Descriptions:
A Bloomberg Tech analyst segment examines Nvidia’s reported 15% price hikes on AI server systems, including Blackwell and the upcoming Vera Rubin platform, driven by increasing HBM (High Bandwidth Memory) costs. Raymond James infrastructure analyst Simon provides context: each Blackwell and Rubin GPU ships with eight corresponding HBM stacks at 12 layers of DRAM, and given the volume of memory required in modern AI platforms, the analyst argues 15% is not demanding relative to overall spending levels. He also notes that published list price increases do not necessarily translate to equivalent changes in contracted customer pricing.
The conversation addresses circular financing — Nvidia investing in consortiums with Blackstone, BlackRock, Apollo, and KKR that in turn purchase Nvidia products — acknowledging the optics problem while defending the logic given Nvidia’s roughly $1 billion in free cash flow every two days. The analyst suggests more aggressive share buybacks would help investor sentiment.
A pointed exchange covers the apparent contradiction between Nvidia executives promoting Vera Rubin as a cutting-edge new platform while simultaneously marketing older Ampere and Hopper generations as long-lived, cash-flow-generating assets. The analyst resolves this by describing a broadening market where leading-edge applications demand the newest accelerators while older GPUs retain useful lives for less demanding workloads. On China, he concludes Nvidia GPU sales there would be discounted by analysts due to revenue unpredictability from potential future export restrictions.
📺 Source: Bloomberg Tech · Published August 24, 2026
🏷️ Format: News Analysis
