At GTC Taipei during COMPUTEX, NVIDIA and Microsoft introduced NVIDIA RTX Spark, a Windows PC platform for personal AI agents. It puts Blackwell RTX graphics, Grace CPU cores, up to 128GB of unified memory and a claimed 1 petaflop of AI performance into slim laptops and compact desktops.
Two other machines come up in the same conversation. NVIDIA DGX Spark is the compact desktop AI supercomputer that started shipping earlier. NVIDIA DGX Station for Windows is a much larger deskside enterprise workstation due in Q4 2026.
If you searched for "NVIDIA Spark workstation laptop", then: RTX Spark is the laptop and compact desktop platform, DGX Spark is the small desktop AI supercomputer, and DGX Station for Windows is the enterprise deskside tier.
What NVIDIA launched
NVIDIA's May 31, 2026 announcement covers RTX Spark, a new class of Windows PCs for local AI agents, creative tools, development and gaming. NVIDIA says RTX Spark laptops and small desktops will be available this fall from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI, with Acer and GIGABYTE models to follow.
| Feature | NVIDIA RTX Spark claim | Why it matters |
|---|---|---|
| AI performance | Up to 1 petaflop | Gives local agents and model workflows more headroom than mainstream thin laptops |
| Memory | Up to 128GB unified memory | Useful for large local models, long context, video, and 3D workloads |
| GPU | Blackwell RTX GPU with 6,144 CUDA cores | Brings CUDA, RTX, Tensor Cores, DLSS, OptiX, and TensorRT into one mobile platform |
| CPU | 20-core NVIDIA Grace CPU | Arm-based CPU design tied to the GPU through NVLink-C2C |
| Form factors | Slim laptops and compact desktops | Targets portable and desk-friendly AI work instead of only tower workstations |
| Availability | Fall 2026 | Hardware is announced, but real buyer decisions should wait for model pricing and benchmarks |
NVIDIA calls RTX Spark a machine for "personal agents": local assistants that reason over your files, automate workflows, generate media, write code and drive Windows apps while staying on your device.
RTX Spark laptops
The laptops are the interesting part. NVIDIA says they can be as slim as 14 millimeters, as light as three pounds, and come in 14- to 16-inch sizes with premium displays. AI developers usually associate local model work with much heavier mobile workstations.
NVIDIA's performance claims are big. It says RTX Spark can run 120-billion-parameter LLMs with up to a 1 million token context locally, edit 12K 4:2:2 video, render very large 3D scenes, and play AAA games at 1440p at over 100 frames per second.
Memory is the real change. A typical creator laptop has 16GB or 32GB of RAM and a discrete GPU with far less VRAM. A 128GB unified pool that the CPU and GPU share is much more useful for large models and long-context agents.
That doesn't make every RTX Spark laptop a good local LLM machine. Before buying, check:
- whether the exact laptop ships with the full 128GB configuration
- how your favorite AI tools run under Windows on Arm
- how CUDA, TensorRT, llama.cpp, ComfyUI, Blender, Adobe tools and local inference frameworks behave on retail hardware
- thermals, noise, battery life and sustained performance under long inference loads
- price against RTX PRO workstations, Mac Studio-class machines or cloud GPU access
DGX Spark
DGX Spark is the compact desktop AI supercomputer NVIDIA started shipping in October 2025. It uses the NVIDIA GB10 Grace Blackwell Superchip and delivers up to 1 petaflop of FP4 AI performance with 128GB of coherent unified memory.
NVIDIA pitches it at developers, researchers, data scientists, robotics teams and students who want to prototype and run AI locally. The company says DGX Spark can run inference on models up to 200 billion parameters and fine-tune models up to 70 billion parameters locally.
It is a dedicated local AI box more than a daily-driver laptop. You put it on a desk to run agents, model experiments, fine-tuning jobs or private inference without renting cloud GPUs for every test.
DGX Station for Windows
DGX Station for Windows is aimed at enterprise developers and AI infrastructure teams. NVIDIA says it arrives in Q4 2026 and is built on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip. It targets frontier agents, local inference, data science, fine-tuning and Windows-managed enterprise workflows, and nobody will confuse it with a laptop replacement.
| Feature | DGX Station for Windows |
|---|---|
| Processor platform | NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip |
| CPU | 72-core NVIDIA Grace CPU |
| AI performance | Up to 20 petaflops FP4 |
| Memory | Up to 748GB coherent memory |
| Model target | Up to 1 trillion-parameter AI models |
| Networking | NVIDIA ConnectX-8 SuperNIC, up to 800Gb/s |
| Optional GPU pairing | NVIDIA RTX PRO 6000 Blackwell Workstation GPU |
| Availability | Coming in Q4 2026 |
The idea is to keep powerful agents next to the Windows applications, files, workflows and security systems employees already use. Teams run large local workloads on managed deskside hardware instead of sending every agent job to a cloud endpoint.
The three products compared
| Product | Best fit | Memory | Performance target | Practical role |
|---|---|---|---|---|
| RTX Spark laptop | Creators, developers, gamers, mobile agent workflows | Up to 128GB unified memory | Up to 1 petaflop AI | Daily laptop that can run local AI agents and creative workloads |
| RTX Spark compact desktop | Desk users who want Spark in a small PC | Up to 128GB unified memory | Up to 1 petaflop AI | Small AI PC for local agents, media, and productivity |
| DGX Spark | AI developers, labs, researchers, robotics teams | 128GB coherent unified memory | Up to 1 petaflop FP4 | Dedicated compact desktop AI supercomputer |
| DGX Station for Windows | Enterprises and frontier AI developers | Up to 748GB coherent memory | Up to 20 petaflops FP4 | Deskside workstation for 1T-parameter agents and heavy local workloads |
The names are close but the buyers differ. RTX Spark is an AI PC, DGX Spark a compact AI supercomputer, and DGX Station a deskside enterprise system.
What it changes for local AI
AI hardware is moving closer to ordinary laptops and workstations. Local AI has usually meant a compromise:
- a cloud model, with API costs, latency and data-sharing limits
- a gaming GPU, with models squeezed into limited VRAM
- a workstation GPU, at a much higher price
- a Mac or other unified-memory machine, with less CUDA ecosystem depth
RTX Spark targets the middle of that list. If it works as NVIDIA claims, developers could run larger local agents on a laptop and keep NVIDIA's software stack. Private coding agents, local document search, media generation, RAG testing and offline prototyping all benefit.
VRAM planning gets more important, not less. Unified memory helps with model size, but you still need to estimate model weights, quantization, KV cache, context length, batch size and concurrent users. Before buying any Spark-class device, compare your expected workload with the AI VRAM Calculator and the local LLM VRAM guide.
Should you buy one
Most people should wait for retail pricing, independent benchmarks, thermal results and software compatibility reports. The first wave will likely go to premium buyers, developers, creators and AI enthusiasts rather than casual laptop shoppers.
Watch RTX Spark laptops if you want a portable machine for local agents and creative AI, need more memory headroom than gaming laptops give, use CUDA, TensorRT, ComfyUI, Blender, Adobe or local inference tools, or want one laptop for development, media and gaming.
Watch DGX Spark if you want a small desk machine dedicated to AI experiments, care more about local model work than portability, and want a more appliance-like NVIDIA AI software stack.
Watch DGX Station for Windows if you are building enterprise AI agents tied to Windows workflows, need very large local model capacity, and want managed deskside infrastructure rather than a personal laptop.
NVIDIA has announced a direction. Whether to buy depends on prices, OEM configurations, software support and sustained workloads, none of which exist yet. Match the machine to the workload: memory configuration, software support, thermals and cost will decide it.
For more, see ToolMintX on best local AI models in 2026, LM Studio vs Ollama, and NVIDIA's AI infrastructure demand.
FAQ
Is RTX Spark the same as DGX Spark?
No. RTX Spark is the laptop and compact desktop PC platform. DGX Spark is NVIDIA's compact desktop AI supercomputer, built around the GB10 Grace Blackwell Superchip.
When will RTX Spark laptops launch?
NVIDIA says laptops and compact desktops will be available in fall 2026 from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI, with Acer and GIGABYTE models to follow.
Should local AI users buy RTX Spark or a normal GPU workstation?
It depends on workload and price. RTX Spark may suit portable local agents and large unified-memory workflows. RTX PRO workstations may still be better if you need discrete workstation GPUs, mature desktop thermals and established professional configurations.
Sources
- NVIDIA Newsroom: NVIDIA and Microsoft reinvent Windows PCs for the age of personal AI
- NVIDIA Investor Relations: RTX Spark announcement
- NVIDIA Blog: Faster local AI agents on RTX PCs and DGX Spark
- NVIDIA Newsroom: DGX Station for Windows
- NVIDIA product page: DGX Spark
- NVIDIA Newsroom: DGX Spark arrives for world's AI developers
