Enough horsepower to run AI models entirely on the device
At Computex 2026, Microsoft unveiled the Surface Laptop Ultra — a machine that quietly marks a turning point in how the industry imagines the relationship between human and computer. Powered by Nvidia's RTX Spark, a single chip carrying the processing weight of what once required server rooms, the device promises to bring large-scale AI reasoning into a form factor light enough to carry on a shoulder. Whether this represents a genuine leap or an ambitious prologue remains a question only time and testing will answer.
- The stakes are unusually high: Microsoft is wagering its flagship Surface line on the premise that users want — and will pay for — AI that runs entirely on their own device, without the cloud.
- A single chip combining a 20-core CPU, RTX 5070-class GPU, and up to 128GB of unified memory is the engineering gamble at the center of it all, promising one petaflop of AI compute in under 4.5 pounds.
- The tension between ambition and reality is visible in the details — some configurations ship with only 16GB of memory, quietly undermining the offline AI promise Microsoft is loudly making.
- Press were not permitted to power the device on at Computex, leaving pricing, availability, and real-world performance shrouded in deliberate vagueness beyond a loose 'this fall' window.
- The industry is watching: the Surface Laptop Ultra is being positioned against a new tier of AI-first premium laptops, and the benchmarks to fairly judge them don't fully exist yet.
Microsoft arrived at Computex with the Surface Laptop Ultra, its most direct statement yet that the next era of personal computing belongs to machines capable of running artificial intelligence locally — no cloud required. At the heart of the device is Nvidia's RTX Spark, a new system-on-a-chip that fuses a 20-core CPU, a GPU on par with the RTX 5070, and up to 128 gigabytes of unified memory into a single piece of silicon. Microsoft claims this translates to one petaflop of AI compute — enough to run 120-billion-parameter models entirely on the device, a capability previously reserved for server hardware or elaborate multi-GPU rigs.
The physical form is deceptively familiar. At under 18 millimeters thin and less than 4.5 pounds, the all-metal chassis in Platinum or Nightfall finishes looks like a refined Surface Laptop at first glance. The real investment is in the 15-inch mini-LED PixelSense Ultra display — the brightest Microsoft has ever shipped in a laptop, reaching 2,000 nits in HDR with a 3:2 aspect ratio, 262 pixels per inch, and 120Hz variable refresh. Ports are generous, the keyboard familiar, and the touchpad is the largest haptic-enabled surface Microsoft has placed in a Surface device — and reportedly user-replaceable.
The thermal engineering required to sustain 6,144 CUDA cores in a thin chassis is substantial. Microsoft doubled the cooling capacity compared to its 2024 Surface Laptop 7th Edition, using dual fans and dual heat pipes with a redesigned airflow path — a detail the company illustrated at Computex using smoke. Internally, the SSD follows the standard M.2 2280 form factor and can be swapped, and the battery is accessible for eventual replacement, with QR codes linking to service guides in a nod to repairability. Memory, however, appears soldered to the board.
The caveats are meaningful. Entry configurations may ship with as little as 16 gigabytes of memory — insufficient for the offline AI experiences being marketed. Microsoft has not yet allowed the press to power the device on, has offered no pricing, and has committed only to a vague fall 2026 launch. The benchmarks needed to fairly evaluate this new class of machine are still being developed. For now, the Surface Laptop Ultra is a compelling promise — one that will only be tested when it reaches the hands of reviewers and, eventually, the people it's built for.
Microsoft brought the Surface Laptop Ultra to Computex this week, and it's the clearest signal yet that the industry is betting hard on a new class of machine: laptops built from the ground up to run artificial intelligence models locally, without sending your data to the cloud. The device is powered by Nvidia's freshly announced RTX Spark processor, a system-on-a-chip that combines a 20-core CPU, a GPU equivalent to an RTX 5070, and up to 128 gigabytes of unified memory into a single piece of silicon optimized for Windows on Arm.
The specs alone are staggering. Microsoft claims the Surface Laptop Ultra will deliver up to one petaflop of AI compute—enough horsepower to run 120-billion-parameter AI models entirely on the device. That's capability that has until now been confined to professionals with server farms or enthusiasts willing to build custom rigs with multiple graphics cards. The RTX Spark chip is a cousin of the hardware Nvidia built for its DGX Spark AI developer box, but retooled for consumer laptops and fitted with an NPU to handle Microsoft's Copilot+ AI features.
The physical design is understated. The laptop measures less than 18 millimeters thick and weighs under 4.5 pounds, wrapped in an all-metal chassis available in Platinum or Nightfall finishes. At first glance, it looks like any other Surface Laptop—the same polished Windows logo, the same refined aesthetic. But look closer and you'll notice the chassis has a raised design that makes the machine appear to float above the desk. The 15-inch display is where Microsoft has invested serious engineering: a mini-LED PixelSense Ultra panel with a 3:2 aspect ratio, capable of reaching 2,000 nits of peak brightness in HDR mode. It's the brightest display Microsoft has ever put in a laptop, with 262 pixels per inch and support for 120Hz refresh rates and variable refresh rate technology. The company says the mini-LED backlight will draw more power than traditional displays, but they're confident they can maintain all-day battery life.
The ports are comprehensive: full-size HDMI, USB Type-C, USB Type-A, a 3.5mm headphone jack, and an SD card reader. The keyboard is familiar—the same satisfying typing feel from previous Surface Laptops—and the touchpad is the largest haptic-enabled pad Microsoft has ever installed in a Surface device. Microsoft says the touchpad will be user-replaceable, and they're working with software partners to customize the haptic feedback for different applications.
Inside, the engineering is where the real story lives. Powering 6,144 CUDA cores while keeping a thin-and-light form factor requires serious thermal management. Microsoft has equipped the Surface Laptop Ultra with dual fans and dual heat pipes—more than twice the thermal capacity of the 15-inch Surface Laptop 7th Edition from 2024. The fans are thinner and more densely packed with fins than previous designs, with airflow routed from the sides and out the back, passing over heat pipes and surrounding components. At Computex, Microsoft even demonstrated the airflow path using smoke to show how air moves through the system.
The internals are partially upgradeable. Four screws on the underside provide access to the SSD, which uses the standard M.2 2280 form factor and can be swapped out. The battery is also accessible for replacement when it eventually degrades. Microsoft has added QR codes next to internal components that link to service instructions, a design choice borrowed from Framework Computer. However, the memory appears to be soldered directly to the motherboard—there are no SO-DIMM slots visible on the sample shown to press. While Microsoft advertises configurations with up to 128 gigabytes of unified memory, some models will ship with as little as 16 gigabytes, which the company acknowledges is not enough to run large AI conversations entirely offline.
Microsoft is keeping details sparse. The company hasn't allowed press to power on the Surface Laptop Ultra yet, and they're being vague about availability beyond saying it will arrive "this fall"—whether that means back-to-school season or holiday shopping remains unclear. Pricing is even more opaque, though it's safe to assume these won't be budget machines. The Surface Laptop Ultra is being positioned as an "Ultra" device, grouped with other premium AI-first laptops like the Lenovo Yoga Pro, Asus ProArt, and MSI Prestige. These are machines for early adopters willing to pay for cutting-edge AI capability.
The fundamental questions—whether the hardware actually delivers the promised performance, how it handles real-world AI workloads, what the user experience feels like, and whether these machines justify their inevitable premium pricing—remain unanswered. We're still in the early hype cycle. Real evaluation will have to wait until the Surface Laptop Ultra reaches retail and lands in testing labs, where new benchmarks will need to be developed to measure whether this new class of machine delivers what Nvidia and Microsoft are claiming.
Notable Quotes
These machines are being purpose-built for agentic AI tasks run locally on the device.— Microsoft briefing team at Computex
The Hearth Conversation Another angle on the story
So this is a laptop that can run AI models entirely offline? That's a significant shift from what we've been doing.
Yes. Until now, if you wanted to run a serious AI model, you either sent your data to the cloud or you built a custom rig with multiple graphics cards. This is the first consumer laptop designed to do it in your lap.
But 16 gigabytes of memory—that's what some of these will ship with—doesn't sound like enough.
It isn't. Microsoft acknowledges that. The real value is in the configurations with more memory, but those will cost more. It's a tiered approach.
What about the thermal design? Fitting that much compute into something under 18 millimeters thick seems like it would get hot.
That's where the engineering shows. Dual fans, dual heat pipes, more than twice the cooling capacity of the previous generation. They've redesigned the entire airflow path. It's not an afterthought.
And you couldn't actually test it. You just looked at it.
Right. No one could power these on at Computex. We saw the internals, the display, the build quality. But whether it actually runs as promised—that's still a mystery.
When will we know?
Fall, according to Microsoft. But they won't say early or late fall. My guess is they're aiming for the holiday shopping season.