Microsoft Surface Laptop Ultra high-memory model sells out amid local AI push

Microsoft is expanding its push to run GitHub Copilot locally on user hardware, a move intended to reduce cloud computing reliance. However, technical disclosures reveal significant hardware requirements for these local AI workloads. On the Surface Laptop Ultra, the local MAI Code 1.1 Flash model has a 53GB footprint, with peak memory usage reaching 75.5GB when utilizing a 256,000-token context.

This high memory demand has impacted product availability. The highest-capacity configuration of the Surface Laptop Ultra, which features 128GB of unified memory and a 1TB SSD, has sold out through Microsoft’s United States store. This model is part of the emerging RTX Spark category, utilizing NVIDIA hardware and CUDA-compatible software to support AI tasks that exceed the capacity of standard laptop GPUs.

While the 128GB unified-memory design allows the accelerator to address much larger workloads than mainstream laptops, the high hardware and management costs may influence customer adoption. Competitors, such as AMD’s Ryzen AI Max 400 family, offer unified-memory configurations reaching up to 192GB.