Nvidia's Huang: Surface Laptop Ultra Took Four Years to Build

What Huang actually said about the Surface Laptop Ultra

Speaking at a developer event in London, Huang framed the Surface Laptop Ultra as the result of a long, deliberate bet on local AI rather than a routine hardware refresh.

"This computer is going to be incredible — it took us four years to build it," Huang told attendees, according to TechRadar Pro. He described the project as "really a work of love," with several thousand engineering years behind it and a brand-new chip line built from scratch.

The pitch is not about benchmarks. Huang's story starts with a question: what if your PC could run an AI agent locally, one that reaches your files, reasons about them and uses your apps as tools? "The ultimate tool is your personal computer," he said.

His imagined workflow is almost casual. You text your laptop from somewhere else, ask it to render images or update slides, and it hands back the results. "You don't have to bring it with you anymore — just leave it at your desk," he said.

The RTX Spark platform: Nvidia's first shared-memory GPU

The Surface Laptop Ultra runs on Nvidia's RTX Spark platform, pairing an N1X processor with RTX Blackwell graphics and up to 128GB of unified memory.

Huang said the chip delivers 1 petaflop of computing and is the first Nvidia GPU with shared memory, a redesign made because "we knew the models were going to be big." Shared memory lets the CPU and GPU draw from the same pool instead of copying data back and forth, which matters when a model needs to sit in memory alongside everything else.

The configuration list starts at 24GB RAM and 512GB storage, with a step-up version at 32GB RAM and 1TB storage, and support for up to 128GB of unified memory. Nvidia is aiming the machine at developers, AI specialists, engineers, scientists and creators running demanding workloads.

Huang also drew a line between three kinds of PCs: the ordinary machine you type on, the GeForce gaming PC, and the Nvidia workstation. The Surface Laptop Ultra is meant to sit above all three, running every AI framework and any AI model. That is a big claim, and independent testing will decide how much of it holds up.

Why four years is the real story

A four-year development cycle for a laptop chip is unusual, and it says as much about Nvidia's priorities as it does about Microsoft's.

Huang said Nvidia had to wait until Blackwell was finished before the laptop chip could be completed. That sequencing explains the timeline: the Surface Laptop Ultra is not a side project, it is a downstream product of Nvidia's data-center roadmap.

It also explains the timing of the pitch. Agentic AI — software that takes actions on your behalf — is the industry's current obsession, and every major chipmaker wants a piece of the local compute that makes it practical. Nvidia's argument is that the PC is where those agents should live, close to your files and your tools.

The catch is that a flagship like this rarely sells in volume. It is a showcase: a machine that proves Windows on Arm can carry serious AI hardware, and a reference point for whatever Nvidia and Microsoft do next. For a sense of how Nvidia's silicon is being pulled toward AI workloads, the RTX 5090 vanishing from Nvidia's own store is a useful parallel.

Availability and pricing

The Surface Laptop Ultra launched on October 16, 2026. Microsoft has not published official US, UK or EU pricing in the material available, so treat any figure you see elsewhere as unconfirmed until Microsoft lists it.

Availability will likely follow Microsoft's usual Surface rollout, starting with major markets before wider distribution. Buyers in the US, UK and EU should check Microsoft's own store pages for configuration and price rather than relying on early retailer listings.

Nvidia's first Surface chip is a statement of intent more than a mass-market product. Whether developers buy the agent-on-your-desk vision depends on real-world performance, battery life and price — none of which Huang addressed. The four-year wait is over; now the benchmarks have to do the talking.