NVIDIA doesn’t usually sell you a laptop. It sells the graphics chip inside someone else’s laptop. That changed yesterday. On October 16, 2026, NVIDIA announced that RTX Spark laptops are actually on sale right now — machines built around NVIDIA’s own “superchip” that pair a Blackwell RTX graphics chip with an NVIDIA Grace CPU and up to 128GB of shared memory. Prices start at $2,599, and the pitch is simple: run serious AI models and agents on your laptop without touching the cloud.
I’ve spent the last year watching every “AI laptop” launch come and go, and most of them are just normal laptops with a sticker on the lid. This one is different hardware with a different promise. Here’s what RTX Spark actually is, what you can buy, and who should care.
What RTX Spark Actually Is
Forget everything you know about a laptop’s insides for a minute. Normally, your CPU and GPU live in separate apartments: the CPU gets the system’s RAM, and the GPU gets its own little slice of VRAM. On most laptops, that GPU slice is tiny — often 6 or 8GB — which is why running a large AI model locally on a normal machine chokes. The model has to fit in memory, full stop.
RTX Spark throws out that arrangement. NVIDIA calls it a superchip, and the idea is one package that holds a Blackwell RTX GPU (the same architecture family as their latest graphics cards) next to a Grace CPU — an Arm-based design NVIDIA built with MediaTek. All of it shares one big pool of up to 128GB of unified memory, wired together with a 600GB/s link between the CPU and GPU.
The headline number NVIDIA gives is up to 1 petaflop of AI performance (measured in FP4, a low-precision format that speeds up AI workloads). I’ll be honest: manufacturer FLOPs numbers are marketing poetry. What matters is the architecture — because the CPU and GPU can read the same memory, the machine doesn’t waste time copying data back and forth. NVIDIA claims these laptops can run “frontier AI models” locally while a creative app runs beside them. Independent reviews are only starting to appear, so treat the performance claims as NVIDIA’s homework until third parties check it.
RTX Spark Specs: The Two Configurations
NVIDIA has listed two RTX Spark laptop configurations, and the difference comes down to how much of everything you get:
- Top configuration: up to 20 Grace CPU cores, 6,144 GPU cores, up to 128GB of unified LPDDR5X memory. Power envelope 45–80W.
- Second configuration: up to 18 Grace CPU cores, 5,120 GPU cores, up to 64GB of unified LPDDR5X memory. Same 45–80W range.
Both run Windows 11, and NVIDIA also announced compact RTX Spark desktops for November — same basic recipe, a 20-core Grace CPU and 6,144-core GPU with up to 128GB of memory in a small chassis with a 140W power budget, designed to sit there 24/7 running background AI agents. There’s also a Windows version of NVIDIA’s DGX Station coming in Q4, but that’s an enterprise deskside machine, not something you’ll see on a shelf at Best Buy.
Which RTX Spark Laptops Can You Buy Right Now
The launch lineup announced October 16 comes from six manufacturers: ASUS, Dell, HP, Lenovo, Microsoft, and MSI. NVIDIA’s retail list includes Amazon, Best Buy, Costco, and B&H Photo Video.
The named models on NVIDIA’s page are the ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yoga Pro 9n, Microsoft Surface Laptop Ultra, and MSI Prestige N16 Flip AI+ and Prestige N16. The Surface Laptop Ultra leads the pack — Microsoft and NVIDIA first showed the RTX Spark platform at a joint October 7 event, where Microsoft demoed the machine running GitHub Copilot and Gears of War: E-Day at 60 frames per second.
Everything starts at $2,599, and there’s a small sweetener: NVIDIA is running a limited-time promotion for new Perplexity users with RTX Spark systems.
The Real Point: Running AI Models Locally
Here’s why anyone’s paying attention. Cloud AI is convenient, but it’s someone else’s computer, billed by the token, holding your data. RTX Spark’s pitch is a laptop that keeps the AI in the house.
The standout stat NVIDIA keeps repeating is memory headroom: with up to 128GB shared between CPU and GPU, you can load models that simply don’t fit on a normal laptop. Reporting around the launch cites a 74-billion-parameter Nemotron model (9 billion active parameters) running in 24GB of memory — that’s the kind of workload that would eat a regular gaming laptop alive. NVIDIA also promises day-one support for local AI agents (they name Hermes and OpenClaw) alongside regular creative software like Adobe Photoshop and Premiere, and Microsoft has made its Execution Containers generally available to help isolate and manage these agents on Windows.
If you’re the kind of person these laptops are aimed at — running local models, wiring up agents, debugging tool calls — you’ll live in JSON payloads. Our free JSON formatter makes tangled model output readable in one click, no signup.
What About Gaming and Creative Work?
NVIDIA isn’t selling a pure “AI laptop” — it wants this to be your everything machine. The claims: edit 8K video, render 3D scenes up to 90GB, and play AAA games at 1440p at over 100 frames per second, with support for DLSS 5 in titles including Gears of War: E-Day and NBA 2K27. The October 7 demo of Gears of War: E-Day at 60fps on the Surface Laptop Ultra suggests gaming is at least taken seriously.
My take: if those numbers hold up in reviews, a machine that’s genuinely good at local AI and plays games at high frame rates is a new category, not just a gimmick. But note the asterisk that should be next to every performance claim in this article — these are NVIDIA’s figures. Nobody has spent a month living with one yet.
Who Should Actually Buy One — My Honest Take
Developers and researchers working with AI: this is the clearest audience. If you’re fine-tuning models, running agents, or building tools on top of local LLMs, a portable 128GB unified-memory machine for $2,599 is genuinely interesting. Cloud GPU bills can eat that in a quarter.
Creators: a maybe. Big-memory unified laptops are great for video and 3D, and Apple‘s MacBooks have proven the unified-memory concept in creative workflows for years. But NVIDIA’s big memory advantage only matters if your software is optimized for it, and early Windows-on-Arm quirks are a real thing to watch.
Everyone else: $2,599 is a lot of money for a laptop whose main trick you’ll never use. If your heaviest workload is a browser with 47 tabs, a $1,000 laptop does the job. And gamers should know that RTX Spark competes with Qualcomm‘s Snapdragon X laptops on the Arm efficiency front, while traditional x86 gaming laptops from the same brands give you proven compatibility at lower prices.
I’d also call it what it is: a first-generation product. First-gen anything has driver quirks, software edges, and a price premium for early adopters. The compact desktops arriving in November might actually be the smarter buy for the always-on agent crowd, since battery life and thermals are where laptops always compromise.
The Bottom Line
NVIDIA RTX Spark laptops went on sale October 16 starting at $2,599, from ASUS, Dell, HP, Lenovo, Microsoft, and MSI. The hardware is genuinely novel — a Grace CPU, Blackwell GPU, and 128GB of unified memory in one package, aimed at running AI models locally instead of renting someone else’s cloud. For developers and researchers, that’s exciting. For everyone else, it’s a fascinating preview of where laptops are going, priced like a first-class ticket.
I’ll be watching the independent reviews closely. If the performance numbers survive contact with reality, NVIDIA just invented a new laptop category. If they don’t, it’s a very expensive demo machine. Either way, the days of assuming your laptop can’t do serious AI are over.

[…] on NVIDIA’s AI pivot: NVIDIA isn’t just shifting chips to data centers — its new RTX Spark laptops run AI models locally without the cloud, starting at […]