Nvidia confirms Vera Rubin AI chips are in production as it trims rack memory to curb costs

Nvidia CEO Jensen Huang told reporters that the company’s Vera Rubin AI chips are already in production, with "giant amounts" slated for shipment. This affirmation comes after reports of delays to the related Kyber rack system, which Nvidia says does not affect the Rubin chip roadmap.

Separately, analysts from GF Securities reported that Nvidia is halving the memory capacity of its Vera Rubin NVL72 rack‑scale AI system to mitigate soaring HBM4 memory prices. The company plans to cut the small‑outline memory modules from 192 GB to 96 GB and reduce total CPU memory from roughly 55 TB to 28 TB, actions that could lower rack costs by up to 50 %. The adjustments aim to keep the high‑performance Vera Rubin platform commercially viable amid a broader memory shortage.