At Computex 2026 in Taipei on Monday, Microsoft did something no one in the Surface division has done in years: it built a laptop that looks like it means it. The Surface Laptop Ultra, packing NVIDIA’s new RTX Spark system-on-a-chip, is a 15-inch slab of aluminum with 2.5 times the thermal headroom of last year’s Surface Laptop 7, a MiniLED display pushing 2,000 nits in HDR, and a port selection — three USB-C, full-size HDMI, SD card reader, USB-A, headphone jack — that reads like a wish list from anyone who has ever dangled a dongle off a MacBook Pro.

It is also, by any honest measure, a workstation wearing a laptop costume. And that’s the part worth paying attention to.

NVIDIA’s RTX Spark is not the kind of silicon you spec for a thinner device. It packs 6,144 Blackwell GPU cores and 20 Arm CPU cores onto one package, and NVIDIA claims it can hit a petaflop of AI performance — in a chip that draws up to 80 watts under load. For context, Apple’s M-series Max chips, the ones that go into the $3,000-plus MacBook Pros, top out around a third of that raw AI number. But they also don’t require the kind of cooling solution that forces a product manager to use phrases like “2.5x the thermal headroom” with a straight face.

The Thin-and-Light AI PC Was Always Marketing Vapor

For two years, the industry has sold a vision of AI computing that runs on whispers. Qualcomm’s Snapdragon X Elite. Intel’s Lunar Lake. AMD’s Ryzen AI. Every single launch event featured a slide showing a 45 TOPS NPU inside a fanless chassis, with a product manager insisting that local AI — real, useful, inference-at-your-fingertips AI — could live on a 15-watt power budget.

It couldn’t. It can’t. The models that matter — the large language models that developers actually run locally, the diffusion models that generate production assets, the CUDA-dependent libraries that power scientific computing — all saturate memory bandwidth and GPU compute. They don’t fit neatly into a neural processing unit that exists mostly to blur your webcam background and summarize three sentences of text.

The Surface Laptop Ultra says the quiet part out loud: serious local AI compute requires serious silicon and serious cooling. The chip draws single-digit watts at idle — comparable to any Arm laptop — but that 80-watt ceiling is the real story. That’s not a smartphone chip that grew up. That’s a laptop GPU that learned to sip power when it’s not working.

One engineer at a competing PC maker, reached in a Slack message during the Computex keynote, put it plainly: “Everyone who shipped an AI PC last year shipped a Chromebook with a better NPU. This is the first one that ships with a real GPU. The difference is the difference between checking the box and running the workload.”

What the MacBook Pro Comparisons Miss

Predictably, the coverage has framed this as a MacBook Pro challenger. Microsoft itself invited the comparison — three USB-C ports, MiniLED display, premium aluminum build, fall 2026 release window. The spec sheet is drawn to overlap with Apple’s on as many lines as possible.

But frame it as a MacBook Pro rival and you miss what Microsoft is actually doing here. Apple’s 16-inch MacBook Pro with an M4 Max — the closest analogue — is a machine for video editors, software developers, and creative professionals whose workflows are well-served by Apple’s vertically integrated stack. It runs cool. It runs quiet. It cannot, and will not, run CUDA.

The Surface Laptop Ultra runs CUDA. That one fact reorients the entire product. It means this machine is aimed not at the Final Cut Pro user but at the machine-learning engineer who wants to iterate on models locally. At the data scientist whose pipeline depends on libraries that Apple Silicon still can’t touch without translation layers and performance penalties. At the researcher who has been dragging a Zephyrus or a Razer Blade to conferences and pretending it’s a professional tool.

This is not a laptop for everyone. It’s not even a laptop for most professionals. It is a laptop for people who know exactly how many GPU cores they need and why. And that audience — small, well-funded, exacting — has been underserved by the thin-and-light-forever design philosophy that captured the Windows ecosystem after the M1 MacBook Air made everyone panic in 2020.

The Surface Finally Stops Apologizing

The broader significance here isn’t about Microsoft versus Apple at all. It’s about Microsoft finally letting the Surface division build a machine that doesn’t apologize for its power draw. The Surface line has spent a decade chasing thinness as a proxy for desirability — the Surface Pro X, the Surface Laptop Studio, the various ARM experiments that promised battery life at the cost of compatibility. Each one was a hedge. Each one compromised on something real to win on something cosmetic.

The Ultra doesn’t hedge. It picks a lane — high-performance local AI compute, CUDA compatibility, real GPU muscle — and accepts the tradeoffs. It will be heavier than a MacBook Air. It will be louder under load. Microsoft hasn’t even announced pricing yet, but anyone who has bought an NVIDIA GPU in the last three years can guess the ballpark, and it won’t be cheap.

That’s fine. The machine that tries to please everyone ends up disappointing the one group that was willing to pay a premium for exactly what they needed.

There is a lesson here that extends beyond laptops. The AI industry spent 2024 and 2025 promising that intelligence would be ambient, invisible, distributed across every device, running on negligible power. The Surface Laptop Ultra is a useful corrective: sometimes the future needs a fan.

Sources