Asus’ New ProArt Laptops Can Run 120B Parameter AI Locally

A sleek Asus ProArt laptop open on a desk displaying an AI workflow with neural network graphics

Running a 120-billion-parameter AI model used to mean renting time on a cloud server farm. Asus’ newest ProArt laptops change that equation entirely. The ProArt P16 and ProArt P14, built on NVIDIA’s new RTX Spark platform, are designed to run massive AI workloads — including LLMs with up to 120 billion parameters — entirely on-device, no internet connection required.

Unveiled at Computex 2026 and showcased again at IFA 2026 ahead of their launch, the P16 and P14 represent one of the most aggressive pushes yet toward local AI computing in a portable form factor.

Key Takeaways

  • Asus’ new ProArt P16 and P14 laptops are powered by NVIDIA’s RTX Spark superchip, pairing a Blackwell RTX GPU with a 20-core Grace CPU.
  • With up to 128GB of unified memory and 1 petaflop of FP4 AI performance, they can run 120-billion-parameter LLMs locally with a 1-million-token context window.
  • The 16-inch P16 is just 12.9mm thick with a 4K 120Hz OLED display; the 14-inch P14 packs a 3K OLED into a 1.48kg chassis.

What Is RTX Spark?

The heart of both laptops is the NVIDIA RTX Spark superchip, a purpose-built platform that combines an Arm-based Grace CPU with a Blackwell RTX GPU. Here’s the headline architecture:

  • Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores supporting FP4 precision
  • 20-core NVIDIA Grace CPU, connected to the GPU via the NVLink-C2C chip-to-chip interconnect
  • Up to 128GB of unified memory, shared across CPU and GPU workloads
  • Up to 1 petaflop of FP4 AI performance

The unified memory design is the real breakthrough. Traditional laptops split memory between the CPU (system RAM) and the GPU (VRAM), creating a bottleneck for AI workloads that need to load enormous model weights. By pooling up to 128GB into a single shared space, RTX Spark removes that bottleneck — the GPU can access the full memory pool directly.

That is what makes the 120-billion-parameter claim possible. Large language models need somewhere to live before the accelerator can process them, and a 120B model in efficient FP4 precision fits within that 128GB pool. Asus says the platform can also handle a context window of up to 1 million tokens — meaning the AI can keep enormous documents, codebases or conversations in working memory.

Not Just AI: Serious Creator Hardware

While local AI is the headline, these are ProArt machines first — built for creators. Asus claims the RTX Spark platform can render 3D scenes exceeding 90GB in size, edit 12K 4:2:2 video footage, and generate 4K AI video. Gaming is in the mix too, with support for AAA titles at 1440p and over 100 frames per second.

The displays are classic ProArt:

ProArt P16 (H7607) ProArt P14 (H7407)
Display 16-inch 4K Lumina Pro OLED 14-inch 3K Lumina Pro OLED
Refresh rate 120Hz variable, G-Sync 120Hz variable
Colour 100% DCI-P3, factory calibrated 100% DCI-P3, factory calibrated
Peak HDR brightness Up to 1,600 nits Up to 1,600 nits
Thickness 12.9mm 13.9mm
Weight 1.77kg 1.48kg
Battery 99.9Wh 90Wh

Both models feature anti-reflection coatings, Wi-Fi 7, USB4 connectivity and extensive I/O — and despite the heavy-duty internals, the P16 is barely thicker than an ultrabook at 12.9mm.

Software Built for Local AI

Asus is bundling a suite of AI tools designed to exploit the hardware:

  • MuseTree — local image generation using FLUX and WAN models, running entirely on the machine
  • StoryCube — AI-powered media organisation and asset management
  • Hermes Agent — an open-source autonomous AI agent from Nous Research that can execute tasks across local apps and files
  • ComfyUI — native support for the popular node-based interface for visual AI workflows

The practical pitch is straightforward: run generative AI workflows, code assistants and local agents without paying per-use cloud API fees — and without sending private files to a remote server.

Why This Matters for Pakistani Users

For developers, researchers and creators in Pakistan, local AI has particular appeal. Cloud GPU access can be expensive, slow or unreliable depending on connectivity, and data-privacy regulations — not to mention client confidentiality — make on-device processing attractive for professional work. A laptop that can run a 120B model locally could be transformative for AI researchers, software houses and content studios that want cutting-edge capabilities without recurring cloud bills.

There are caveats. Asus has not announced Pakistani pricing or local availability, and flagship AI laptops like these typically command premium prices in Pakistan once import duties and taxes are factored in. The 1-petaflop figure also reflects FP4 precision — ideal for inference, less so for training large models from scratch.

The Bigger Picture

The ProArt P16 and P14 — alongside a new ProArt Mini PC on the same platform — signal where the industry is heading. For years, the most capable AI was locked behind data-centre doors. RTX Spark brings data-centre-class local AI into a 1.48kg laptop.

Whether most creators need a 120-billion-parameter model on their desk is an open question. But for those who do — AI developers, researchers, studios working with sensitive data — Asus has just made the cloud optional.