Surface Laptop Ultra: 1 Petaflop AI, 128GB RAM — And It Runs CUDA Natively

Microsoft just dropped the thing many of us have been waiting for since the original Surface.

At Computex 2026, Microsoft and NVIDIA jointly announced the Surface Laptop Ultra — the most powerful Windows on Arm laptop ever made, powered by NVIDIA's RTX Spark "superchip" with up to 128GB of unified memory and a staggering 1 petaflop of AI compute. That's enough to run 120-billion-parameter models entirely on-device, no cloud required.

What's happening?

This isn't Microsoft's first attempt at putting an NVIDIA chip in a Surface. Back in 2012, the Surface RT shipped with an NVIDIA Tegra — and failed so spectacularly Microsoft had to write off $900 million. But this time is different.

The Surface Laptop Ultra runs Windows 11 on Arm with full native CUDA support — something AI developers have been waiting for ever since Apple Silicon proved Arm could match x86 in performance. Microsoft claims it's "the most powerful thing we've ever made." Andrew Hill, head of Surface, doesn't mince words: "This is the most powerful thing we've ever made."

Hardware Breakdown — Not Your Office Worker's Surface

The NVIDIA RTX Spark Superchip

At the heart of the Surface Laptop Ultra is the RTX Spark Superchip, a combined CPU+GPU architecture connected via NVLink-C2C:

Component	Specification
CPU	20-core NVIDIA Grace (Arm, co-developed with MediaTek)
GPU	NVIDIA Blackwell RTX, up to 6,144 CUDA cores
Tensor Cores	5th-gen, FP4 precision
AI Performance	1 Petaflop
Unified Memory	Up to 128GB, dynamically allocated between CPU and GPU

The RTX Spark is essentially a variant of the chip already found in NVIDIA's DGX Spark AI mini-PC — but this time optimized for Windows 11 and squeezed into a laptop chassis weighing under 4.5 pounds (~2kg).

Display and Design

• Display: 15-inch mini-LED PixelSense Ultra, 2880×1920 resolution (262 ppi)
• Brightness: 2,000 nits peak HDR — the brightest Surface display ever
• Touchpad: Largest haptic trackpad on any Surface device
• Weight: Under 4.5 lbs (~2kg)
• Ports: HDMI, USB-C, USB-A, full-size SD card slot, headphone jack — no dongle required
• Colors: Platinum and Nightfall

Microsoft is also emphasizing repairability: user-replaceable SSD, official repair guides, and replacement parts available for purchase.

Why AI Developers Should Care

1. Run 120B-Parameter Models Locally

128GB of unified memory means you can load models like Llama 3 70B — and potentially 120B-parameter models — entirely on-device. No cloud GPU instances, no inference API costs, no data leaving your machine.

This is a game-changer for anyone working with sensitive data, wanting to fine-tune private models, or simply tired of being at the mercy of third-party API providers.

2. Native CUDA on Windows on Arm

For the first time, CUDA runs natively on a Windows Arm machine. No translation layers, no WSL gymnastics. The entire AI ecosystem — PyTorch, TensorFlow, JAX — runs natively on the same device you use for Zoom calls and writing code.

In short: you can prototype AI models on the exact same machine you use for your day job.

3. Dual-Fan Cooling for Sustained Workloads

AI inference and training aren't burst workloads — they demand sustained performance. The dual-fan cooling system is specifically designed to prevent aggressive thermal throttling during long-running rendering or model workloads. Microsoft's tagline: "No walls. No compromises."

Windows on Arm: Finally Ready?

Microsoft has spent years getting Windows ready for Arm. With the Surface Laptop Ultra, they're not just showing off hardware — they're signaling that Windows on Arm is ready for professional workloads.

This matters enormously. Apple Silicon has spent the last five years proving that Arm can beat x86 in performance-per-watt. The Surface Laptop Ultra is a direct answer to the MacBook Pro — but with one weapon Apple doesn't have: CUDA.

Apple's M-series GPUs are powerful, but they don't run CUDA. And in the AI world, CUDA is the standard. Every framework, every model, every optimization targets CUDA first. A laptop with native CUDA and 128GB of unified memory is something no MacBook can replicate right now.

But What About the Price?

That's the big question. Microsoft hasn't announced pricing, but with this hardware — 128GB unified memory, mini-LED display, custom NVIDIA silicon — don't expect it to be cheap. NVIDIA's DGX Spark (the desktop version of the same chip) cost several thousand dollars at launch.

However, NVIDIA has indicated the RTX Spark family will span multiple configurations, starting from 16GB, to cover a range of price points. The Surface Laptop Ultra is expected to ship in fall 2026.

The Marketing Is... A Lot

Let's be honest: Microsoft's blog post about the Surface Laptop Ultra is filled with phrases like "Every micron matters and every choice is deliberate" and "It belongs in the hands of world makers." It sounds like Apple circa 2015. But beneath the marketing fluff are real specs, and specs don't lie.

On paper, this hardware represents the biggest leap the Surface line has ever taken.

The Bottom Line

The Surface Laptop Ultra signals three important shifts:

• Windows on Arm has grown up — it's no longer a "good enough" platform but a genuine option for power users
• Local AI is becoming the standard — you no longer need the cloud to run large models
• Microsoft + NVIDIA are taking direct aim at Apple — and this time they've brought CUDA

The only remaining question: what will it cost? And are developers willing to pay that much for a laptop that runs 120B-parameter models right on their desk?

---

Content assisted by AI (Amy 🌸). Reviewed by the author.