AI Chips

HBM Market Share by Supplier

HBM (the stacked-DRAM chips paired with every AI accelerator) market share by supplier. SK Hynix dominates HBM3E with ~60% share; Samsung trails on yield, Micron is the credible #3.
Updated at 2026-05-06

HBM Industry Revenue Growth

Total quarterly revenue across all HBM (High-Bandwidth Memory) suppliers. Up roughly 20x since early 2022 because every AI accelerator now ships with 80–192 GB of stacked DRAM, making HBM the highest-margin memory product in the industry.
Updated at 2026-05-06

HBM Generation Comparison

How each HBM (High-Bandwidth Memory) generation compares on bandwidth, capacity per stack, layer count, and which AI chips use it. HBM3E is the volume product today (NVIDIA H200, B100); HBM4 begins shipping in late 2025 with roughly 2x the bandwidth.
GenerationYearBandwidthCapacity/StackLayersTSV PitchUsed InStatus
HBM22018256 GB/s8 GB/stack8-Hi~40 μm
NVIDIA V100AMD MI60
Legacy
HBM2E2020460 GB/s16 GB/stack8-Hi~36 μm
NVIDIA A100Intel Gaudi 2
Mature
HBM32022665 GB/s24 GB/stack12-Hi~30 μm
NVIDIA H100AMD MI300X
Mass Production
HBM3E20241.18 TB/s36 GB/stack12-Hi~26 μm
NVIDIA B200AMD MI325X
Mass Production
HBM42025–20262.0 TB/s48 GB/stack16-Hi~20 μm
NVIDIA Rubin (expected)AMD MI400 (expected)
Sampling
HBM4E2026–20273.25 TB/s64 GB/stack16-Hi~18 μm
Next-gen AI accelerators
Development
Updated at 2026-03-18

HBM Bit Shipments by Supplier (Quarterly)

Quarterly HBM bit shipments across SK Hynix, Samsung, Micron in petabits. SK Hynix dominates from NVDA HBM3E first-mover qualification; Samsung NVDA HBM3E qualification partial; Micron ramping HBM3E + HBM4.
Updated at 2026-05-31

HBM3E + HBM4 Qualification Status by Customer

Qualification matrix: rows are HBM suppliers, columns are customer platforms (NVIDIA, AMD, Google TPU). SK Hynix's first-mover NVIDIA HBM3E qualification is the marquee milestone.
SupplierNVDA H200 (HBM3E)NVDA B100/B200 (HBM3E)NVDA Rubin (HBM4)AMD MI300/325 (HBM3E)AMD MI350/400 (HBM4)Google TPU (HBM3E + HBM4)
SK HynixQualified (sole supplier 2024-H1; volume ramp Q2-Q3 2024)Qualified (volume ramp Q4 2024 onward)Sampling + qualifying (Q1-Q2 2026); targeting volume H2 2026Qualified (volume ramp Q3 2024 onward)Sampling + qualifying; targeting volume late 2026Qualified (TPU v5e + v6e); HBM4 sampling for v7
SamsungPartial qualification Q4 2024 (limited volumes); full status disputed in trade pressPartial qualification; NVDA Q1 2026 reportedly accepted some HBM3E volumesSampling (Q1 2026); qualification timing uncertainQualified (alternate source to SK Hynix for MI300X + MI325X)Sampling + qualifying; competitive with SK HynixQualified for TPU v6e + v7 (Google Samsung HBM3E supplier)
MicronQualified Q3 2024 (NVDA validated; production ramp from late 2024)Qualified Q4 2024 (production ramp continuing through 2025)Sampling Q1 2026; volume target H2 2026Not engaged at material volume (SK Hynix + Samsung sources)Sampling; emerging as third sourceLimited engagement; not a primary Google supplier
Editorial. Why this mattersQualification is the gating event for HBM revenue. SK Hynix first-mover NVDA qualification is the marquee 2024 industry milestone. The Samsung HBM3E NVDA qualification timing is the most-watched cross-supplier comparison in memory. Micron's catch-up qualification is the marquee 2024-2025 turnaround story.
Updated at 2026-05-28

Ask Sterling

Register for a premium account to gain access to Sterling AI.

Get Started

Things you can ask Sterling:

Summarize Tesla's latest earnings reportWhy did NVIDIA's margins expand?Compare Apple vs Microsoft's cash flowWhat's driving the EV sector growth?
AI Chips - Memory | Sterling