Surface Laptop Ultra Starts at $2,599 With RTX Spark

Microsoft has opened US preorders for Surface Laptop Ultra, turning the RTX Spark laptop it introduced earlier in 2026 into a priced retail product. The consumer model starts at $2,599.99 and Microsoft lists the release for October 16, 2026.

The entry configuration combines an NVIDIA RTX Spark N1X with an 18-core CPU and 5,120-core Blackwell RTX GPU, 24GB of LPDDR5x unified memory, a 512GB SSD and Windows 11 Home. Microsoft lists the higher N1X with a 20-core CPU and 6,144-core GPU from $3,699.99.

Surface Laptop Ultra offers 24GB, 32GB, 48GB, 64GB and 128GB of unified LPDDR5x memory. Unlike a conventional gaming laptop with separate system RAM and dedicated GPU VRAM pools, RTX Spark is designed around memory that can be allocated across CPU and GPU workloads. That is particularly relevant to local AI and large creative workloads whose working sets can exceed the VRAM capacity of ordinary notebook GPUs.

Microsoft says the platform can deliver up to one petaflop of peak FP4 AI compute on select configurations and run models up to 120 billion parameters locally. Those are platform/vendor capability claims, not independent Core Tech Tips performance measurements, and actual model capacity and speed depend on configuration, software, quantization and workload.

Microsoft's current US configuration page lists Windows 11 Home on the 24GB configuration and Windows 11 Pro on 32GB, 48GB, 64GB and 128GB models. Storage begins at 512GB, while 1TB and 2TB configurations use removable Gen 5 SSDs according to Microsoft's specification table.

The laptop has a 15-inch PixelSense Ultra touchscreen rated for up to 1,000 nits SDR and 2,000 nits peak HDR, Wi-Fi 7, HDMI 2.1b, USB-A, USB-C/USB4 ports, a full-size SD card reader and a 3.5mm audio jack. Microsoft lists a weight of 4.41 pounds (2.0kg).

The retail milestone matters because RTX Spark is moving from an announced platform into shipping Windows PCs. Independent testing after October 16 should show how the Arm-based Grace CPU, Blackwell RTX GPU and shared memory architecture behave across native Windows software, translated applications, CUDA and local-AI workloads.

Until those measurements exist, specifications and vendor demonstrations should not be converted into general claims about gaming performance, battery life or superiority over x86 laptops and Apple silicon.