Vera Rubin GPUs to train MoE models vs Blackwell — one-fourth the NUMBER of GPUs (~75% fewer, ~4×); a TRAINING metric, not inference

Name: Vera Rubin GPUs to train MoE models vs Blackwell — one-fourth the NUMBER of GPUs (~75% fewer, ~4×); a TRAINING metric, not inference
License: https://aidatacenterguide.com/terms#intellectual-property

~1/4 the number (~75% fewer, ~4×)observed

Value kind	observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range.
Caveat	NVIDIA's phrasing is 'one-fourth the number of GPUs' (= ~75% fewer / ~4×), NOT a 25% cut, and it is a training claim, not an inference metric.
As of	2026-07
Source	NVIDIA (Vera Rubin NVL72 product page) — training
Review	checking…review by 2026-09-11 · fast cadence
Recorded changes	last 2026-07-22
Claim id	vera-rubin-gpus-to-train-moe-vs-blackwell

Where the guide uses it

Not quoted in a chapter yet; it is kept in the curated register.

← Full numbers register — every date-stamped figure in the guide, with revision history.