Microsoft's New Laptop Runs Big AI Without the Cloud

Microsoft's Surface Laptop Ultra is the first laptop to use Nvidia's RTX Spark chips, Nvidia's first chips for Arm-based PCs. The Verge
Testing happens in a plain, windowless lab on Microsoft's campus in Redmond, Washington. Robots press buttons thousands of times. Big antennas blast devices with radio signals. The work is slow and careful. It is meant to find breaks and wireless problems before buyers do.
The goal is different from Microsoft's last Arm effort. The 2024 Copilot Plus PCs started with Qualcomm chips and were aimed at the MacBook Air. Microsoft said those earlier chips were not strong enough for MacBook Pro-level work. Brett Ostrum is corporate vice president of Surface. Jit Hirani is lead designer for Surface devices.
Microsoft and Nvidia spent years building this laptop together after Nvidia privately showed what its future chips could do for AI.
The broader context here is the long timeline, which points to joint design of the electronics and body, not a last-minute chip change.
Silicon and thermals
The RTX Spark puts the main processor and graphics processor together and uses unified memory. Think of it as one large shared worktable instead of two small desks that force you to carry papers back and forth. That setup lets large AI jobs on the laptop use more memory than a separate graphics chip would normally allow.
Microsoft says the chip in this laptop can run very large, compressed AI systems with 280 billion settings locally, without the cloud. That claim will need outside testing with details on compression type, conversation length, and speed.
Looking at what this means for regular use, the pattern is simple. Running huge AI systems on a laptop is mostly about having enough memory and moving data fast, less about raw math speed.
Putting the two chips together let Microsoft use a smaller main board and fit bigger fans. Smaller board, bigger fans. That leaves more room to stay cool during long jobs. Short bursts are easy, like a sprint. Long AI writing sessions are like a marathon. Fan size and airflow decide if the laptop keeps speed or slows down to cool off after a few minutes.
Chassis and serviceability
The Surface Laptop Ultra has a body shaped from aluminum by machine tools, with a sharp touchscreen. It includes an HDMI port and an SD card slot. It is less than 18mm thick and weighs under 4.5 pounds (2 kg). It comes in Platinum. Microsoft
The broader context here is that those ports matter for the planned buyer. HDMI and SD mean no extra adapters in meeting rooms or for moving photos and video. The size and weight make it a portable workstation, not a super-light travel laptop.
Repair gets unusual attention. No parts are glued down, and the laptop has printed guides and QR codes that explain how to open it. That helps trained repair shops and simplifies mail-in repair. Screws and labels do not guarantee cheap parts or parts for many years, but they remove the first barrier.
Positioning and price
A model with 24GB of memory starts at $2,599.99. The top version with 128GB costs $5,899.99. Microsoft is aiming it at developers who want local AI power and at creative professionals who would buy a MacBook Pro.
The broader context here is a split in Windows laptops. The earlier Arm machines focused on battery life and daily tasks. This one focuses on heavy AI work on the device itself. With shared memory, the memory size you buy is the power level you get.
In my view, the practical questions matter most. What steady speed does it hold on a mid-size compressed model versus a model over 200 billion. How much of the 128GB is truly free for AI data during real work. How loud the fans get. How well everyday developer tools, containers, drivers, and testing programs run on Arm.
Looking at what this means for buyers, the appeal is control. Running AI on the laptop keeps files, questions, and private data on the device. That avoids internet delays and per-use online fees. It also means the user handles tuning. Online services hide that work. A local machine shows it. That tradeoff has long fit a certain user, one who watches memory and heat and accepts extra setup for steady results. If Microsoft and Nvidia have made that loop simpler, with joint design, bigger fans, easier repair, and familiar ports, the Ultra name fits how it works and not only its specs.


