Surface Laptop Ultra review: a different kind of MacBook Pro

If you need to run complex AI models or games locally, this laptop now offers an option with greater privacy and potentially lower long-term costs.

Microsoft has launched the Surface Laptop Ultra, a high-end Windows on Arm laptop co-developed with Nvidia. This premium device, starting at $2,599 and reaching $5,899 for the top configuration, features Nvidia's new Arm-based RTX Spark chips designed to run demanding AI models, AAA games, and creative applications locally. The laptop boasts an all-aluminum design, a 15-inch 2000-nit MicroLED touchscreen display, a large haptic trackpad, and an array of ports including USB-C, USB-A, HDMI 2.1b, and an SD card reader. A unique Magnetic Connect feature utilizes the USB-C port for charging, snapping into place magnetically. Early reviews indicate that while the hardware is capable of rivalling Apple's MacBook Pro, effectively utilizing the local AI capabilities requires significant technical expertise or considerable experimentation, and some users have encountered issues with battery reporting, display lag, speaker interference, and slow reboot times.

The Surface Laptop Ultra represents a significant push for Windows on Arm, combining Microsoft's hardware expertise with Nvidia's powerful Arm-based RTX Spark GPUs. This collaboration aims to bridge the gap between high-performance computing and local AI processing, enabling users to run sophisticated models and demanding applications without relying on cloud services. The device's premium build and display specifications position it as a direct competitor to Apple's MacBook Pro, targeting a niche of developers and power users who prioritize local processing power and data privacy.

While the hardware is impressive, the user experience for local AI on the Surface Laptop Ultra is currently geared towards those with coding backgrounds or a high tolerance for troubleshooting. The practical application of running complex AI models locally often results in high resource utilization, loud fan noise, and extended processing times for even basic tasks, suggesting that the software ecosystem and AI model optimization for this platform are still in their nascent stages. Future developments will likely focus on simplifying the interface and improving the efficiency of local AI operations to make this powerful hardware more accessible to a broader audience.