HBM4 ramp accelerating with Nvidia Vera Rubin; HBM4E debut 2027; HBM TAM $100B+ by 2027; Anthropic partnership
Key Questions
What is the expected timeline for HBM4 volume ramp and HBM4E debut?
HBM4 volume ramp is accelerating twice as fast as prior generations and is sold out through 2028, with Micron confirmed as a supplier for Nvidia's Vera Rubin platform. HBM4E is on track for debut in 2027, supported by Hiroshima fab expansion including 1γ/1δ nodes.
What is the projected HBM TAM by 2027 and key growth drivers?
Micron pegs the HBM TAM at $100B+ by 2027, driven by HBM now representing 63% of AI chip component spend and a 40% CAGR through 2030 versus 21% for standard DRAM. SK Hynix plans to double wafer capacity to meet demand.
How is Micron involved with Nvidia's Vera Rubin and Hiroshima production?
Micron is confirmed as an HBM4 supplier for Nvidia Vera Rubin, with the Hiroshima fab expansion set to produce HBM4 and equipment installation targeted for 2H28. Applied Materials tools are enabling HBM scaling at the site.
What is the nature of the Micron-Anthropic partnership?
Micron and Anthropic have formed a partnership to optimize HBM, DRAM, and SSD solutions specifically for AI workloads, including Claude models. This collaboration aligns hardware design with AI memory requirements.
How does Gavin Baker's thesis relate to the HBM outlook?
Gavin Baker's memory crunch thesis highlights HBM as a key bottleneck in AI hardware, reinforcing the structural demand and capacity constraints described in the highlight. It supports the view of sustained tightness through the decade.
HBM4 volume ramp twice as fast; sold out through 2028. Micron confirmed as HBM4 supplier for Nvidia Vera Rubin. HBM4E on track for 2027, with Hiroshima expansion including 1γ/1δ nodes and HBM4E production. HBM now 63% of AI chip component spend. SK Hynix plans to double wafer capacity. HBM CAGR 40% through 2030 vs 21% standard DRAM. Micron explicitly pegs HBM TAM at $100B+ by 2027. Applied Materials new tools enable HBM scaling. Anthropic partnership to optimize HBM, DRAM, and SSD for AI. Hiroshima fab will produce HBM4 for NVIDIA Vera Rubin, equipment 2H28. Gavin Baker's memory crunch thesis reinforces HBM as key bottleneck.