The PC just got reinvented — or so Jensen Huang says
Nvidia unveiled RTX Spark at Computex 2026 in Taiwan, and it’s the most significant shift in personal computing hardware since the smartphone. An Arm-based “superchip” that fuses a Blackwell GPU with a Grace CPU, running up to 1 petaflop of AI compute with 128GB of unified memory. The pitch: instead of clicking and typing, you ask your computer to do things, and AI agents handle the rest.
“You set the objective. The machine handles the rest,” Nvidia’s marketing declares. “There’s intelligence on both sides of the keyboard now.”
🔍 THE BOTTOM LINE
Nvidia is betting that the next era of personal computing isn’t about faster apps — it’s about AI agents that work for you locally, 24/7, without cloud dependency. If RTX Spark delivers, the laptop becomes a teammate, not a tool.
What is RTX Spark?
What is RTX Spark? RTX Spark is Nvidia’s first Arm-based consumer processor — a “superchip” that combines a GPU and CPU on a single package, designed specifically to run personal AI agents locally on Windows laptops and desktops. It builds on the same GB10 architecture as Nvidia’s DGX Spark (a $3,999 mini PC for AI researchers) but targets consumers running Windows 11 instead of Linux.
Key specs:
| Spec | RTX Spark |
|---|---|
| GPU architecture | Nvidia Blackwell |
| CUDA cores | 6,144 |
| CPU | 20-core Nvidia Grace (Arm-based) |
| Manufacturing | TSMC 3nm |
| Unified memory | Up to 128GB LPDDR5X |
| AI compute | Up to 1 petaflop |
| Max local model size | ~120 billion parameters |
| Interconnect | NVLink-C2C chip-to-chip |
| OS | Windows 11 only |
The 128GB unified memory is the standout number. That’s shared between CPU and GPU, meaning you can load large AI models entirely into memory and run them locally — no cloud API, no latency, no data leaving your device. For context, that’s enough to run models with roughly 120 billion parameters on-device. That’s not GPT-4 territory, but it’s well beyond what any current consumer laptop can manage.
The agent pitch
Huang’s vision goes way beyond running ChatGPT offline. He’s talking about autonomous AI agents that run 24/7 on your machine, handling tasks across applications without you touching the mouse or keyboard.
“I could totally imagine some day there is an AI super computer in your house, and it’s running all of your agents, it’s running all of your assistants,” Huang said at the keynote. “And you have to have it in your house, just like you have a home theater in your house.”
This is the same direction we covered in AI Agents Deployed: What Happens to Us? — the shift from AI as a chat interface to AI as an always-on worker. The difference is RTX Spark makes it local and private by default.
The hardware ecosystem
RTX Spark laptops are coming this fall from the major PC makers:
- Launch partners: Asus, Dell, HP, Lenovo, MSI, Microsoft Surface
- Following later: Acer, Gigabyte
- Form factors: 14-16 inch laptops (as slim as 14mm, as light as 3 pounds), plus compact desktops
- Initial rollout: Six premium laptop models, expanding to 30 laptop models and 10 mini desktops
- Target market: Content creators, AI developers, gamers
Microsoft CEO Satya Nadella endorsed the launch: “Our goal is to deliver unmetered intelligence to every home and every desk with Windows. RTX Spark marks a real breakthrough towards that vision.”
The “unmetered” bit is key. Cloud AI APIs charge per token. RTX Spark runs your agents for free after you buy the hardware.
Nvidia’s OpenShell — the software side
Nvidia isn’t just shipping silicon. They’ve announced NVIDIA OpenShell, a new runtime for Windows that provides a secure environment for on-device agents. This is the software layer that lets AI agents interact with your apps, files, and system safely — or at least, that’s the promise.
The company also claims 2X inference performance improvements on top agentic models via optimised llama.cpp and vLLM runtimes. Adobe and Blender are rebuilding apps for RTX Spark.
How it compares
RTX Spark enters a market already occupied by Apple’s M-series chips and Qualcomm’s Snapdragon X. The key differentiator:
| Feature | RTX Spark | Apple M4 Max | Snapdragon X Elite |
|---|---|---|---|
| Architecture | Arm (Grace + Blackwell GPU) | Arm (Apple Silicon) | Arm (Oryon) |
| Unified memory | Up to 128GB | Up to 128GB | Up to 64GB |
| AI compute | ~1 petaflop | ~38 TOPS (NPU) | ~45 TOPS (NPU) |
| Discrete-class GPU | Yes (6,144 CUDA cores) | No | No |
| Local AI agents | Purpose-built | Limited | Limited |
| Gaming | RTX 5070-equivalent | Moderate | Light |
That petaflop number is on a different planet from what Apple and Qualcomm offer. It’s because RTX Spark has a full discrete GPU on the same chip — 6,144 CUDA cores with fifth-gen Tensor Cores. Digital Foundry reports gaming performance equivalent to an RTX 5070 laptop GPU. This isn’t an NPU bolted onto a CPU. It’s a gaming-grade GPU fused with an AI-grade CPU.
The NZ angle
RTX Spark won’t be cheap. Nvidia hasn’t announced pricing, but given the specs and the premium laptop positioning, expect these machines to start well north of NZ$3,000 and climb fast from there. The 128GB unified memory configuration will likely be the halo product at a price that makes your eyes water.
For NZ’s growing AI developer community, the appeal is obvious: run frontier models locally, keep data in-country (important for AI compliance in New Zealand), and skip the API costs. For everyone else, the question is whether the AI agent vision is compelling enough to justify the premium over a regular laptop.
The bigger picture
As we noted in Nvidia Concedes China AI Chip Market to Huawei, Nvidia’s China business has effectively collapsed to zero under US export controls. RTX Spark is where they’re hedging — consumer hardware, Windows ecosystem, personal AI computing. If you can’t sell data centre GPUs to China, sell personal AI chips to everyone else.
Nvidia’s stock rose 6% on the RTX Spark announcement. The company’s market cap has risen more than 10x since ChatGPT launched in 2022. Huang is comparing this moment to the smartphone revolution. He might be right — or he might be selling you a very expensive vision of the future. Either way, the hardware is real, and it’s coming this fall.
❓ Frequently Asked Questions
Q: What does this mean for NZ? NZ developers and businesses could benefit from local AI compute that keeps data onshore — important for privacy regulations and data sovereignty. But pricing will be a barrier. Expect early adoption from AI startups and content creators before it reaches mainstream.
Q: Can I run ChatGPT locally on RTX Spark? Not ChatGPT itself — that’s an OpenAI cloud service. But you can run open-source models up to ~120 billion parameters locally. That includes many capable models from Meta (Llama), Mistral, and others. The experience won’t be identical, but for many tasks, it’ll be close — and private.
Q: What should I do? If you’re an AI developer, start planning for on-device deployment now. The hardware is coming; the question is whether your agents can run without cloud dependency. If you’re a regular user, wait for reviews this fall before committing. The first-gen pricing will be steep, and the AI agent ecosystem is still immature.
🔍 THE BOTTOM LINE
RTX Spark is the most credible attempt yet to put serious AI compute in consumer hands. Whether it “reinvents the PC” or just makes expensive laptops slightly more expensive depends entirely on whether AI agents become something people actually need running locally 24/7. Jensen Huang is making the biggest bet in computing since the iPhone. He’s been right before.
SOURCES
- The Independent — Nvidia ‘reinvents PC’ with AI chip
- PCMag — Nvidia Unveils RTX Spark, an Arm-Based Superchip for Windows PCs
- Nvidia — COMPUTEX 2026 announcements
- CNBC — Nvidia’s new chip to power fresh line of Windows laptops
- Digital Foundry — RTX Spark N1/N1X gaming performance equivalent to RTX 5070
