurance, model response speed or performance under sustained load. Those questions need evidence beyond a product announcement before a purchase recommendation would be justified.
