Microsoft's Surface Laptop Ultra is the first big Windows laptop built on Nvidia's RTX Spark platform, and it is aimed squarely at people who run AI models on their own machine. TechRadar's review calls it a beast at local AI and multi-threaded work, but a hard sell for almost everyone else. Here is what the testing shows and who should consider spending $2,599.99 or more.
Price and configurations
| 18-core model | 20-core model | |
|---|---|---|
| Starting price | $2,599.99 (24GB, 512GB) | Up to $5,899.99 (128GB, 1TB) |
| CPU cores | 9 Cortex-X925 + 9 Cortex-A725 | 10 Cortex-X925 + 10 Cortex-A725 |
| GPU | 5,120 CUDA cores | 6,144 Blackwell CUDA cores |
| Max memory | 64GB LPDDR5X | 128GB LPDDR5X |
Both versions use Nvidia's N1X chip on TSMC's 3nm process, with 9,400 MT/s memory and up to 300GB/s of bandwidth. TechRadar reports lead times for the top model stretching to late December, and rising DRAM and NAND costs are not helping prices; our guide to the 2026 memory shortage explains why.
Local AI performance: where it shines
Up to 96GB of memory can be assigned to the GPU, which is enough to keep models of up to 120 billion parameters entirely on the graphics processor. In TechRadar's agentic AI tests the Surface Laptop Ultra was the fastest system overall, finishing about 25% quicker than an M5 Max MacBook Pro and in roughly half the time of AMD's Ryzen AI Max+ 395. It generated up to 89 tokens per second on 35-billion-parameter models.
There is a catch: for models small enough to fit in 24GB, the RTX 5090 in a Razer Blade 16 generated tokens 2.3 times faster. The Ultra's advantage is capacity, not raw speed. If you are deciding what hardware you need for local models, start with our guide to running AI models locally.
CPU performance: strong multi-core, weak single-core
Multi-threaded results are excellent. The 20-core chip finished 34% ahead of the Razer Blade 16 in TechRadar's multi-core testing and compiled LLVM in 206 seconds, against 327 seconds for the next-fastest machine. Single-threaded performance is the weak spot, trailing x64 chips by 15 to 30% across most of the test suite, and memory latency was the worst of the group.
Gaming performance
Gaming is merely decent. TechRadar puts it roughly on par with an RTX 4070 or RTX 5060 laptop GPU, with Cyberpunk 2077 holding 60fps at 1440p only with frame generation turned on, and input latency is high. If games matter, a dedicated gaming laptop is the better buy; see our best graphics cards of 2026 for the GPUs worth targeting.
Should you buy the Surface Laptop Ultra?
- Buy it if you develop, test or analyze with local AI models larger than 24GB and want everything on one portable machine.
- Skip it if you mainly need fast single-threaded performance, video encoding or gaming.
- Consider a dedicated-GPU laptop if your models fit in 24GB of VRAM, since an RTX 5090 laptop is faster for those workloads.
How much does the Surface Laptop Ultra cost?
It starts at $2,599.99 for the 18-core model with 24GB of memory and 512GB of storage, and tops out at $5,899.99 for the 20-core model with 128GB of memory and 1TB of storage.
What chip does the Surface Laptop Ultra use?
It uses Nvidia's N1X chip from the RTX Spark platform, with up to 20 Arm CPU cores and up to 6,144 Blackwell CUDA cores.
Is the Surface Laptop Ultra good for gaming?
It is adequate rather than great. Gaming performance is roughly in line with an RTX 4070 or RTX 5060 laptop, with high input latency.
