Overview

• Orders open October 15 on NVIDIA.com and through partners, with systems from Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo and MSI, plus U.S. retail availability at Micro Center.
• NVIDIA set the base price at $3,999, updating earlier expectations that referenced a $3,000 figure at the product’s initial reveal.
• The unit features a GB10 Grace Blackwell Superchip, NVLink‑C2C and ConnectX‑7 200 Gb/s networking, 128GB unified CPU‑GPU memory and up to 4TB of NVMe storage, with up to 1 petaflop of AI performance.
• NVIDIA’s guidance says the system can run inference on models up to about 200 billion parameters and fine‑tune models up to roughly 70 billion locally.
• The rollout included a hand‑delivery by CEO Jensen Huang to Elon Musk at SpaceX’s Starbase, early units are with organizations like Anaconda, Google, Hugging Face, Meta and Microsoft for validation, and NVIDIA says two linked Sparks can extend inference to models around 405 billion parameters at 4‑bit precision.