Samsung vs. SK hynix vs. Micron: Who Leads the HBM Race in 2026?

Ask who leads the HBM (high-bandwidth memory) market in 2026, and the answer depends on what exactly is being measured. By revenue share in Q2 2026, SK hynix remained No. 1 with about 50%, Samsung climbed to about 33%, and Micron held about 18%, according to Counterpoint Research.

But market share alone does not tell the whole story. HBM4 production status, logic base-die strategy, packaging, customer relationships, capacity, and the transition to HBM4E all show different competitive strengths.

This article compares Samsung, SK hynix, and Micron across those dimensions. For the generation-by-generation specifications, see HBM3E vs HBM4 vs HBM4E: What Actually Changes for AI?. If you are new to HBM itself, start with HBM Explained: Why AI GPUs Need High-Bandwidth Memory.

Market Share in 2026: SK hynix Still Leads, Samsung Is Closing the Gap

Counterpoint Research's own quarterly series gives the cleanest like-for-like comparison. SK hynix's HBM revenue share fell from about 58% in Q1 2026 to 50% in Q2. Samsung rose from 21% to 33%, while Micron moved from 21% to 18%.

Company	Q1 2026	Q2 2026
SK hynix	58%	50%
Samsung	21%	33%
Micron	21%	18%

That matters because comparing market-share figures from different research firms can create a misleading trend. SK hynix's June 2026 U.S. Form F-1, for example, cited IDC data showing a 56.4% HBM revenue share in Q1 2026. That is a valid snapshot, but it uses a different dataset from Counterpoint and should not be directly subtracted from Counterpoint's Q2 number.

The safer conclusion is therefore based on Counterpoint's consistent series: SK hynix remained the largest HBM supplier by revenue in Q2, while Samsung made a large quarter-over-quarter gain.

HBM4 Production: All Three Are Now Shipping

The HBM4 race has moved beyond sample announcements. All three major suppliers have now reached commercial or volume shipment, although their timelines and public disclosures differ.

Samsung announced on February 12, 2026 that it had begun mass production and commercial shipment of HBM4. The product uses Samsung's 1c DRAM and a 4nm logic base die.

Micron announced in March that its 36GB 12-layer HBM4 had entered high-volume production, with volume shipment beginning in the first quarter of calendar 2026. Micron later said in its fiscal Q3 results that HBM4 was in high-volume shipments for its lead customer's platform.

SK hynix confirmed in its Q2 2026 earnings release that it had begun mass shipments of HBM4 during the second quarter and would ramp production in the second half.

Company	Current HBM4 status
Samsung	Mass production and commercial shipment since February 2026
SK hynix	Mass shipments began in Q2 2026
Micron	High-volume production / shipments since Q1 2026

Yield Numbers Need More Caution Than They Usually Get

Yield is one of the most frequently quoted HBM metrics, but public numbers are often industry estimates rather than audited company disclosures.

TrendForce News, citing Korean industry reporting, said in August that Samsung's HBM4 yield had reached roughly 80%, up from below 60% around the start of mass production. The same report cited industry estimates placing SK hynix's HBM4 yield in the 80% range.

Those numbers are useful indicators, but neither should be treated like an official audited yield disclosure from the companies themselves.

Micron has said its HBM4 ramp is operating at high volume, but no comparable public percentage was found in the company materials reviewed for this article. It is more accurate to leave that field blank than to infer a yield from production status.

Base Die: Three Different Manufacturing Strategies

HBM4 marks an important change because the base die has become a more capable logic component rather than simply a relatively basic interface layer.

Samsung uses its own 4nm Samsung Foundry logic process for HBM4, alongside its 1c DRAM. That gives Samsung an unusually integrated model: memory, logic foundry, and advanced packaging capabilities can all come from within the same company.

SK hynix takes a different route. The company has officially confirmed that starting with HBM4 it is adopting TSMC's advanced logic process for the base die. SK hynix remains responsible for the memory product while using an external leading foundry for the logic layer.

Micron uses an in-house advanced CMOS base logic die for its current HBM4. Micron's own investor materials say both the base logic die and DRAM core dies are designed and manufactured internally. For HBM4E, however, Micron plans to work with TSMC on both standard and customized logic base-die options.

Company	HBM4 base-die approach
Samsung	In-house Samsung Foundry 4nm logic
SK hynix	TSMC advanced logic process
Micron	In-house advanced CMOS logic die

Packaging: Don't Mix Current Production With Future Hybrid Bonding

Packaging is another area where headlines can become misleading because current production technology and future roadmaps are often discussed together.

Industry reporting attributes Samsung's current HBM4 production improvements to its TC-NCF bonding process. SK hynix continues to build its HBM roadmap around its Advanced MR-MUF technology, while Micron publicly describes its HBM4 as using advanced packaging without consistently naming one equivalent branded bonding process.

SK hynix also unveiled iHBM in May 2026, a thermal-management concept for next-generation HBM that embeds integrated cooling elements into the package. The company says it can reduce thermal resistance by about 30%. Importantly, iHBM should not be confused with the bonding method used in today's HBM4 production — it is a separate thermal solution aimed at future HBM designs.

Hybrid bonding belongs in the roadmap discussion rather than being described as the universal packaging method of current HBM4.

Samsung demonstrated its hybrid copper bonding technology at GTC 2026 and said it could enable HBM with 16 or more layers while reducing thermal resistance compared with conventional thermal-compression bonding. SK hynix has also discussed hybrid bonding as one option for higher-layer future HBM alongside Advanced MR-MUF.

Who's Buying HBM4? NVIDIA Gives the Clearest Answer

Customer relationships are especially difficult to compare because memory suppliers do not always identify customers publicly. In 2026, however, NVIDIA gave unusually clear confirmation.

During its GTC Taipei keynote, NVIDIA explicitly identified HBM4 memory from Micron, SK hynix, and Samsung while presenting the Vera Rubin compute board. That gives direct platform-level confirmation that all three memory companies are part of the Vera Rubin HBM4 ecosystem.

The companies also provide additional confirmation from their side. Micron says its 36GB HBM4 was designed for NVIDIA Vera Rubin. Samsung described its HBM4 at GTC 2026 as designed for the Vera Rubin platform. NVIDIA and SK hynix announced a multiyear technology partnership in June, followed in July by another agreement stating that an SK Group AI factory using Vera Rubin will be powered by SK hynix HBM4.

This is more reliable than trying to infer NVIDIA relationships solely from anonymous supply-chain reports.

AMD Is a Clear Samsung HBM4 Win

Samsung also has one customer relationship that is unusually explicit.

On March 18, 2026, Samsung and AMD announced a memorandum of understanding covering HBM4 supply for AMD's Instinct MI455X GPU. Samsung's Korean announcement describes the company as a preferred HBM4 supplier for the MI455X.

AMD's own specifications now list the MI455X with 432GB of HBM4 across 12 stacks and up to 23.3 TB/s of memory bandwidth. A 72-GPU Helios rack therefore contains about 31TB of HBM4 memory.

That makes Samsung's AMD position a confirmed competitive strength. It does not, however, prove that Samsung is the only HBM4 supplier AMD will ever use for the platform, so this article avoids making an exclusivity claim.

Google and Custom Accelerators: More Reporting, Less Disclosure

The picture is less definitive around Google's TPUs and other custom accelerators.

TrendForce News, citing Korean industry reporting, said Samsung supplied more than 60% of Google's HBM shipments through Broadcom in 2025 and was expected to remain an important supplier in 2026. The same report said SK hynix had been stronger in Google/Broadcom orders earlier in 2025.

Because those customer-allocation figures come from industry reporting rather than direct Google or Broadcom disclosures, they should be treated as reported supply-chain estimates rather than confirmed platform shares.

Capacity: Scale Matters Almost as Much as Technology

HBM demand is growing fast enough that having a qualified product is not sufficient — suppliers also need enough DRAM, TSV, packaging, and testing capacity to ship it at scale.

Industry reporting cited by TrendForce News says Samsung has been working toward roughly 250,000 HBM wafer starts per month by the end of 2026, up from roughly 170,000, although that remains a reported capacity plan rather than a company-confirmed completed figure.

SK hynix is also committing enormous capital to future memory production. In August 2026, the company approved approximately 54 trillion won for new Yongin Y2 and Cheongju M17 fabs, including 35.2 trillion won for Y2 and 19.1 trillion won for M17.

These fabs are broader DRAM and NAND investments rather than dedicated HBM4 capacity figures, so they should not be directly compared with Samsung's monthly HBM wafer estimate. They do, however, show how aggressively the memory companies are preparing for longer-term AI demand.

HBM4E: Samsung and SK hynix Have Already Shipped Samples

The next race is already underway.

Samsung announced on May 29, 2026 that it had begun shipping 12-layer HBM4E samples to major customers. SK hynix followed on June 18 with 12-layer HBM4E sample shipments of its own.

Micron's latest financial disclosures say HBM4E development is underway using 1-gamma DRAM, with volume production expected in calendar 2027. Micron has not publicly announced HBM4E sample shipment in the materials reviewed here.

Company	HBM4E status
Samsung	12-layer samples shipping since May 29, 2026
SK hynix	12-layer samples shipping since June 18, 2026
Micron	In development; volume production expected in 2027

Industry forecasts cited by TrendForce News have estimated that HBM4E could account for roughly 40% of HBM demand in 2027. That is a forecast, not an achieved market share, but it explains why sample timing and customer qualification already matter.

So, Who Is Actually Ahead?

There is no single answer across every metric.

SK hynix remains the revenue-share leader, with 50% of the HBM market in Counterpoint's Q2 2026 data. It also has a deep and now publicly confirmed technology relationship with NVIDIA.

Samsung made the biggest Q2 market-share gain, moving from 21% to 33% in Counterpoint's series. It was first to publicly announce commercial HBM4 shipment in February, uses its own 4nm foundry base die, has already shipped HBM4E samples, and has an explicit HBM4 collaboration with AMD for MI455X.

Micron is smaller by HBM revenue share but is already in high-volume HBM4 production, uses an internally designed and manufactured CMOS base die for HBM4, and is directly confirmed as one of the HBM4 suppliers in NVIDIA's Vera Rubin ecosystem.

The most useful way to follow the HBM race through the rest of 2026 is therefore not to ask simply "Who is No. 1?" but to ask which company leads which metric, for which product generation, and in which quarter.

The answer is moving too quickly for a single ranking to capture the whole market.

For the technical differences between the generations, see HBM3E vs HBM4 vs HBM4E: What Actually Changes for AI?.

For the broader reason HBM has become so important to AI systems, see The AI Memory Wall Explained.