At Computex 2026, Microsoft introduced the Surface Laptop Ultra — a machine built around Nvidia's new RTX Spark processor — as a considered argument about where serious computing is heading. Aimed at those who run AI models, edit video, and demand desktop-class power in a portable form, the laptop pairs a 20-core CPU with RTX 5070-equivalent graphics and up to 128GB of unified memory. The announcement is as much a philosophical statement as a product launch: that the future of high-end work belongs on the edge, untethered from the cloud. Yet the most consequential details — price, minimum memo
Microsoft's Surface Laptop Ultra: A Powerhouse with Questions
A machine with impressive specs and thoughtful engineering, but it remains a promise.
Mark
So this is the machine Microsoft is leading with at Computex. What actually makes it different from what came before?
Mimi
The processor is the real shift. The RTX Spark is designed specifically for AI workloads running locally—128 gigabytes of unified memory means you can run large language models on the machine itself without sending data to the cloud. That's a different class of computing than a typical laptop.
Luke
But we haven't seen benchmarks yet, right? The hands-on was in a demo environment.
Mimi
Correct. The testing was controlled, no real-world performance data. But the thermal design is solid—dual fans, dual heat pipes, and the machine stayed quiet even under load during the show.
Mark
What about the price? Is this a $2,000 machine or a $4,000 machine?
Mimi
That depends entirely on the RAM configuration, which Microsoft hasn't announced. The minimum is probably 32 gigabytes, more likely 64, which puts the base model at $2,500 or higher.
Luke
So we're estimating based on what makes sense, not what Microsoft has actually said.
Mimi
Right. They've shown the hardware, but the actual configurations and pricing are still unknown.
Mark
The display sounds exceptional. Two thousand nits is genuinely bright.
Mimi
It is. The mini-LED panel with that brightness creates visuals that really pop. But that brightness also draws power, which raises questions about battery life that haven't been answered yet.
Luke
And availability is vague too—late summer or early fall for pre-orders, shipping sometime after that.
Mimi
Yes. It's a machine with impressive specs and thoughtful engineering, but it's still a promise at this point.
Le Pouls
- Microsoft arrived at Computex 2026 with a single machine powered on — a deliberate act of control that signals both confidence and caution around the Surface Laptop Ultra.
- The RTX Spark chip at its core promises up to one petaflop of AI performance, positioning this laptop as a direct challenge to cloud dependency for creative and machine-learning workloads.
- A redesigned thermal system — raised chassis, dual fans, dual heat pipes — had to work hard to contain the heat generated by this much processing power, and early demos suggest it mostly does.
- Repairability details like a removable backplate, accessible SSD and battery, and QR-coded internal components hint at a longer-term product philosophy unusual for premium consumer hardware.
- Pricing likely starting above $2,500 and climbing past $4,000 for top configurations, combined with vague late-summer availability, means the machine remains a well-staged promise rather than a purchasable reality.
At Computex 2026, Microsoft introduced the Surface Laptop Ultra — a machine built around Nvidia's new RTX Spark processor — as a considered argument about where serious computing is heading. Aimed at those who run AI models, edit video, and demand desktop-class power in a portable form, the laptop pairs a 20-core CPU with RTX 5070-equivalent graphics and up to 128GB of unified memory. The announcement is as much a philosophical statement as a product launch: that the future of high-end work belongs on the edge, untethered from the cloud. Yet the most consequential details — price, minimum memory, battery life, and availability — remain unresolved, leaving the machine, for now, more vision than certainty.
Microsoft used Computex 2026 to unveil the Surface Laptop Ultra, a laptop centered on Nvidia's RTX Spark — an ARM-based processor designed to handle up to one petaflop of AI performance. The specifications are serious: a 20-core CPU, a GPU equivalent to the GeForce RTX 5070, and up to 128GB of unified memory. The target audience is equally specific — video editors, machine learning engineers, and anyone running large language models locally without relying on a cloud connection.
The physical design is refined without being showy. A 15-inch mini-LED PixelSense Ultra touchscreen delivers 262 pixels per inch and peaks at 2,000 nits of brightness — vivid enough to genuinely impress during hands-on time at the show. The aluminum chassis, edge-to-edge glass, and a full port selection including USB-C, USB-A, HDMI, SD card reader, and headphone jack mean users won't need a bag full of adapters.
Thermal management received particular attention, and for good reason. Microsoft raised the chassis slightly to allow airflow underneath, and dual fans with dual heat pipes draw cool air in through the sides and exhaust it out the back. During extended demo sessions running graphically demanding games, the machines stayed warm but the fans remained quieter than expected. More notably, Microsoft built in a removable backplate giving access to the SSD and battery, with QR codes on internal components for individual replacement — a meaningful gesture toward repairability.
Still, Computex is a controlled environment, and the Surface Laptop Ultra was the only RTX Spark machine allowed to power on during the show. The open questions are significant: minimum RAM configurations, real-world battery life under a 2,000-nit display, and pricing that will likely start around $2,500 and potentially exceed $4,000 fully loaded. Pre-orders are expected in late summer or early fall, with no confirmed ship date. The engineering is thoughtful and the ambition is clear — but the months ahead will determine whether the Surface Laptop Ultra becomes the machine its Computex debut suggested it could be.
Microsoft walked into Computex 2026 with a statement: the Surface Laptop Ultra, a laptop built around Nvidia's new RTX Spark processor, an ARM-based chip that represents a deliberate bet on what high-end computing should look like in 2026. The hardware inside is substantial. A 20-core CPU pairs with a GPU equivalent to Nvidia's GeForce RTX 5070, backed by up to 128 gigabytes of unified memory—the kind of specification sheet that speaks directly to video editors, machine learning engineers, and anyone running large AI models locally on their machine. The RTX Spark itself is positioned as a new category of processor, designed to handle up to one petaflop of AI performance, a figure that translates to real work: video upscaling, intelligent masking, billions of parameters of language models running without a cloud connection.
The physical machine itself is polished. The 15-inch display is a mini-LED PixelSense Ultra touchscreen with 262 pixels per inch, a 3:2 aspect ratio, and peak brightness of 2,000 nits—bright enough that during hands-on testing at the show, the visuals genuinely stood out. The chassis borrows from the playbook that works: aluminum body, edge-to-edge glass, recessed black keys. The touchpad is responsive and appropriately sized. The port selection is comprehensive for creators—two USB-C ports, one USB-A, HDMI, an SD card reader, and a headphone jack. It's the kind of machine that doesn't force you to buy adapters.
The thermal engineering deserves its own paragraph because it had to. This much processing power generates heat, and Microsoft redesigned the cooling system to manage it. The laptop sits slightly raised off the desk to allow air to flow underneath. Inside, dual fans and dual heat pipes pull cool air in through the sides and exhaust it out the back. During the demo at Computex, multiple Surface Ultra units ran graphically demanding games—Pragmata and Indiana Jones and the Great Circle—for hours. The machines were warm to the touch, but the fans running at maximum speed remained surprisingly quiet. Microsoft even planned a smoke machine demonstration to visualize the airflow, though the machine malfunctioned when powered on. Beyond cooling, Microsoft included a detail that suggests longer-term thinking: the backplate is removable, giving users access to the SSD and battery, and internal components are tagged with QR codes for individual replacement. It's a nod to repairability in an industry not always known for it.
But Computex is a stage, and stages are built for spectacle. The Surface Laptop Ultra was the only RTX Spark laptop allowed to power on during the show—all other new models remained dark. That control of the narrative matters when trying to understand what actually remains unknown. The biggest open question is configuration and pricing. Yes, the RTX Spark supports up to 128 gigabytes of unified memory, but what will the minimum RAM configuration actually be? A 16-gigabyte model would undercut the entire purpose of the processor and wouldn't support the AI workflows this machine is designed for. The realistic floor is probably 32 gigabytes, though 64 gigabytes seems more likely, which would push the base price into the $2,500 range or higher. Fully loaded configurations could exceed $4,000.
Battery life is another question mark. Everyone Microsoft spoke to expressed confidence in the Surface Ultra's endurance, but a 2,000-nit mini-LED display demands power regardless of how efficient the processor is. Real-world testing hasn't happened yet—the hands-on at Computex took place in a controlled demo environment, no benchmarking, no sustained use. Availability is similarly vague. Pre-orders are expected in late summer or early fall, with shipping sometime after that, but nothing is confirmed. The Surface Laptop Ultra is a machine with impressive specifications and thoughtful engineering, but it remains, for now, a promise. The months ahead will determine whether it delivers.
Citations marquantes
The RTX Spark is a new class of GPU for AI, designed explicitly to run large models and access datasets locally.— Microsoft
Envie de l'histoire complète ?
Lire l'original sur ZDNet ↗
Technology