---
format: "aidr-story-markdown/v1"
id: "8264c497c6f8aaef176b432d15f8e17480e4d0803d2676958409eab4609690e6"
canonical_url: "https://aidr.today/8264c497?lang=en"
title: "Nvidia Swaps 12 High HBM4 for 8 High Stacks to Resolve Server Heat"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-26T09:01:21.000Z"
category: "Chips"
topics: ["nvidia","chips","infra"]
source_urls: ["https://huggingnews.com/ai/update-nvidia-swaps-12-high-hbm4-for-8-high-stacks-to-resolve-server-hea-11959010","https://x.com/jukan05/status/2092484503080206760","https://x.com/JulienTechInvst/status/2092515467860619557"]
summary: "Samsung and SK Hynix will expand production of memory components with fewer stacked DRAM dies to stabilize the supply chain. The companies are shifting the primary product configuration for Nvidia from 12 high to 8 high starting in the second half of the year. This pivot addresses thermal-management concerns resulting from HBM4 doubling its input/output channels to 2,048, which increases heat generation at the chip level and across server-rack systems. Lower yields for 12 die stacks and tight AI infrastructure demand prompted the move toward more stable 8 high procurement. Next-generation HBM4E products intended for Rubin Ultra AI accelerators are also likely to be predominantly 8 high, though final specifications for those GPUs remain unresolved."
---

# Nvidia Swaps 12 High HBM4 for 8 High Stacks to Resolve Server Heat

> [Open the canonical story](<https://aidr.today/8264c497?lang=en>)

**Published:** 2026-08-26T09:01:21.000Z
**Category:** Chips
**Topics:** nvidia, chips, infra

## Summary

Samsung and SK Hynix will expand production of memory components with fewer stacked DRAM dies to stabilize the supply chain\. The companies are shifting the primary product configuration for Nvidia from 12 high to 8 high starting in the second half of the year\. This pivot addresses thermal\-management concerns resulting from HBM4 doubling its input/output channels to 2,048, which increases heat generation at the chip level and across server\-rack systems\. Lower yields for 12 die stacks and tight AI infrastructure demand prompted the move toward more stable 8 high procurement\. Next\-generation HBM4E products intended for Rubin Ultra AI accelerators are also likely to be predominantly 8 high, though final specifications for those GPUs remain unresolved\.

## Sources

- [Story source](<https://huggingnews.com/ai/update-nvidia-swaps-12-high-hbm4-for-8-high-stacks-to-resolve-server-hea-11959010>)
- [Story source](<https://x.com/jukan05/status/2092484503080206760>)
- [Supporting source](<https://x.com/JulienTechInvst/status/2092515467860619557>)