HomeAIAI Chips & ComputeCounterpoint: Most Nvidia Rubin Products Likely to Use 12-Layer HBM4, Not the...

Counterpoint: Most Nvidia Rubin Products Likely to Use 12-Layer HBM4, Not the Rumoured Eight

Counterpoint Research says most of Nvidia’s Rubin-based products will likely use 12-layer HBM4 memory, not an eight-layer version that had been rumoured, the Korea Herald reported on 9 October 2026. HBM, high-bandwidth memory, is a tower of memory chips stacked beside an AI processor; the “layers” are those chips. TSN explains HBM here.

What Counterpoint said

Reported (Korea Herald and Businesskorea), Counterpoint’s own assessment. In the Korea Herald’s words, Counterpoint “said Friday that most Rubin-based products will likely use 12-layer HBM4, the sixth generation of high-bandwidth memory, rather than the eight-layer configuration that had been rumored in the industry.” [1] Businesskorea quotes Counterpoint: “The concern about slowing memory demand is a misunderstanding.” [2] The dates differ: the Korea Herald says Friday, 9 October; Businesskorea says 8 October.

The Korea Herald credits the change to Counterpoint itself: its latest assessment “marks a shift from its August report, which had pointed to Nvidia adopting an eight-layer HBM4 variant for Rubin.” [1]

Why the rumour mattered

The rumour was that Nvidia had asked suppliers for fewer stacked chips, “raising concerns that lower HBM requirements could free up production capacity for conventional DRAM and ease pressure on memory prices.” [1] Businesskorea’s paraphrase of Counterpoint is that HBM “requires more than three times the production capacity of commodity DRAM” [2]. TSN covered Samsung’s record quarter.

Company claims, as the Korea Herald reports them

SK hynix “unveiled its 48-gigabyte, 16-layer HBM4 at CES in Las Vegas in January”. Samsung “has been shipping commercial HBM4 products since February”. A September Silicon Analysts report “suggested that SK hynix had begun volume shipments of 16-layer HBM4 for Rubin, but the company has not publicly confirmed those deliveries.” [1]

What this does not show

  • A Nvidia decision. TSN found no Nvidia comment on layer counts in either report.
  • A certainty. Counterpoint’s view is a forecast: “most” and “likely”.
  • Counterpoint’s own words. TSN did not find its release, so it relies on these two press accounts.

The Bottom Line

Counterpoint now expects most Rubin products to use 12-layer HBM4, a change from its August view, according to two Korean outlets. It is a forecast.

Related on TSN: What Is Custom HBM? Why Nvidia Wants to Move the Memory Controller Off the Chip; Samsung’s Record Quarter and AWS’s Linux Foundation Seat: Two Signs of How AI Is Reshaping Big Tech

Sources

  1. Jie Ye-eun, “SK hynix, Samsung push 16-layer HBM4 as Rubin favors 12”, The Korea Herald, 9 October 2026, 11:51 KST (newspaper; relays Counterpoint’s assessment and company facts). https://www.koreaherald.com/article/10897751
  2. Shin So-hyeon, “HBM4 Demand Holds Firm As Rubin Keeps 12 Layers”, Businesskorea, 8 October 2026, 19:11 KST (trade press; relays Counterpoint’s assessment). https://www.businesskorea.co.kr/news/articleView.html?idxno=278536

Share this story

More in this category

Latest on TSN

Free TSN tools: crypto calculator, Flux dashboard and more.