Nvidia this week shipped general availability of Project DIGITS, a developer-scale AI workstation built around the new GB10 Grace Blackwell Superchip. The box delivers up to one petaflop of FP4 AI performance, 128 GB of unified coherent memory, and up to 4 TB of NVMe storage — and draws power from a standard wall outlet.
Technically, DIGITS is a scale-down of the same Grace Blackwell architecture shipping in Nvidia's data-center-grade HGX systems, linked via NVLink-C2C to a 20-core Arm CPU. Practically, it is a distribution bet: Nvidia has watched Apple, AMD, and Qualcomm nibble at the edges of on-device inference and has decided to come downmarket with Blackwell before any of them can build a credible training alternative.
The list price lands at $3,999 for the base config and $4,999 for the upgraded SKU. Early access units went to a short list of frontier labs and selected research universities in January; broad availability begins this week.
Key facts
- 01Nvidia. A key thread of reporting in this story.
- 02GB10. A key thread of reporting in this story.
- 03DIGITS. A key thread of reporting in this story.
Developer reaction has been strong but measured. In a thread on Hacker News, one commenter noted that DIGITS effectively turns what used to be a three-person DGX cluster into a single desk. Others pointed out that at a petaflop, DIGITS is still roughly two orders of magnitude short of the compute needed to train a Claude- or GPT-class frontier model — and that Nvidia is likely more than happy to keep that gap where it is.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
