NVIDIA CorporationMemory supply constraints force Nvidia to cut host-memory capacity in Vera Rubin systems, reducing rack memory from 55TB to 28TB.

Wedbush analyst Matt Bryson warns that Nvidia is cutting host-memory capacity in its upcoming Vera Rubin systems as surging memory prices and tight supply reshape product design, while passing higher component costs into consumer GPU prices. Memory now represents roughly 29% of system cost, above Nvidia's preferred 20%, leading the Vera CPU configuration in the NVL72 to carry 96GB rather than the previously planned 192GB, reducing reported rack memory to about 28TB from 55TB, while GPU-side HBM4 capacity remains unchanged at 20.7TB. Bryson notes the choice appears predominantly dictated by supply constraints, with Nvidia simply not having enough memory to support its targeted volumes despite committing more capital than any other company to supply chain certainty. On the consumer side, Wedbush cites reported graphics-card price increases of 20% to 30%, following rises of 10% to 15% in January and May, suggesting Nvidia is protecting margins by passing higher GDDR costs downstream. Nvidia's August 26 earnings report, with guidance for fiscal second-quarter revenue of $91 billion and a 75% non-GAAP gross margin, will be a key test of whether memory inflation pressures margins, delays Rubin shipments, or constrains gaming volumes.
NVIDIA CorporationMemory supply constraints force Nvidia to cut host-memory capacity in Vera Rubin systems, reducing rack memory from 55TB to 28TB.