Run 120-billion-parameter AI models directly on the device without needing the cloud
At Computex 2026, Microsoft unveiled the Surface Laptop Ultra — a flagship device built around Nvidia's RTX Spark chip — staking its claim at the intersection of portable computing and artificial intelligence. The machine is designed to run vast AI models locally, without the cloud, placing serious computational power directly in the hands of developers and creators. It is a bold wager from a company that once lost nearly a billion dollars on a similar Nvidia partnership, now returning to that alliance in an era where AI has moved from speculation to daily practice.
- Microsoft is betting its flagship Surface line on a chip capable of 1 petaflop of AI compute — enough to run 120-billion-parameter models without ever touching a cloud server.
- The RTX Spark's 20 CPU cores, 6,144 GPU cores, and up to 128GB of unified memory signal a fundamental shift in what a thin laptop is expected to carry.
- A ghost haunts the launch: Microsoft's 2012 Surface RT with Nvidia ended in a $900 million write-down, and the industry has not forgotten.
- The device's brightest-ever Surface display, haptic touchpad, and repairability focus suggest Microsoft is courting creators and professionals — not just spec-hunters.
- With no price announced and a vague 2026 launch window, the critical question of whether this is a specialist tool or a mainstream statement remains deliberately unanswered.
Microsoft announced the Surface Laptop Ultra at Computex 2026, positioning it as the most powerful Surface device ever made. Built around Nvidia's RTX Spark chip, the laptop is aimed at AI developers, creative professionals, and power users who need serious compute without being tethered to the cloud.
The RTX Spark delivers up to 20 CPU cores and 6,144 GPU cores, producing up to 1 petaflop of AI compute — sufficient to run 120-billion-parameter AI models entirely on-device. Up to 128GB of unified memory allows the CPU and GPU to share resources fluidly, smoothing workflows across 3D rendering, software compilation, and local AI inference. Full CUDA support and a dual-fan cooling system round out the technical foundation.
The 15-inch mini-LED touchscreen reaches 2,000 nits of peak brightness — Microsoft's most luminous Surface display to date — at 262 pixels per inch. The machine weighs just under 4.5 pounds, features the largest touchpad ever on a Surface with haptic feedback, and offers a full complement of ports including USB-C, USB-A, HDMI, and a full-size SD card slot. Microsoft has emphasized both durability and repairability in the design.
The device runs Windows 11 adapted for the RTX Spark architecture, with Nvidia's OpenShell platform enabling local agentic AI with built-in security measures. Battery life is promised to last a full day, though capacity and charging details remain undisclosed.
The shadow of history looms over the launch. Microsoft's 2012 Surface RT — also built on an Nvidia ARM chip — resulted in nearly $900 million in losses. The landscape has changed: AI is now a working tool, not a promise. But with no price announced and a general 2026 release window, the Surface Laptop Ultra's true ambition — niche instrument or mainstream challenger — will only become clear when that number finally arrives.
Microsoft has entered the ring with a new flagship laptop, the Surface Laptop Ultra, built around Nvidia's RTX Spark chip—a processor designed to reshape what a thin, portable computer can do. The announcement came during Computex 2026, where Nvidia first revealed the RTX Spark as part of a broader push to reinvent the personal computer around artificial intelligence. For Microsoft, this marks a significant bet: the company is positioning the Surface Laptop Ultra as its most powerful Surface device ever, aimed squarely at AI developers, creative professionals, and power users who need serious compute in a portable form.
The hardware inside tells the story of where computing is heading. The RTX Spark chip delivers up to 20 CPU cores and 6,144 GPU cores, capable of producing up to 1 petaflop of AI compute—enough to run 120-billion-parameter AI models directly on the device without needing a cloud connection. The machine comes with up to 128GB of unified memory, a design choice that allows the CPU and GPU to share RAM dynamically depending on what task is running. This architecture is meant to smooth the workflow for 3D work, long software compilation cycles, local AI inference, and heavy multitasking. The laptop also includes full CUDA support, the standard toolkit for GPU-accelerated computing, and a dual-fan cooling system to manage the thermal load.
The display is built for creators. A 15-inch mini-LED touchscreen with 262 pixels per inch reaches 2,000 nits of peak brightness in HDR mode—Microsoft's brightest Surface display to date. The machine weighs just under 4.5 pounds and comes in Platinum and Nightfall finishes. The touchpad is the largest ever fitted to a Surface device, and it uses haptic feedback. Ports include USB-C, USB-A, HDMI, a full-size SD card slot, and a headphone jack, though Microsoft has not yet specified which versions or speeds these connections will support. The company has also emphasized durability and repairability in the design.
Battery life is promised to be all-day, though Microsoft has not disclosed the battery capacity or charging speeds. The device runs Windows 11, adapted specifically for the RTX Spark architecture. One of Nvidia's key focuses with the RTX Spark is agentic AI—the ability to run AI agents locally on the device using Nvidia's OpenShell platform, with new security measures built in.
This is not Microsoft's first dance with Nvidia on an ARM-based Surface device. In 2012, the company launched the Surface RT with an Nvidia ARM processor. That venture cost Microsoft roughly $900 million in losses. The stakes are different now—AI is no longer a speculative technology but a practical tool that developers and creators are actively building with. Still, the success of the Surface Laptop Ultra will depend heavily on pricing and market adoption. Microsoft has said the device will launch later in 2026 but has not announced a price. That number, when it comes, will signal whether this is a niche device for specialists or a genuine attempt to reshape the mainstream laptop market.
Notable Quotes
The Surface Laptop Ultra is the most powerful Surface Laptop ever built.— Microsoft
The RTX Spark is designed for slim laptops with a thickness of as much as 14mm.— Nvidia
The Hearth Conversation Another angle on the story
Why does Microsoft need to build a laptop around Nvidia's chip instead of using their own processors?
Microsoft doesn't make their own chips for consumer laptops the way Apple does. Nvidia's RTX Spark is purpose-built for AI workloads, and rather than spend years developing that capability in-house, Microsoft is partnering with Nvidia to get to market faster. It's a pragmatic choice.
The unified memory thing—why does that matter so much?
Normally, your CPU and GPU have separate memory pools, and moving data between them creates a bottleneck. Unified memory lets them share the same pool, so when you're doing something like 3D rendering or running an AI model, the data doesn't have to make that expensive trip back and forth. It's faster and more efficient.
Can you actually run serious AI models on this thing without the cloud?
Yes, that's the whole point. You can run 120-billion-parameter models locally. That means an AI developer or a creative professional can iterate on their work without uploading data to a server, waiting for results, and downloading them back. It's a different experience entirely.
What about that $900 million loss on the Surface RT? Isn't Microsoft worried about history repeating?
They should be thinking about it. But the context is completely different now. In 2012, ARM laptops were a curiosity. Today, AI is something people actually need to work with. If the pricing is reasonable and the performance is real, there's a genuine market here.
So what's the catch?
Price. We don't know what this costs yet, and that will determine whether it's a specialty tool for professionals or something that actually changes the market. A $3,000 laptop for developers is one thing. A $5,000 or $6,000 laptop is another.