Back to the stories

NVIDIA RTX Spark launches as a local development platform for AI coding agents

Score 9.7,

Agent infrastructure and tooling

You can now build, test and run coding agents entirely on a single laptop.

NVIDIA packaged a hardware and software stack called RTX Spark that puts a Blackwell RTX GPU and a Grace CPU together with shared memory, and a developer toolchain that supports CUDA and PyTorch as well as popular local runtimes.

It includes NVIDIA Skills, an open source repository of routines that teach agents how to use NVIDIA's GPU libraries, and NVIDIA says it worked with GitHub Copilot, Pi, Hermes and OpenClaw to improve agent support.

Until now, building and iterating on coding agents mostly meant renting powerful cloud machines or stitching together separate development and test environments. RTX Spark lets teams keep the whole loop local, which speeds iteration, reduces cloud costs and keeps sensitive data on the device.

Think of the system as a mini supercomputer in a laptop: a fast AI GPU and a matching CPU share a large pool of memory so models and their working data can run together instead of being copied back and forth. NVIDIA says the platform delivers about one petaflop of FP4 AI compute and up to 128 GB of unified memory. That bridge between compute and memory is what makes running models, adjusting them for your data and testing agents practical on one machine.

Developers, startups and product teams can prototype coding agents faster, test them on real local data and ship proofs of concept without first buying cloud time or waiting for approvals. The systems are already available from OEMs including ASUS, Dell, HP, Lenovo, Microsoft and MSI. This is not cheap consumer hardware, so serious fine-tuning and very large models will still live in data centers.

The real test will be independent benchmarks and developer stories: can teams turn this hardware and the open agent tools into useful, on-device products that compete with cloud-first services? Watch for performance numbers and real-world apps in the months ahead.