Summary
- ASUS’s new ProArt P14, P16, and GR1X use Nvidia RTX Spark to run demanding AI workloads locally.
- Up to 128GB of unified memory enables 120-billion-parameter models, 4K AI video generation, and complex agentic workflows.
- Local processing can eliminate token charges and keep sensitive files on-device, while cloud services remain available when needed.
Tired of juggling monthly AI subscriptions and sending your private data to big companies? ASUS thinks it has the answer. The company has introduced three ProArt computers specifically designed to run serious AI workloads locally instead of sending every prompt and file to somebody else’s server. The ProArt P14 and P16 are laptops, while the GR1X is a palm-sized mini PC. All three are powered by Nvidia’s new RTX Spark platform.
According to ASUS’s official IFA 2026 announcement, RTX Spark delivers up to one petaflop of AI performance and up to 128GB of unified memory. ASUS says that is enough to run language models with as many as 120 billion parameters, render 90GB-plus 3D scenes, and generate 4K AI video directly on the computer.
Cloud-based AI finally has some serious competition
No tokens, no queue, and no need to upload your data
Just about every tech company on Earth is pushing AI jargon alongside its latest products. Agentic this, AI-assisted that — we’ve heard it all. What makes the AI label on these new ASUS ProArt machines different is the sheer computing power behind it. The specifications suggest you could repeatedly run local models and agents to do serious work without paying for every API call or stressing over a monthly token counter.
If you’re comfortable setting everything up yourself, you can already run powerful AI locally without a subscription using free tools such as Ollama and LM Studio. The advantage of these RTX Spark machines is that they not only raise the ceiling considerably but also arrive in preconfigured, portable form factors. That gives developers and creators a middle ground between accepting the limitations of consumer hardware and the hassles of building a powerful AI machine themselves.
For context, several useful local AI models can run with 8GB of VRAM or less, although smaller and heavily quantized models sacrifice some capability. But thanks to up to 128GB of unified memory available in these ProArt computers, local video generation and complicated agentic workflows become considerably more plausible.
There’s also the question of where your work goes. If your work includes dealing with confidential documents, switching to local AI should be a priority. With ProArt computers, ASUS says all your files, prompts, and creative assets remain on the device when using local models. Those models can continue working offline, too, without requiring a connection to a cloud provider.
How much you could save with a local setup depends on how heavily you use AI and what you use it for. As our local Claude Code experiment demonstrated, the software coordinating an agent can be free while the cloud model processing each request keeps generating charges. Local hardware replaces that variable expense with one potentially enormous bill upfront — one that still remains a mystery because ASUS did not include pricing in its announcement.
The ProArt systems preload MuseTree for token-free local image generation using FLUX and WAN, while StoryCube handles AI-powered media organization. They can still connect to cloud services, including Adobe Creative Cloud and, on eligible laptops, an optional Google AI bundle. If you’re already pushing Claude or other such AI tools to their limits, that hybrid approach, for now, is probably the most realistic version of personal AI.