Bringing the Spark to Europe: Local Intelligence Powers Next Wave of Agents

NVIDIA laid out its whole local AI lineup in Berlin: RTX Spark laptops for personal agents on Windows, a new 64GB DGX Spark shipping from partners on October 23, clustering of up to eight DGX Sparks, and DGX Station for teams, with a Windows version due in December. Around the hardware came a pile of software: a Model Launcher in NVIDIA Sync, seven enterprise management integrations for DGX Station, robotics reference workflows, and day-zero support for a long list of open models.

The headline for us is not any single box. It is that a credible stack for running capable agents on hardware an organization owns is filling in fast, from a 4GB voice model up to a desk-side machine NVIDIA says can serve more than 30 users.

We like this direction, and we want to say why plainly. Local inference changes who holds the data. Underdog, the desktop agent app NVIDIA highlighted, says it runs models entirely on device so personal data never leaves the PC. For lawyers, clinicians, engineers with export-controlled designs, or anyone who simply does not want their inbox flowing through someone else's servers, that is a real benefit, not a marketing line.

The most interesting item to us is Recursive Flint, a new open-weights coding model under the Apache 2.0 license. It is a 40-billion-parameter mixture-of-experts model with 2.5 billion active parameters, and NVIDIA claims over 220 tokens per second on a 64GB RTX Spark. Its maker says an automated research system built custom kernels that sped up generation by 55% without touching the weights. Those are vendor numbers and we want independent benchmarks, but a permissively licensed coding model that runs fast on a laptop is the kind of thing that genuinely widens access.

Now the caveats. This is NVIDIA's own blog at its own conference, so every performance figure here is self-reported. And "local" is not the same as "independent." The software layer, from Sync to Model Launcher to Cloud Functions on DGX Station, is NVIDIA's, which means the local AI future being sketched still runs through one company's ecosystem. Owning the hardware does not mean owning the stack.

There is a security angle too. Many RTX Spark laptops ship with unlocked bootloaders to encourage tinkering, and agents running on a desk with access to local files and tools are exactly the setups that need careful permissioning. Moving agents closer to sensitive data cuts some risks and creates others. We would like to see the same energy put into guardrails for local agents that is going into speed.

Where this leads if it scales: a split market, with frontier models in the cloud and a growing share of everyday agent work, coding, drafting, robotics training loops, happening on machines that sit in the office. That is healthier than total cloud concentration, as long as the open-weights part of the story stays open.

NVIDIA says RTX Spark launched October 16, offering up to 128GB of unified memory and 1 petaflop of AI compute. DGX Station is quoted at 748GB of coherent memory and up to 20 petaflops, and an eight-unit DGX Spark cluster at 1TB of unified memory. Laptops from ASUS, Dell, HP, Lenovo, Microsoft and MSI are listed as available, with Acer coming.