Microsoft unveiled the Surface Laptop Ultra back on June 1, 2026. Microsoft is only now, in October 2026, actually telling us what it costs. Those two facts sit about four months apart, and the gap between them is the most interesting thing about this launch.
Because when a company shows you a product and then goes quiet about the price for a third of a year, that silence is usually doing work. The price, as of this week’s Windows and Surface event, is $2,599.99 for the model with a 5120-core GPU and an 18-core CPU, or $3,699.99 if you want the 6144-core GPU and 20-core CPU. Preorders are open on Microsoft’s site. You get Platinum or Nightfall. That’s the whole menu.
What Microsoft Is Actually Selling
The silicon is Nvidia’s RTX Spark chip, and Nvidia’s own framing is that it “delivers amazing creating, AI development, and gaming” in slim laptops and small desktops. Microsoft’s pitch, per its own announcement video, is aimed at “world makers and creative pros who refuse to choose between power and portability.”
Strip the marketing off and you have a thin laptop with a lot of GPU in it, sold to people who want to run heavy workloads without a desk. The same RTX Spark chips also show up in the Surface RTX Spark Dev Box, a separate machine The Verge’s Tom Warren got hands-on with at Build earlier this year. Two devices, one chip family, one obvious strategic intent: Microsoft wants the hardware you run local AI work on to have a Surface logo on it.
Where I Get Skeptical
I review AI tools for a living, which means I spend a lot of time watching companies describe capability in adjectives instead of numbers. “Amazing creating, AI development, and gaming” is three categories and zero benchmarks. Nvidia is not a company that struggles to produce performance charts when the performance is good. The absence of them in the public messaging so far is not proof of anything, but it’s a gap I’d want filled before spending $3,700.
Here’s what I can’t tell you from the verified information available, and what nobody buying this should pretend to know yet:
- How much memory the thing has, and how much of it the GPU can actually address — the single most important number for anyone running models locally
- Sustained thermal performance in a slim chassis, which is where every ambitious thin-and-powerful laptop historically goes to die
- Battery life under real AI workloads, as opposed to idle or video playback
- Whether the software stack you already use runs on it without a weekend of fighting drivers
- What the $1,100 jump between the two configurations buys you in practice beyond 1024 extra GPU cores and two CPU cores
That last one bugs me most. On paper, the step up is roughly a 20% core increase on the GPU side for a 42% price increase. Core counts are a bad proxy for real performance, so that ratio might be unfair in either direction. But Microsoft is the one asking for the money, so Microsoft is the one who should explain the math.
The Strategic Read
Set the specs aside and the move makes sense. Microsoft spent the last few years putting Copilot into everything and renting AI compute by the hour. Local AI hardware is the hedge. If inference keeps moving toward devices — for latency, for cost, for privacy, for all three — then owning the premium device tier matters. Partnering with Nvidia rather than waiting on its own silicon to catch up is the fast way there.
The risk is that $2,599.99 is a professional tool price for a category that hasn’t finished proving it needs a professional tool. Creative pros doing video and 3D work have a clear, measurable reason to want more GPU. “AI development” on a laptop is a fuzzier buyer. Plenty of people doing serious model work still train in the cloud and use whatever machine is in front of them as a terminal. Microsoft is betting that changes. It might. It also might stay a niche inside a niche.
My Advice Right Now
Don’t preorder. Not because this looks bad, but because you have no information to judge it with. You have two prices, two colors, four spec numbers, and a company’s description of its own target customer. That’s a pitch, not a review.
If you are a creative professional whose work is GPU-bound and whose current machine is actively costing you hours, put this on the watch list and wait for independent testing — thermals under sustained load, memory configuration, and real throughput on the specific tools you use. If you’re buying it because you want to run models locally and the idea sounds cool, wait longer. The gap between “this chip is good at AI” and “this laptop is good at your AI work” is where a lot of expensive disappointment lives.
Microsoft took four months to talk about the price. Taking a few weeks to see the benchmarks seems fair.
🕒 Published: