Nvidia and Microsoft Just Teased a New Era of PC: What N1X and the GB10 Superchip Actually Mean

Nvidia and Microsoft just posted the exact same thing on the exact same day, and if you have been watching this corner of the industry as long as I have, that is not a coincidence. Both accounts promised “a new era of PC” and dropped the coordinates of the Taipei Music Center, where Jensen Huang gives his GTC Taipei keynote at Computex 2026 next week. When the Windows account mirrors Nvidia’s marketing word for word, the message underneath is hard to miss: the long-rumored N1X laptop platform is probably about to show up, and it looks like it is running Windows on Arm.

I have been digging into the chip this is all built around since the DGX Spark reviews landed, so I want to explain what is actually coming, where the hype outruns the silicon, and what I would buy today if you want to build for this platform before everyone else does.

What N1X actually is

N1X is widely understood to be the mobile version of the GB10 Superchip, the same part sitting at the heart of Nvidia’s DGX Spark mini-PC. GB10 pairs an RTX 5070-class Blackwell GPU with a 20-core Arm CPU complex that Mediatek helped design (ten Cortex-X925 performance cores and ten Cortex-A725 efficiency cores), and it hangs 128GB of LPDDR5X off a single unified memory pool that both the CPU and GPU share.

That unified-memory design is the whole point. On a normal gaming laptop your discrete GPU gets its own walled-off pool of fast GDDR memory, and anything bigger than that pool has to be shuffled across the PCIe bus. On GB10 the GPU can address all 128GB directly. For AI work that is the difference between loading a 70-billion-parameter model and watching it crash with an out-of-memory error. You can run quantized large language models locally, on your desk, that simply will not fit on a 16GB or 24GB consumer card.

Why Microsoft showing up changes the story

Until now GB10 has only existed as the DGX Spark, and the DGX Spark runs DGX OS, which is Ubuntu Linux. It is an AI developer sandbox, not a machine your accountant can use. The moment Microsoft puts its weight behind N1X, the entire Windows application ecosystem comes along through Windows on Arm, and this stops being a niche Linux box and starts being a general-purpose computer that also happens to have a serious local-AI engine inside it.

This matters for Microsoft more than people realize. None of its existing Windows on Arm partners have shipped anything close to the GB10’s raw AI capability. Today’s Copilot+ PCs run small neural accelerators rated in the tens of TOPS; GB10 is a different weight class entirely, advertised around a petaFLOP of AI performance. If Microsoft has hardware this capable to target, it can finally build first-party local AI features that were never possible on the underpowered NPUs it has been shipping. It is worth remembering that Microsoft has already confirmed Windows 11 26H1 will launch as an Arm-only release, so the company is clearly committing engineering runway to this architecture.

Where I would pump the brakes

I am not going to pretend the first wave of these machines will be for everyone. Two specifics keep me grounded.

First, bandwidth. Because the CPU and GPU share one LPDDR5X pool, the GB10 GPU gets about 273 GB/s of memory bandwidth. That is generous for a unified design, but it is far below what a traditional laptop GPU pulls from dedicated GDDR. In practice you can game on GB10, but gaming is not what it is good at. The bottleneck shows up exactly where you would expect it to.

Second, price. Every DGX Spark-class GB10 box I have seen is selling in the rough neighborhood of $4,000 to $5,000. Some of that premium comes from an exotic ConnectX networking card that almost certainly will not survive the trip into a laptop chassis, so N1X notebooks could shave cost there. But with memory and SSD prices where they are during the current silicon crunch, nobody should expect these to be cheap. A smarter product stack with 64GB options and trimmed core counts would help, and I suspect that is coming, but the launch hardware will be a premium purchase aimed at developers and AI professionals, not students.

Who this is actually for

If you fit one of these descriptions, the GB10 platform is genuinely interesting right now rather than a year from now:

• You build with local LLMs. 128GB of unified memory lets you run and fine-tune models that will not fit on any single consumer GPU, with no cloud bill and no data leaving your desk.
• You are an AI/ML engineer who wants the dev environment that ships first. The DGX Spark’s CUDA stack is the same architecture N1X laptops will inherit, so anything you build now ports forward.
• You want a quiet, compact AI workstation. This is a desktop-class AI engine in a footprint smaller than a stack of books, not a screaming tower.
• You are a researcher or founder prototyping agentic systems. A petaFLOP of on-desk compute means you iterate locally instead of renting H100 time by the hour.
• You teach or demo AI. A self-contained box that runs big models offline is far easier to deploy in a classroom or a conference room than cloud credentials.

What I would buy today

Here is the honest part. N1X laptops are not on sale yet, and based on Nvidia’s own history the first units will be supply-constrained and expensive when they do arrive. If you want to start building for this architecture now, the shipping hardware is the DGX Spark and its OEM twins, which use the identical GB10 Superchip. Anything you develop on them moves straight onto N1X when the laptops land.

The reference machine is the Nvidia DGX Spark with the GB10 Grace Blackwell Superchip, 128GB of LPDDR5X, and 4TB of NVMe storage. It is the cleanest way to get the full unified-memory experience and Nvidia’s own DGX software stack.

If you would rather buy the same silicon from an OEM, the ASUS Ascent GX10 packs the identical GB10 Superchip, 128GB of unified memory, and a stackable chassis so you can cluster two of them over Nvidia’s high-speed interconnect later. Same chip, same 128GB pool, slightly different box.

Both are premium purchases, and I would only pull the trigger if local AI is real work for you rather than a curiosity. But if it is, owning the GB10 platform now means you are already fluent on the architecture the rest of the Windows world is about to discover at Computex.

The bottom line

A coordinated Nvidia and Microsoft teaser is not proof of anything, and I will update this once Jensen actually holds the thing up on stage. But the direction is clear enough to plan around: the GB10 architecture is leaving the Linux sandbox and walking into the Windows mainstream. The DGX Spark is how you get a head start, and N1X is what happens when that same engine fits in a backpack.

Disclosure: Links to Amazon are affiliate links. If you buy through them I earn a small commission at no extra cost to you. I do not accept payment from manufacturers for placement. Specifications cited here are based on Nvidia’s published GB10 platform details and independent reporting; prices and availability change, so confirm current details on the product page before buying.