Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
InfrastructureInsight · On RadarTNW→ original

Nvidia RTX Spark Mini PCs With 128 GB Memory Can Run 120B-Parameter AI Models, Starting at $4,399

Nvidia has started selling mini PCs built on its Arm-based RTX Spark chip, which pairs a 20-core Grace CPU and a Blackwell GPU that share up to 128 GB of unified memory. Nvidia says that is enough to run a 120-billion-parameter model directly on the desk, with no cloud bills. Models from ASUS, Dell and MSI start at $4,399, roughly $700 more than laptops with the same chip. For an always-on agent, the price is worth comparing against an annual cloud bill.

AI-processed from TNW; edited by Hamidun News
Nvidia RTX Spark Mini PCs With 128 GB Memory Can Run 120B-Parameter AI Models, Starting at $4,399
Source: TNW. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTNW ↗
Editor-in-Chief
MSI EdgeMesa N AI+$4,399 with 64 GB of memory and $7,499 with 128 GB
Key metric #2ASUS ProArt GR1X starts at $4,499; Acer, GIGABYTE and HP are coming later
Windows Centralgraphics are 22% faster than the laptop chip, the CPU 5% faster

Cloud AI services are convenient, until the token bill becomes a permanent line item and your data has to go to someone else's servers. Nvidia is offering another route: put a compact computer next to your monitor that runs language models and AI agents on its own. Machines built on the new RTX Spark chip have already gone on sale.

RTX Spark is Nvidia's Arm chip for Windows PCs. Its CPU and GPU share a common pool of memory, up to 128 GB. A "120-billion-parameter model" is a very large language model with 120 billion adjustable numbers; the more parameters, the more capable the model, and the more memory it needs.

Nvidia claims 128 GB is enough for a model of this size. And $4,399 is the price of the cheapest of the first models, MSI's version with 64 GB of memory.

What Is Actually Being Sold

ASUS, Dell and MSI were first to release RTX Spark mini PCs. Acer, GIGABYTE and HP are expected to follow, along with Microsoft's Surface RTX Spark Dev Box. All of these are Windows machines on the Arm architecture. They include a 20-core Nvidia Grace CPU, a Blackwell-generation GPU and up to 128 GB of memory shared between the CPU and graphics.

Shared memory is the key point here. A typical graphics card is limited by its own memory capacity, and a large model simply does not fit. With unified memory, almost the entire pool can be allocated to the model. Nvidia states that 128 GB is sufficient to run 120-billion-parameter models directly on the device, without paying for cloud compute. This is the manufacturer's claim: the material contains no independent measurements of generation speed.

Nvidia also positions these computers as "always-on" boxes for AI agents. The logic is straightforward: an agent that works through email around the clock, monitors logs or runs tests is cheaper to host on your own hardware than to pay for every request it makes.

What It Costs

The prices are steep. The MSI EdgeMesa N AI+ costs $4,399 with 64 GB of memory and $7,499 with 128 GB. The ASUS ProArt GR1X starts at $4,499. According to Windows Central's calculations, that is about $700 more than RTX Spark laptops, which went on sale in October starting at $2,599.

The premium buys a modest amount: the outlet measured up to 22% higher graphics performance than the laptop chip, thanks to higher power draw and better cooling. The CPU is only about 5% faster. In other words, the mini PC wins mainly on graphics and on sustaining performance without overheating, not on general-purpose speed.

Software and Who It Is For

Nvidia has also prepared the software side. Adobe Illustrator and InDesign now run natively on the RTX Spark Arm chip, without emulation. Blender 5.3 added DLSS, Nvidia's technology for accelerating ray tracing, in the viewport preview. So the machine is aimed not only at enthusiasts running local models, but also at designers and 3D artists.

For developers and small teams, the takeaway is practical. If you spend heavily on cloud models, need to keep sensitive data in-house, or are building an agent that must run continuously, a 128 GB mini PC is worth comparing against an annual cloud bill. The 64 GB version is cheaper, but its ceiling for the largest models will be lower.

There are caveats. The platform is new and based on Arm, so before buying, check whether your working stack runs on it: libraries, drivers and inference tools. The 120-billion-parameter figure currently comes from Nvidia itself, and real response speed depends on quantization and context length. For occasional tasks, the cloud may still be cheaper, but for sustained workloads a local machine starts to pay for itself.

The arrival of these computers shows where the market is heading: Nvidia is trying to establish itself in Windows PCs not only as a supplier of gaming graphics cards, but also as a maker of complete platforms for local AI.

Story

Key Questions Answered

What Is Actually Being Sold?

ASUS, Dell and MSI were first to release RTX Spark mini PCs. Acer, GIGABYTE and HP are expected to follow, along with Microsoft's Surface RTX Spark Dev Box. All of these are Windows machines on the Arm architecture. They include a 20-core Nvidia Grace CPU, a Blackwell-generation GPU and up to 128 GB of memory shared between the CPU and graphics. Shared memory is the key point here. A typical graphics card is limited by its own memory capacity, and a large model simply does not fit.

What It Costs?

The prices are steep. The MSI EdgeMesa N AI+ costs $4,399 with 64 GB of memory and $7,499 with 128 GB. The ASUS ProArt GR1X starts at $4,499. According to Windows Central's calculations, that is about $700 more than RTX Spark laptops, which went on sale in October starting at $2,599. The premium buys a modest amount: the outlet measured up to 22% higher graphics performance than the laptop chip, thanks to higher power draw and better cooling. The CPU is only about 5% faster.

What are the key details on Software and Who It Is For?

Nvidia has also prepared the software side. Adobe Illustrator and InDesign now run natively on the RTX Spark Arm chip, without emulation. Blender 5.3 added DLSS, Nvidia's technology for accelerating ray tracing, in the viewport preview. So the machine is aimed not only at enthusiasts running local models, but also at designers and 3D artists. For developers and small teams, the takeaway is practical.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…