Nvidia RTX Spark AI PCs: The Windows Laptops That Run Frontier AI Locally

Nvidia spent Friday showing off a new kind of Windows PC. At its RTX Spark Experience event in New Delhi on November 20, 2026, the company gathered Asus, Dell, HP, Lenovo and MSI to showcase laptops and compact desktops built around its RTX Spark superchip — machines designed to run frontier AI models locally, without sending your work to the cloud. The laptops have actually been on sale globally since October 16, starting at $2,599, but the New Delhi event was the platform’s big moment in one of the world’s fastest-growing PC markets, with local prices and availability finally confirmed.

The pitch is simple: a laptop that is also an AI workstation. Instead of splitting memory between the processor and the graphics card, RTX Spark machines share one big pool of memory between both, so large AI models, video projects and games can all draw on the same resources at once.

What RTX Spark actually is

At the heart of every RTX Spark machine is a single superchip that fuses two very different processors. On one side is a Blackwell RTX GPU with up to 6,144 cores and fifth-generation Tensor Cores; on the other is a 20-core Grace CPU, a custom design Nvidia built with MediaTek. The two are joined by Nvidia’s NVLink-C2C interconnect, and they share up to 128GB of unified LPDDR5X memory. The headline number Nvidia quotes is up to 1 petaflop of FP4 AI performance — enough, the company says, to run large language models and AI agents entirely on the device.

There is a lower tier too, and it matters for the price. It pairs a 5,120-core Blackwell GPU with an 18-core Grace CPU and tops out at 32GB of unified memory, with around 827 teraflops of AI performance. Both tiers run Windows 11, both support CUDA, and both fit into laptops Nvidia describes as the slimmest and most power-efficient RTX laptops ever made — as thin as 14mm and as light as 3 pounds in 14- to 16-inch sizes, with tandem OLED displays on some models.

A close-up of a glowing computer processor chip with green circuits on a dark circuit board

The lineup and the prices

The global laptop lineup, on sale since October 16, includes 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 the MSI Prestige N16 Flip AI+ and Prestige N16. Prices start at $2,599 in the US, with configurations varying by region. Full details are in Nvidia’s availability announcement and on the Nvidia RTX Spark product page.

The New Delhi event added India pricing, covered in Gadgets360’s report on the showcase and Gizbot’s price roundup. Lenovo’s Yoga Pro 9n starts at Rs. 2,99,399, with the Yoga 9n 2-in-1 at Rs. 3,32,801. The ASUS ProArt P14 starts at Rs. 3,19,990 and the ProArt P16 at Rs. 4,89,990, while the compact ProArt GR1X desktop starts at Rs. 9,00,000. Dell’s XPS 16 Creator Edition starts at Rs. 4,49,990 and went on sale through Dell’s online store on November 20, in a 17.8mm chassis with a 3.2K tandem OLED display. HP’s OmniBook Ultra starts at Rs. 5,49,999 with pre-bookings from November 24, and MSI’s EdgeMesa N AI+ compact desktop starts at Rs. 3,89,999. Gigabyte’s AI TOP Aero is expected in December, starting at Rs. 7,35,499.

One warning before you shop: the headline 128GB of memory is the top configuration, and it costs far more than the entry price. Retail listings seen by reviewers put 128GB machines nearer $6,000 to $7,000, so the $2,599 starting price buys a much smaller memory pool. Check the memory figure on the exact configuration before you buy, because that number decides which AI models the machine can actually run.

Local AI agents: from tool to teammate

The feature Nvidia talks about most is not gaming or video editing — it is local AI agents. RTX Spark is built to run agents on the device, so the PC can complete tasks, generate assets and write code without round-tripping to a cloud service. Guided setup and easy installers get users started with agents including GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy, without needing the command line.

Nvidia’s own Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model tuned for tool calling and multi-step tasks, is available for supported RTX Spark configurations. Security gets a mention too: the OpenShell secure runtime and Microsoft Execution Containers are meant to give users control over what agents are allowed to do, and compatible agents can work with tools such as Nvidia Broadcast, RTX Remix and Omniverse libraries. Through Nvidia PAIR, an RTX Spark machine can even serve AI inference to other Windows, Linux and macOS computers on a home network.

That local-first angle is the same instinct behind other recent moves in the industry. Apple is building its own AI agent that can operate a Mac on the user’s behalf, and messaging apps are racing to keep on-device AI from reading private data — a reminder that the most interesting fight in AI right now is over what happens on your own machine, not in a distant data centre.

Built for creators and gamers too

RTX Spark is not only an AI machine. For creators, Nvidia says the systems can edit 8K 4:2:2 video and render 3D scenes of up to 90GB, even on battery power, with Adobe enhancing Photoshop, Premiere and Substance 3D with TensorRT acceleration for AI editing features. Broadcast 2.3 can clean up recorded audio and shape voices with Studio Voice Mic Profiles, and new RTX Video features in ComfyUI can smooth AI-generated video from 15 frames per second to 60.

For gamers, the same Blackwell graphics bring DLSS 5, ray tracing, G-SYNC and Reflex to a portable machine, with Nvidia claiming AAA games at 1440p and over 100 frames per second. Launch support includes Cyberpunk 2077, Fortnite, Gears of War: E-Day, Marvel Rivals, World of Warcraft and NBA 2K27, the last of which uses DLSS 5 with 3D-guided neural rendering.

A compact mini desktop PC beside a laptop showing a colourful 3D scene on a creator's desk

Who should buy one — and who should wait

RTX Spark makes the most sense for a specific buyer: an AI developer who wants to prototype agents and run models locally, or a creator who edits heavy video and 3D work and is tired of renting cloud GPU time. For those users, a machine that runs a 120-billion-parameter model with a million tokens of context on a laptop is genuinely new — that workload used to need a workstation or a cloud bill. Nvidia’s own chip ambitions keep growing, too: the company is nearing a $500M deal to buy Inception’s diffusion AI models, even as it halts H200 AI chip shipments to China amid export controls.

For everyone else, the honest answer is to wait. If your work is documents, browsing and video calls, a 128GB AI laptop is vast overkill, and even the entry configuration is expensive for what it does. The platform is also young: it runs Windows on Arm, which means some software still arrives through emulation or beta builds, and the real-world battery life and app compatibility of these first machines are only now being tested by reviewers. The compact desktops, which only started reaching partners in mid-November, are even earlier in their lifecycle. The pace of chip breakthroughs is relentless — Google’s Willow quantum chip just beat a game classical computers can’t win, and OpenAI’s AI redesigned AMD’s GPU circuits — so today’s flagship can age fast.

The bottom line

RTX Spark is Nvidia’s bet that the next great PC is not defined by its processor speed but by how much AI it can run without the cloud. The hardware is real, it is shipping, and the November 20 showcase in New Delhi showed that every major PC maker is on board. The open question is the same one that follows every new platform: whether the software — the agents, the creative apps, the local models — grows fast enough to justify the price. For AI developers and creators, that answer may already be yes. For the rest of us, the smart move is to watch the reviews, check the memory on the configuration you would actually buy, and let the second wave arrive.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top