Microsoft put an Arm-based Nvidia processor inside a Surface laptop and asked $2,599.99 for the privilege. Microsoft also, as of the October 7 event in San Francisco, has not given anyone outside its own labs a reason to believe that number is justified.
Both of those things are true at once, and that gap is the entire story of the Surface Laptop Ultra.
What was actually announced
The hardware details are straightforward enough. The Surface Laptop Ultra went up for preorder on Microsoft’s site in two colorways, Platinum and Nightfall, in two configurations:
- $2,599.99 for the 5120-core GPU and 18-core CPU model
- $3,699.99 for the 6144-core GPU and 20-core CPU model
Inside both is the Arm-based Nvidia RTX Spark, a chip Nvidia says “delivers amazing creating, AI development, and gaming” on slim laptops and smaller desktops. Microsoft’s target is explicit: Apple’s MacBook Pro, across AI work, video workloads, and gaming. Satya Nadella and Jensen Huang shared the stage. The machine itself was not a total surprise, having made a brief appearance at Build earlier in the year, where it was billed as the most powerful Surface ever made.
That is the announcement. Notice what is missing from it.
Core counts are not performance
Five thousand one hundred and twenty GPU cores is a big number. It is also a number that tells you almost nothing on its own. Core counts without clock speeds, memory bandwidth, thermal headroom, or sustained-load figures are marketing arithmetic. Two laptops with identical core counts can differ by a wide margin depending on how long they can hold boost before the chassis gives up.
This matters more than usual for AI work specifically. Local model inference is dominated by memory bandwidth and available memory, not raw compute. If you are running a quantized model on-device, the question is how much memory the thing has and how fast it moves data, not how many cores are listed on the spec sheet. Microsoft’s announcement, as reported, leads with cores. Draw your own conclusion about which numbers were the most flattering.
The Arm problem has not gone away
Here is the part that should give any professional buyer pause. This is an Arm-based Windows machine. Microsoft has been shipping Arm Windows devices for years, and compatibility has improved considerably, but the AI and creative tooling space is where the rough edges live.
Think about what an actual AI developer workflow touches: Python environments with compiled native extensions, CUDA-dependent libraries, Docker images built for x86, model runtimes with hand-tuned kernels, plugin ecosystems in video editors that vendors update on their own schedule. Every one of those is a place where an Arm build either exists, runs through emulation at a performance penalty, or does not run at all. Nvidia being the silicon partner is genuinely interesting here, because nobody has deeper hooks into the AI software stack. But partnership announcements do not recompile anyone’s dependency tree.
Apple went through this transition and ate years of pain to get through it. The difference is that Apple controlled the whole stack and could force the ecosystem to follow. Microsoft is asking a more fragmented Windows ecosystem to do the same work, on a chip that is new to the market.
Who should preorder this
Almost nobody, yet. Not because the machine is bad, since nobody can honestly say whether it is bad or excellent at this point, but because preordering unbenchmarked silicon at $2,599.99 is paying full price to be a beta tester. The $3,699.99 configuration is a $1,100 jump for roughly 20% more GPU cores and two more CPU cores, which is the kind of pricing curve that only makes sense if you already know the workload scales with it. You do not know that. Neither do I.
What I would want before spending that money:
- Sustained performance numbers under real workloads, not short benchmark bursts
- Memory capacity and bandwidth figures, stated plainly
- An honest compatibility list for the specific AI, video, and gaming software you depend on
- Independent reviews from people who did not get a briefing
The part I am actually optimistic about
Credit where it is due: Microsoft and Nvidia going after the MacBook Pro directly is good for everyone who buys laptops. Apple has had the high-end creative and AI-development segment largely uncontested, and uncontested segments get expensive and complacent. A serious Windows competitor with Nvidia silicon behind it changes that math, even if this first attempt lands imperfectly.
First-generation hardware on new architecture is historically where the compromises hide. Wait for the reviews. The preorder button will still be there, and if the machine is as good as the stage show suggested, so will the laptop.
🕒 Published: