Microsoft’s first laptop built on Nvidia silicon is now in reviewers’ hands, and the verdict is complicated. The Surface Laptop Ultra, powered by the Nvidia RTX Spark N1X superchip, is the most capable local-AI laptop anyone has tested, and also a machine that shut itself off twice in one lab.
The reviews landed within days of each other from StorageReview, The Shortcut, CNET, and PCWorld, with WindowsForum compiling a detailed megathread. Together they describe the same machine from different angles: a genuine breakthrough for on-device AI, wrapped in first-generation rough edges.
What the RTX Spark N1X actually is
The N1X is an Arm-based superchip that pairs an Nvidia Grace CPU, up to 20 cores split between 10 Cortex-X925 performance cores and 10 Cortex-A725 efficiency cores, with an integrated Blackwell RTX GPU of up to 6,144 cores and unified LPDDR5 memory. Because the memory is shared, the user can allocate up to 96GB of it to the GPU on the top configuration, more than any Nvidia laptop GPU before it. Microsoft claims up to 1 petaflop of FP4 AI performance and says top configurations can run local models above 120 billion parameters.
For background on the machine’s positioning and launch pricing, see our earlier Surface Laptop Ultra overview.
The config ladder runs $2,600 to $5,900
| Tier | CPU / GPU | Memory | Storage | Price |
|---|---|---|---|---|
| Entry | 18-core / 5,120-core | 24GB | 512GB | $2,599.99 |
| Reviewed by CNET, PCWorld, Shortcut | 20-core / 6,144-core | 32GB | 1TB | $3,700 |
| Top memory tier | 20-core / 6,144-core | 128GB | 1TB | $5,099 to $5,900 |
A few quirks stand out. Want the 2TB SSD? Microsoft restricts it to the 64GB memory tier. Step up to the full 128GB for heavy local inference and internal storage is capped at 1TB, for as much as $5,899.99. Microsoft has not confirmed India pricing, but at current exchange rates the range works out to roughly Rs. 2.3 lakh to Rs. 5.3 lakh. To soften the sticker shock, Microsoft is offering up to $1,000 back on MacBook Pro trade-ins between October 7 and November 23, 2026.
89 tokens a second from a 35-billion-parameter model
This is where the N1X earns its keep. StorageReview’s lab ran a 35-billion-parameter coding agent locally at up to 89 tokens per second, with gpt-oss:20b and Qwen3.6-35B-A3B generating between 58 and 89 tokens per second. For comparison, a Razer Blade 16 with an RTX 5090 is about 2.3 times faster on workloads that fit in its 24GB of VRAM, but it simply cannot load the larger models at all.
The CPU impresses too. The 20-core N1X won nearly every multi-threaded test in StorageReview’s comparison group, usually by 30% or more, against Intel Panther Lake and Arrow Lake H, AMD Strix Point, and Nvidia’s own RTX 5090 and RTX PRO 2000 laptop GPUs. The Shortcut’s Kevin Lee calls it the most flexible modern PC: an all-day work machine that can also game with DLSS 5 and path tracing, and run local AI without spending a cent on tokens. He judges it the better machine for developers, power users, and gamers, while conceding the MacBook Pro remains the better pick for most people.
The local-model angle matters beyond benchmarks. Open-weight models keep improving, most recently with Meta’s 816-billion-parameter Muse Spark Open release, and a laptop that can actually run serious models locally is a genuinely new category of machine.
Two review units died in the lab
Now the caveats, and they are significant. StorageReview reports that two separate Surface Laptop Ultra review units suffered sudden, hard power-offs during a sustained all-core double-precision floating-point workload, three shutdowns in one evening on the second unit after the first had to be replaced. The lab had not seen that failure mode before.
Other trade-offs: memory latency is more than double the fastest x64 competitor, and Windows on Arm’s storage stack throttles random 4K SSD throughput to less than a fifth of the raw drive’s capability. Battery life trails the MacBook Pro by a clear margin, which stings at these prices. The 15-inch mini-LED display draws praise for brightness and color accuracy, but reflections and blooming undercut it for professional color work, and there is no stylus support.
Who should buy it, and who should not
Buy it if you are a developer or researcher who needs more local AI memory than any other laptop offers, and you understand you are buying into a first-generation platform. Skip it if gaming is the priority, where a dedicated gaming laptop costs less and performs better, or if you want maximum battery life and polish, where the MacBook Pro still wins. For everyone in between, the $3,700 32GB configuration that CNET, PCWorld, and The Shortcut tested looks like the sensible middle of a very tall ladder.
Sources: StorageReview; The Shortcut; CNET; PCWorld; WindowsForum; Microsoft.
