On October 7, NVIDIA and Microsoft jointly announced what the companies call "a new beginning for Windows PCs": DGX Station is getting a Windows version for the first time, packing 748 GB of memory and 20 PFLOPS-class compute into a desktop workstation, while the new RTX Spark laptop line starts at $2,599 — described by commentators as "a data-center chip inside a laptop."
The Launch List: From Desktop Supercomputer to "Data Center in a Backpack"
- DGX Station for Windows: extends beyond Linux for the first time; Blackwell architecture, 748 GB memory, 20 PFLOPS, capable of running trillion-parameter models locally (NVIDIA official / Sina Finance);
- RTX Spark laptops: starting at $2,599, alongside new Nvidia-chip AI PCs from Microsoft and a revamped Windows 11 (TechCrunch / Startup Fortune);
- Surface Laptop Ultra: Microsoft's new machine, reportedly able to run 120-billion-parameter models offline (single source, pending verification);
- ASUS ProArt RTX Spark: pre-orders open now, shipping an integrated ecosystem of AI agents and creative apps (ASUS pressroom).
Multi-Source Facts at a Glance (as of 2026-10-08)
· Launch date: 2026-10-07, joint NVIDIA + Microsoft event (NVIDIA blog)
· Desktop specs: 748 GB memory / 20 PFLOPS / Blackwell (Wccftech / Technetbook)
· Laptop starting price: $2,599 (Startup Fortune / TechCrunch)
· Core pitch: local Agentic AI — agents running entirely on-device
Why Now: The Window for Local Agents
The move marks phase two of the AI PC race. Phase one was about NPUs and offline "parameter counts"; phase two centers on local agents — the revamped Windows 11 redesigns its interaction layer for persistent, cross-app agents. The NVIDIA-Microsoft compute alliance now goes head-to-head with Qualcomm's Snapdragon X efficiency camp and Apple's vertically integrated M-series line.
The real value of "running locally" is not parameter count but data sovereignty: for enterprises, it means sensitive data can complete its inference and decision loop on the endpoint.
A Sober View: Marketing vs. Reality
"Desktop data center" is unmistakably marketing spin — full-power compute still collides with real-world constraints of power draw, thermals and software ecosystem. The more useful takeaway for decision-makers: a hybrid cloud-and-local agent architecture is becoming the default. NineZenith's Zenith-Act platform supports hybrid cloud-local scheduling, while Zenith-Safety sets the security baseline for endpoint agents — precisely aligned with this trend.
(Compiled from public reports by the NVIDIA blog, TechCrunch, Wccftech, ASUS pressroom and Sina Finance, among others)