The HBM4 qualification race is not a uniform technology transition — it is a supplier-by-supplier, customer-by-customer negotiation playing out across NVIDIA, AMD, Google, and custom ASIC programs simultaneously. Who is qualified where, at what volume, and at what price determines the memory line item in the most expensive compute hardware ever mass-produced. This article maps the qualification matrix as of mid-2026, quantifies the BOM impact, and projects share trajectories for the remainder of the year.
The Qualification Matrix: Who Is Qualified Where
HBM4 represents a meaningful generational leap: interface width doubles to 2,048 bits (from 1,024 bits in HBM3E), per-stack bandwidth reaches 1.5–2 TB/s versus 896–1,280 GB/s for HBM3E, and maximum stack capacity rises to 64 GB from 48 GB. SK Hynix claims a 40% power-efficiency improvement at 10 Gbps data rate; Micron's samples run up to 11 Gbps [1]. These specs matter because qualification is not binary — customers test to specific bandwidth and yield targets, and a 1 Gbps edge can be the difference between primary and secondary supplier status.
The table below summarizes publicly reported qualification and production status as of mid-2026:
| Supplier | HBM4 Status (mid-2026) | Primary Customers | HBM Revenue Share, Q1 → Q2 2026 (Counterpoint) | HBM4 Segment Share Outlook |
|---|---|---|---|---|
| SK Hynix | Mass production; development completed and production readied Sept 2025 [10] | NVIDIA (primary), AMD, Google | 58% → 50% [8] | ~70% (NVIDIA Rubin, UBS est.) [5] |
| Samsung | Mass production and commercial shipments announced Feb 12, 2026, at 11.7 Gbps [9] | NVIDIA (secondary), AMD, custom AI | 21% → 33% [8] | ~20–25% (recovering) [1] |
| Micron | Qualification progressing; 11 Gbps samples demonstrated | NVIDIA (~20% allocation), others | 21% → 18% [8] | ~10–15% (stable to slight upside) [1] |
Two suppliers say they are in HBM4 mass production. SK Hynix completed HBM4 development and readied mass production in September 2025 [10]; Samsung announced mass production and commercial shipments on February 12, 2026, at 11.7 Gbps [9]. Volume still favours SK Hynix, but the gap narrowed sharply between Q1 and Q2, and Counterpoint expects HBM4 shipments to gain significant traction in the second half [8]. Micron's position is the most asymmetric: its 11 Gbps samples edge SK Hynix's 10 Gbps claim [1], which could accelerate qualification at customers whose system architecture is bandwidth-constrained rather than capacity-constrained.
Correction, October 4, 2026: This article said SK Hynix was the only supplier confirmed in HBM4 mass production and put Samsung's Q1 2026 HBM share at 25–30%. Samsung announced HBM4 mass production and commercial shipments on February 12, 2026, and Counterpoint puts its Q1 2026 HBM revenue share at 21% (33% in Q2). The share table, charts and risk discussion now use Counterpoint's figures.
Samsung's HBM share rose from 21% to 33% between Q1 and Q2 2026 as SK Hynix fell from 58% to 50%
Source: Counterpoint Research, via EDAILY (Sep 3, 2026) [8]
2026 Share Trajectory: The Diverging Paths
The quarter-by-quarter picture is important because HBM4 shipments are expected to ramp materially in H2 2026, with H1 still dominated by HBM3E revenue [8]. That sequencing means the overall 2026 share numbers will blend two different competitive dynamics: HBM3E (where SK Hynix's lead is most entrenched) and HBM4 (where Samsung and Micron have more room to close ground).
For the full year, the publicly available estimates bracket as follows [1][3][5]:
| Supplier | HBM Overall Share Range (2026) | HBM4-Specific Share (NVIDIA Rubin, est.) |
|---|---|---|
| SK Hynix | 50–62% | ~70% |
| Samsung | 25–40% | ~20–25% |
| Micron | 5–20% | ~10–15% |
Counterpoint's actuals so far (SK Hynix 58% → 50%, Samsung 21% → 33%, Micron 21% → 18% from Q1 to Q2) put Samsung near the top of its bracket and Micron above most of its range [8].
The wide bands reflect genuine uncertainty around Samsung's yield ramp cadence and Micron's qualification timeline at NVIDIA. UBS's 70% HBM4 share projection for SK Hynix on the Rubin platform [5] is the most specific public forecast and is directionally consistent with SK Hynix's current production lead. Barclays' view that SK Hynix retains >50% HBM share for several years [6] implies the gap does not close rapidly even as Samsung and Micron scale.
Samsung's capex signal is worth noting — we covered the spending race in depth in SK Hynix vs Samsung: HBM4 and the 2026 Capex Race — the company is targeting approximately 50% growth in HBM production capacity by end-2026 [2][3]. That is a large absolute increment given Samsung's existing scale, but capacity growth and qualification status are different constraints — Samsung can build the supply before it finishes winning the socket.
SK Hynix's HBM4 share premium over its overall HBM share signals qualification is outpacing capacity
Source: Counterpoint via EDAILY [8], UBS via multiple sources [5], Presenc AI [1]
BOM Impact: What Supplier Mix Does to Accelerator Economics
HBM is not a peripheral cost line in AI accelerator manufacturing — it is the single largest or second-largest cost component depending on SKU. Using Silicon Analysts' canonical BOM estimates (tracked live in our HBM Market Analysis tool):
| Accelerator | HBM Generation | HBM Capacity | Est. HBM Cost | HBM as % of Mfg Cost |
|---|---|---|---|---|
| NVIDIA H200 SXM5 | HBM3e | 141 GB | ~$2,400 | ~47% |
| NVIDIA B100 / B200 | HBM3e | 192 GB | ~$3,250 | ~47–48% |
| NVIDIA GB200 Superchip | HBM3e | 384 GB | ~$6,500 | ~46% |
| AMD MI300X | HBM3 | 192 GB | ~$2,900 | ~55% |
| AMD MI325X | HBM3e | 256 GB | ~$4,350 | ~64% |
HBM4's combination of higher per-stack capacity (64 GB vs 48 GB) and higher bandwidth (1.5–2 TB/s) will likely allow next-generation accelerators to achieve more capability per stack, but the per-GB and per-bit pricing dynamics are the key variable. SK Hynix reportedly set HBM4 supply pricing roughly 70% above HBM3E in early NVIDIA negotiations [7]. If that premium holds through H1 2026 — plausible given SK Hynix's lead in HBM4 supply — the HBM line item in Rubin-class accelerator BOMs could expand materially relative to the Blackwell-generation figures above.
The competitive relief valve is the H2 2026 multi-supplier scenario. If Samsung's HBM4 volume keeps growing and Micron progresses through customer gates before year-end, procurement teams will have negotiating leverage they lack today. BofA estimates the total 2026 HBM market at $54.6 billion, a 58% increase year-over-year [5]; Goldman Sachs projects HBM demand for custom ASIC AI chips growing 82%, accounting for one-third of total HBM demand [5]. That demand curve provides pricing support for all three suppliers even if share shifts — the market is expanding fast enough that Samsung and Micron can grow revenue while SK Hynix defends its share premium.
Procurement and Platform Risk Implications
For platform architects and procurement teams building HBM4 sourcing strategies, three structural risks stand out from this qualification map:
Concentration on HBM4 in H1 2026. Samsung announced HBM4 mass production in February [9], but SK Hynix still supplies most HBM, so any yield disruption, geopolitical event, or allocation decision by SK Hynix flows through to accelerator build plans. Samsung's rise from 21% to 33% of HBM revenue between Q1 and Q2 [8] is the first sign of a second volume source.
Samsung's recovery shows in share, not yet in disclosed yields. Samsung announced HBM4 mass production and commercial shipments in February [9] and gained 12 points of HBM revenue share in Q2 [8], but it has not disclosed HBM4 yields, and the step from qualification to volume is where it stumbled on HBM3E. Procurement teams should weight its H2 2026 ramp with that history in mind.
Micron's specification edge is real but commercially unproven at scale. An 11 Gbps sample is a meaningful data point, but customers require qualification across full operating temperature ranges, stacked configurations, and interposer integration — all of which take time. Micron's share upside is real; its timeline is the uncertainty.
The $54.6 billion HBM market projected for 2026 [5] means these qualification gaps have direct dollar consequences for both suppliers missing revenue and customers paying premium ASPs. The qualification race is, in practical terms, a pricing race.
References & Sources
[1] Ramanath, Presenc AI. "HBM Market Share 2026: SK hynix vs Samsung vs Micron." May 2026. https://presenc.ai/research/hbm-market-share-samsung-skhynix-micron-2026
[2] Silicon Analysts. "SK Hynix vs Samsung: HBM4 and the 2026 Capex Race." siliconanalysts.com
[3] Korea JoongAng Daily. "Samsung, SK hynix dramatically scaling up HBM production to meet rising demand, recent Nvidia deals." Lee Jae-lim, January 4, 2026. https://www.koreajoongangdaily.com/business/samsung-sk-hynix-dramatically-scaling-up-hbm-production-to-meet-rising-demand-recent-nvidia-deals/12391367
[4] Counterpoint Research. "Global DRAM and HBM Market Share: Quarterly." https://counterpointresearch.com/en/insights/global-dram-and-hbm-market-share
[5] Multiple sell-side sources (UBS, BofA, Goldman Sachs). "2026 Market Outlook — Focus on the HBM-Led Memory Supercycle." 2026.
[6] Barclays equity research (analyst coverage), via Yahoo Finance. "Barclays says SK Hynix stock could more than double on 'significant growth'." 2026.
[7] TweakTown, reporting Korean media. "SK hynix 'drastically' raises next-gen HBM4 supply price in a 'war of nerves' with NVIDIA." 2026.
[8] EDAILY, reporting Counterpoint Research. "SamsungElectronics Expands HBM Market Share to 33%… SK Hynix Remains No. 1 with 50%." Soyeon Kim, September 3, 2026. https://en.edaily.co.kr/news/eda202609035608/
[9] Samsung Electronics. "Samsung Ships Industry-First Commercial HBM4 With Ultimate Performance for AI Computing." February 12, 2026. https://news.samsung.com/global/samsung-ships-industry-first-commercial-hbm4
[10] SK hynix. "SK hynix Completes World's First HBM4 Development and Readies Mass Production." September 12, 2025. https://news.skhynix.com/en/sk-hynix-completes-worlds-first-hbm4-development-and-readies-mass-production/