Nvidia RTX Spark: the ARM chip that wants to kill Intel, AMD, and Apple in one shot

June 1, 2026, Taipei Computex. Jensen Huang takes the stage, usual leather jacket, and pulls out of his pocket something we weren't necessarily expecting from Nvidia: a complete processor for laptops . Not a graphics card, not a server module. A chip that runs Windows on its own, like an M4 at Apple or a Snapdragon X2 at Qualcomm.

The RTX Spark - codenamed N1X - is the first time Nvidia has set foot in the consumer PC market as a complete SoC provider. This refers to a chip that integrates both the CPU (a 20-core ARM Grace, co-developed with MediaTek) and the GPU (a Blackwell RTX with 6,144 CUDA cores), all connected by an internal NVLink-C2C. 128 GB of unified memory . Engraving in 3 nm at TSMC.

Said differently: it's the same bet that Apple made in 2020 with the M1, but Windows version. And with the GPU power of a desktop RTX 5070 inside.

We will present the figures, because they speak for themselves.

6,144 CUDA cores, that's exactly the same number as a desktop GeForce RTX 5070. Except here, it's in a 14mm thick laptop weighing 1.3kg. The 80W TDP is roughly half of what a typical CPU+GPU combo consumes in a gaming laptop. And the 128GB unified memory is what allows loading monstrous AI models directly locally — we'll come back to that just after.

The parallel with Apple Silicon is obvious. Shared memory between CPU and GPU, ARM architecture, fine engraving, low thermal envelope. The difference is that Nvidia brings with it the entire CUDA, RTX, DLSS, TensorRT ecosystem - the software stack that the whole world uses for AI and gaming.

Forget about chatbots in the cloud at $20/month. The RTX Spark can run an LLM of 120 billion parameters with a context window of one million tokens directly on your machine. Without internet, without a subscription, without sending your data somewhere. To give you an idea: the best models available to the general public today require entire servers for this.

Nvidia and Microsoft worked together on a dedicated security layer for local AI agents. The idea: you launch an agent that can navigate through your Windows apps, handle complex tasks in the background, code things, search through your files — all without your data leaving the machine. Nvidia calls it OpenShell , a runtime that defines what the agent is allowed to do or not, and that automatically anonymizes personal information if a request still goes to the cloud.

Jensen Huang summed it up in one sentence: "For forty years, you launched apps. You clicked, you typed. With RTX Spark, you ask — and the PC does the job." A bit salesy as a pitch, but technically the hardware is there to do it.

Even the MacBook Pro with their M4 Ultra and 192GB of unified RAM do not yet offer AI agents integrated natively at this level in the OS. This is where Nvidia scores a real point.

That's the question everyone is asking. With 6,144 CUDA Blackwell cores, RT Cores, and full DLSS (including Multi Frame Generation), the RTX Spark promises 1440p at over 100 fps with ray tracing on the AAA headlines. Nvidia showed Doom The Dark Ages and Fortnite running properly on the prototypes.

Let's not kid ourselves: we're not going to compete with an RTX 5090 in a tower. The RTX Spark targets ultra-thin laptops, not desktop gaming setups. But for a 1.3 kg laptop that lasts all day on battery, the gaming performance announced is frankly impressive. We're far from Qualcomm's Snapdragon X Elite, which struggled with any slightly demanding game.

Nvidia takes advantage of this to announce the DLSS 4.5 with an enhanced Ray Reconstruction (second-generation transformer model), scheduled for August 2026. And support is coming to Blender 5.3 this fall, along with a dozen additional games including Marvel Rivals, Phantom Blade Zero, and Gothic 1 Remake.

Click to enlarge

Adobe has announced a complete overhaul of Photoshop and Premiere Pro for RTX Spark. Not just a simple ARM port - a rewrite of the rendering pipeline to take advantage of unified memory, the Blackwell GPU, and TensorRT. The announced result: gains up to 2x faster on AI, colorimetry and effects.

Premiere will integrate a new video pipeline that leverages unified memory to handle complex timelines in real time. Photoshop, for its part, is moving to a GPU-accelerated compositing engine with live filters, native HDR, and natural brushing. Firefly's Generative Fill runs locally on Spark without going through the Adobe cloud.

Substance 3D Painter and Stager will also be natively supported on ARM. That's the kind of thing that can really tip the scales for creators who were hesitating between Mac and PC — especially if the performance lives up to its promises.

The first RTX Spark machine is the Surface Laptop Ultra from Microsoft, presented as the most powerful Surface laptop to date. Aluminum chassis, OLED tandem screen with G-SYNC, 14 mm thick. We are clearly in the premium segment positioned against the MacBook Pro.

Behind Microsoft, all the major manufacturers follow:

• ASUS — laptops and compact desktops
• Dell — XPS 16 Creator Edition (oriented towards creatives)
• HP — OmniBook RTX Spark (one of the thinnest laptops)
• Lenovo — full range (creators, gamers, devs)
• MSI — performance-oriented compact desktop
• Acer and GIGABYTE — announced for later

Expected availability: autumn 2026 . No official price for the moment, but given the positioning (128 GB unified memory, OLED screens, premium chassis), expect high-end. We are probably talking about 1,500-2,500€ depending on the configurations, but this is speculation at this stage.

It's the touchy subject. Windows on ARM, it's been complicated for years. Qualcomm's Snapdragon X Elite last year looked nice on paper but in practice, lots of apps didn't run, missing drivers, incompatible anti-cheat that made Fortnite and Valorant unplayable.

Nvidia promises that the RTX Spark will run "every application that Windows has ever run". The trick: Prism , Microsoft's emulator that translates x86 instructions to ARM on the fly. Nvidia says it has optimized it specifically for its chip, with support for AVX and AVX2 instructions that were missing before.

On anti-cheat — the real puzzle of gaming on ARM — Nvidia announces working with Easy Anti-Cheat, BattlEye, and Denuvo for native ports. It's promising, but each publisher will have to do the work on their own. We've already seen Qualcomm try the same move without reaching critical mass.

What changes this time is Nvidia's weight in the industry. When the manufacturer that controls 80% of the AI data centers market tells developers to recompile for ARM, they have a little more interest in getting moving than when Qualcomm asked politely. But as long as we haven't had real test machines, we remain cautious.

On the day of the announcement, AMD lost more than 6% on the Stock Exchange. Intel the same. Nvidia gained 3.8%. The market understood the message: if Nvidia succeeds with the RTX Spark, it's a whole section of the business laptop that shifts.

AMD and Intel are not going to sit idly by, obviously. AMD has its Ryzen AI which is progressing well, and Intel is working on its upcoming Arrow Lake-H generations. But the CPU+GPU+AI combo in a single 80W chip with the entire CUDA stack behind it, neither can offer that today. And Apple, which has been leading the efficient ARM since 2020, finds itself with a direct competitor in the field of unified memory — but with a much more powerful GPU.

The RTX Spark is probably the most significant hardware announcement of 2026. Nvidia is not just putting a GPU in a laptop — they are redefining what a laptop can do by merging CPU, GPU, and AI into a single low-power chip.

The promises are huge: local AI without cloud, 1440p gaming in an ultrabook, Adobe is redoing its apps in native, all major manufacturers are on board. But we are still in unknown territory regarding Windows on ARM software compatibility, and we have no independent benchmarks. We have already seen Computex announcements not survive contact with the real world.

Meet in the fall of 2026 for the verdict. In the meantime, if you were planning to buy a high-end gaming or creative laptop, it might be worth waiting a few months.

A super ARM chip that combines a Grace 20-core CPU and a Blackwell RTX GPU (6,144 CUDA cores) with 128GB of unified memory. Nvidia's first complete SoC for laptops and compact desktops running Windows.

The first laptops and desktops equipped with RTX Spark will arrive in the fall of 2026, with machines from Microsoft (Surface Laptop Ultra), ASUS, Dell, HP, Lenovo, and MSI. Acer and GIGABYTE will follow a little later.

No official price has been communicated. Given the premium positioning (128 GB unified memory, OLED screens, aluminum chassis), we can expect machines between €1,500 and €2,500 depending on the configurations.

Nvidia promises full compatibility thanks to Microsoft's Prism emulator (translation x86 to ARM). In practice, most apps should work, but some very old software or specific drivers could pose a problem. Anti-cheat systems are currently being natively ported.

Yes. With the same 6,144 CUDA cores as an RTX 5070, the Spark aims for 1440p at 100+ fps with ray tracing and DLSS. Nvidia showcased demos with Doom The Dark Ages and Fortnite. DLSS 4.5 is coming in August 2026.

Both use ARM architecture with unified memory, but the RTX Spark embeds a much more powerful GPU (6,144 CUDA cores vs Apple's integrated GPU), the entire CUDA/RTX/DLSS stack for gaming and AI, and can run AI models with 120 billion parameters locally. However, Apple chips have the advantage of a perfectly optimized macOS ecosystem.