Corpus Agentis
The field book to agent ecosystems
The field book to agent ecosystems
Compute · HBM

HBM3e supplier share

High-bandwidth memory is the deepest bottleneck under the agent stack, share by supplier.

SK Hynix~62%
Micron~21%
Samsung~17%
HBM suppliers in the network · in the supply-chain network
SK Hynix (HBM) · ~62% HBM3EMicron (HBM) · ~21% HBMSamsung (HBM) · HBM3E in AMD MI350; cleared NVIDIA qualification Sept 2025
Common questions
What is HBM and why does it matter for AI?

HBM is high-bandwidth memory: DRAM chips stacked vertically and placed right beside the processor, connected by a very wide interface. It matters because large AI models are limited by how fast their weights can be moved, not by raw maths. Without HBM the compute units sit idle waiting for data.

Who makes HBM?

Three companies make it in volume: SK hynix, Samsung and Micron. That short list is why memory supply, rather than chip design, often decides how many AI accelerators actually reach the market.

What is HBM3e?

HBM3e is the current high-volume generation of high-bandwidth memory, with more bandwidth and capacity per stack than HBM3 before it. It is the memory paired with today's leading AI accelerators.

Why is there an HBM shortage?

Every accelerator vendor is competing for output from the same three suppliers. Yields are difficult, capacity takes years to add, and the advanced packaging that bonds the memory to the processor is separately constrained. Supply is typically committed well ahead of production.

Next in the learning path