Surface Laptop Ultra: Microsoft's $2,599 Nvidia AI PC

Surface Laptop Ultra is Microsoft's $2,599 laptop built with Nvidia's RTX Spark chip, packing up to 128GB of unified memory so it can run giant AI models entirely on-device. Microsoft opened pre-orders on October 7, with shipments starting October 16.

At an event in San Francisco on Wednesday, Microsoft CEO Satya Nadella took the stage with Nvidia CEO Jensen Huang to unveil what both companies describe as a new kind of personal computer: a Windows laptop that behaves more like a portable AI workstation. Huang put it boldly, saying the team had "reinvented the computer as we know it."

What is the Surface Laptop Ultra?

Image: Microsoft

The Surface Laptop Ultra is Microsoft's most powerful Surface laptop ever, and the first one built around an Nvidia chip. It's a 15-inch machine aimed squarely at developers, creators, and people who build with AI — not at shoppers looking for a faster everyday laptop.

The pitch is simple: instead of sending every AI task to the cloud (and paying per use), you run the models right on the machine, for free, with your data staying on your device. The laptop comes preloaded with tools like VS Code, GitHub Copilot, WSL, PowerShell 7, Python, and Node, so developers can start building the moment they unbox it.

Nvidia RTX Spark: the superchip inside

Image: Microsoft

At the heart of the machine is Nvidia's RTX Spark, an Arm-based "superchip" that fuses a 20-core Grace CPU with a Blackwell RTX GPU (up to 6,144 CUDA cores) into a single package. Both halves share one pool of up to 128GB of unified LPDDR5X memory, so the CPU and GPU can grab RAM wherever the workload needs it most.

Why does unified memory matter? AI models are hungry for memory. Microsoft says the top configuration can run models with more than 120 billion parameters locally — that includes the company's own MAI Code 1.1 Flash coding model, which was compressed to 53GB specifically so it can live on the device. The headline "1 petaflop of AI performance" figure is theoretical FP4 performance with sparsity, so treat it as a ceiling rather than an everyday number.

It's also worth noting this is Windows on Arm, with x86 apps running through Microsoft's Prism emulation. Historically that's been a weak spot for Arm Windows machines, but Nvidia argues the chip is powerful enough that even emulated apps clear the bar — and the gaming push (60 FPS in Gears of War: E-Day, with Call of Duty support coming) should help the whole Arm ecosystem.

Surface Laptop Ultra price, specs, and release date

Image: Microsoft

Microsoft is selling three main configurations:

• $2,599 — 18-core CPU, 5,120-core GPU, 24GB unified memory, 512GB SSD
• $3,699 — 20-core CPU, 6,144-core GPU, 32GB unified memory, 1TB SSD
• $5,899 — top configuration with 128GB unified memory (already out of stock, per Microsoft)

Display and design

The 15-inch PixelSense Ultra touchscreen is a mini-LED panel with up to 2,000 nits of peak HDR brightness, which Microsoft claims is 25% brighter than the MacBook Pro M5 — though touch-only, with no pen support. The haptic touchpad is 30% larger than the previous Surface Laptop, and the cooling system has 2.5 times the thermal capacity to handle sustained AI workloads. The chassis stays under 18mm thick and weighs about 2kg.

Ports Apple won't give you

In a direct swipe at MacBook buyers, the laptop includes three USB-C ports, a USB-A port, HDMI, a full-size SD card reader, and a headphone jack. A new Magnetic Connect cable snaps onto the USB-C port for charging — Microsoft calls it the world's first built-in magnetic USB-C charging on a laptop. The SSD is user-replaceable.

The developer desktop

Alongside the laptop, Microsoft priced the Surface RTX Spark Dev Box at $5,999. It's a compact desktop with the same RTX Spark platform and 128GB of memory, meant to compete with the Mac Studio and Mac mini. It's sold only through Microsoft's US store, with shipping starting in November.

Windows 11 learns to babysit AI agents

The hardware is only half the story. Microsoft also announced a revamped Windows 11 with a feature called Execution Containers — a way to sandbox AI agents so they can't wander where they shouldn't. Nadella said the feature will roll out to all Windows 11 users, not just buyers of the new hardware.

The broader idea is what Microsoft calls "hybrid intelligence": Windows decides whether an AI task should run locally or move to the cloud. GitHub Copilot will start splitting coding work between local and cloud models later this month, and Meta's Muse AI agent is coming to Windows as a native app.

Nadella summed up the philosophy on stage: having a model alone "doesn't do much for anything" — you need memory outside the model, an orchestration layer, and a platform that any developer's agents can build on. "That's what the Windows platform is all about going forward."

Practical takeaways

• For developers: a laptop that runs 120B-parameter models locally is genuinely new. If you experiment with AI models daily, local inference with no per-token cloud bill is the big win.
• For creators: the display, CUDA acceleration, and ports make it a strong portable workstation — but check that your tools are Arm-ready or run well under emulation before pre-ordering.
• For MacBook owners: Microsoft is offering up to $1,000 cash back on eligible MacBook Pro trade-ins through November 23 (US and Canada) — an unusually aggressive lure.
• For everyone else: the $2,599 entry price puts this firmly in "pro workstation" territory, not "college laptop" territory. RTX Spark laptops from Asus, Dell, HP, Lenovo, and MSI also ship October 16, so cheaper options are coming.

Between Microsoft's Surface Laptop Ultra and the coming wave of RTX Spark machines from other PC makers, the "agentic PC" era is officially underway. The next question is whether everyday software — and everyday users — are ready to meet it.

If you want to keep up with where AI hardware is heading, check out our more AI news and more Technology news sections.

Sources: TechCrunch, Microsoft Devices Blog, TechMyMoney, iTechPost