Why Korea Dominates Memory Semiconductors: Samsung, SK Hynix and the HBM Era

Almost every headline about the AI boom eventually points back to two companies most people outside Korea couldn't name a decade ago. Samsung Electronics and SK Hynix don't design the chatbots or the AI accelerators everyone talks about. But without the memory chips they make, none of those accelerators would work.

That's not an exaggeration. As of the second quarter of 2026, these two Korean companies together control roughly 64% of the world's DRAM market and close to half of the NAND flash market. In high-bandwidth memory (HBM) — the specialized chip that feeds data to AI processors like Nvidia's GPUs — the picture is even more lopsided, with Korean firms holding the overwhelming majority of the market between them.

This piece walks through how dominant Korea's position actually is right now, how the Samsung–SK Hynix rivalry inside HBM is playing out, how Korea got here in the first place, and what's coming next.

So How Dominant Is Korea in Memory Chips Right Now?

Start with DRAM, the standard memory chip used in nearly every computer, phone, and server. According to TrendForce, global DRAM revenue hit $154.73 billion in the second quarter of 2026, up 59.5% from the previous quarter — a jump driven almost entirely by AI server demand.

Samsung Electronics led with 39.4% of that revenue ($60.98 billion), up from 38.5% the quarter before. SK Hynix held 24.9% ($38.59 billion), down from 28.8%. Micron, the only non-Korean company in the top three, came in at 23.3% ($36.0 billion). TrendForce attributed Samsung's gain largely to early HBM4 mass production and the largest shipment growth of the three companies that quarter.

Combined, Samsung and SK Hynix controlled about 64% of the global DRAM market in Q2 2026. For context, one Q1 2026 report put the split closer to Samsung 38–38.6%, SK Hynix 28.8–29%, and Micron around 22.4% — the exact numbers move quarter to quarter, but the two-Korean-companies-on-top pattern doesn't.

NAND flash tells a similar story. Counterpoint Research put Samsung at 28% and SK Hynix at 19% of the Q2 2026 NAND market, a combined 47%, with Micron close behind at 15% — its highest share of the quarter. The overall NAND market grew about 70% quarter over quarter, and while Samsung and SK Hynix still lead, Micron is narrowing the gap.

Who's Actually Winning the HBM Race?

This is the part of the memory market that matters most for AI right now. HBM stacks multiple DRAM dies vertically and connects them with ultra-fast interconnects, letting AI accelerators move data in and out far faster than standard memory allows. Every major AI chip — Nvidia's GPUs included — depends on it.

According to Counterpoint Research, SK Hynix held 50% of the HBM market by revenue in Q2 2026, down from 64% a year earlier. Samsung came in at 33%, a sharp jump from just 21% in Q1 2026. Micron held 18%, down 3 percentage points from the prior quarter. Samsung reportedly crossed $1 billion in HBM revenue within four months of starting HBM4 mass production — a fast ramp by any standard.

It's worth flagging that not everyone agrees on the exact numbers here. Some other market trackers — reportedly citing TrendForce data through outlets like Chosun Biz — put SK Hynix's share as high as 58–62% for the same quarter, with Micron at 18–21% (ahead of Samsung) and Samsung at just 17–21%. We weren't able to verify the original TrendForce report behind those figures directly, and the sites republishing them carry less verification than Counterpoint's own published data. So treat those alternate numbers as lower-confidence estimates. What every source agrees on is the direction: SK Hynix leads HBM outright, and Samsung is closing the gap fast on the strength of its HBM4 ramp.

Where Do HBM3E, HBM4, and HBM4E Stand?

HBM3E is still the workhorse — it accounted for roughly 66% of total HBM output in 2026, driven by continued demand for Nvidia's H200 and other AI accelerators. Reportedly, both Samsung and SK Hynix raised HBM3E prices by around 20% for 2026 given how tight supply remains.

SK Hynix officially confirmed the start of HBM4 mass production in its Q2 2026 earnings release, saying it met customer-required operating speeds while achieving competitive power efficiency and cost. The company also laid out plans to accelerate HBM4 production and begin shipping HBM4E samples in the second half of 2026.

Samsung, for its part, was reported to begin official HBM4 shipments to Nvidia and AMD starting in February 2026, after passing both companies' final qualification testing. Samsung's HBM4, built on its sixth-generation 10-nanometer-class DRAM process, reportedly runs at 11.7 Gbps with headroom up to 13 Gbps. In March 2026, at Nvidia's GTC conference, Samsung unveiled HBM4E and said it had shipped samples to major customers — a claim it described as an industry first. Samsung repeated that "HBM4E samples shipped first in the industry" language in its own Q2 2026 earnings release alongside news of expanding HBM4 sales.

What About Nvidia's Next-Gen Chips?

Nvidia's upcoming Vera Rubin platform, expected in the second half of 2026, is expected to use eight HBM4 stacks per processor. TrendForce reported — explicitly framing it as an industry estimate rather than a confirmed allocation — that SK Hynix is expected to supply roughly two-thirds of Nvidia's HBM4 volume, possibly as much as 70%, with Samsung competing on the strength of earlier shipment timing. Neither company has publicly confirmed a specific allocation split, so this figure should be read as informed industry speculation, not an official number.

Separately, SK Hynix said in its October 2025 earnings call that it had already sold out its entire 2026 chip production slate — a strong, if indirect, signal of AI-driven demand.

How Did Korea Get Here?

None of this happened overnight. Korea's semiconductor push traces back to 1968, when the government passed its first Electronics Industry Promotion Law, and accelerated in the early 1980s with a dedicated semiconductor development plan.

Samsung entered the chip business in 1974 by acquiring Korea Semiconductor. The real turning point came in 1983, when founder Lee Byung-chul made the decision to enter the memory chip market, licensing 64K DRAM design technology from Micron Technology in the US for $3 million. By the end of that year, Samsung had successfully developed its own 64K DRAM, making Korea only the third country in the world to do so.

Less than a decade later, in 1992, Samsung launched the world's first 64M DRAM chip, catapulting it to the top of the global memory chip industry — a position it has never really relinquished. In 1996, the company developed the industry's first 1Gb DRAM and began mass production of 64Mb DRAM. Korea's share of the global DRAM market rose from under 5% in the 1980s to more than 30% by the mid-1990s.

Academic research, including a UC Berkeley working paper on Samsung's technological capabilities, points to both formal government support and less visible mechanisms — like government pressure on banks to extend credit — as part of what fueled this rise.

Korean chipmakers also became known for a distinct strategy: investing aggressively in new capacity even during industry downturns, a pattern often described as a "chicken game" that squeezed Japanese, Taiwanese, and American rivals out of the market over time.

That government-industry partnership hasn't stopped. Korea recently signed power and water supply agreements with Samsung and SK Hynix for its planned Yongin semiconductor cluster, introduced a first-of-its-kind production-linked tax credit for six strategic industries including semiconductors in its 2026 tax reform, and reportedly earmarked more than 14 trillion won in policy financing for the sector, including a 4.25 trillion won low-interest loan program through the Korea Development Bank. A semiconductor special law aimed at direct subsidies and infrastructure support has also reportedly cleared a National Assembly standing committee.

Samsung vs SK Hynix: Whose Q2 2026 Was Better?

Both companies posted record numbers in the second quarter of 2026, though they measure success somewhat differently.

Metric	Samsung Electronics DS Division	SK Hynix
Revenue	127.5 trillion won (+56% QoQ)	79.32 trillion won (+257% YoY)
Operating profit	89.2 trillion won	60.54 trillion won (+557% YoY)
Operating margin	~70.0%	76% (+5pp QoQ)
Net income	Not separately disclosed for DS division	93.92 trillion won (includes 63.3 trillion won in investment valuation gains)
Notable detail	Record quarterly memory revenue and profit; server-related revenue share at an all-time high	First-half revenue topped 100 trillion won for the first time; long-term supply agreements signed with roughly 10 major customers

(Source: Samsung Electronics Q2 2026 earnings release; SK Hynix Q2 2026 earnings release)

The two numbers aren't perfectly apples-to-apples. Samsung's figures cover its entire Device Solutions division, which includes foundry and system LSI business alongside memory, while SK Hynix's numbers are almost entirely memory. On a memory-only basis, some reporting suggests SK Hynix's operating margin (76%) actually runs higher than Samsung's DS-wide margin (~70%) — though that particular comparison is based on a report from before Q2 results came out, so the timing isn't a perfect match.

Strategically, Samsung has publicly emphasized high-value products — HBM4/HBM4E, DDR5, and server modules like SOCAMM2 — as it targets server infrastructure and "agentic AI" demand in the second half of the year. SK Hynix has emphasized accelerating HBM4 production, shipping HBM4E samples, and locking in multi-year supply contracts with customers.

How Much Are They Investing?

The investment race between the two companies — and the Korean government behind them — has been striking in 2026.

In March, SK Hynix's board approved an additional roughly 21.6 trillion won (about $15 billion) for fab expansion, including its Cheongju M15X line. In August, the company went further, announcing 54 trillion won (about $38 billion) to build two new plants: Yongin "Y2" and Cheongju "M17."

At the national level, Korea's government announced an 800 trillion won (roughly $520 billion) public-private investment plan in June 2026, with President Lee Jae-myung directly involved alongside Samsung and SK Hynix. Four new fabs — two each from Samsung and SK Hynix — sit at the core of that plan.

What About Micron and China's Chipmakers?

Korea's lead isn't unchallenged. Micron, the largest non-Korean player, held 23.3% of the DRAM market in Q2 2026 — closing in on SK Hynix's 24.9% — and 18% of the HBM market, good for third place. Some industry watchers expect Micron to receive the third-largest allocation of Nvidia's Vera Rubin HBM4 volume, though again, this isn't an official, confirmed figure. (A handful of lower-confidence secondary sources have claimed Micron briefly overtook Samsung in specific HBM allocations; those claims conflict with Counterpoint's own Q2 2026 figures showing Samsung well ahead of Micron, so we're treating them as unverified.)

China's chipmakers are the more closely watched long-term threat. CXMT is reportedly targeting HBM3 mass production in 2026, supplying samples to Chinese AI chip designers including those linked to Huawei, with an eye toward reaching HBM3E-level technology by 2027. Some analysts are more cautious, suggesting mass production at competitive yields is more realistically a 2028-or-later story. CXMT is also said to be building a new DRAM fab in Shanghai up to three times the size of its Hefei headquarters, with equipment installation starting in 2026 and full production targeted for the first half of 2027. The US Department of Defense has designated CXMT a Chinese military company.

YMTC, meanwhile, has reportedly grown to roughly 10–13% of the global NAND market by shipment volume and is said to be preparing, in cooperation with CXMT, to enter the DRAM and HBM markets aimed at AI accelerators. YMTC is on the US Commerce Department's Bureau of Industry and Security Entity List.

The broader backdrop here is US export controls. In December 2024, Washington introduced a new control package specifically targeting China's access to HBM and related chipmaking equipment, adding more than 140 Chinese semiconductor firms to the Entity List. Analysts covering the topic believe CXMT has secured enough equipment to sustain HBM-related production through 2026–2027, but is likely to hit real obstacles scaling further or developing next-generation HBM technology beyond that point.

Where Is the Memory Market Headed Next?

TrendForce's July 2026 outlook expects DRAM and NAND to diverge in 2027. DRAM supply is expected to stay tight, driven by continued HBM capacity allocation and strong AI server demand — because HBM production requires substantially more wafer input than standard DRAM for the same bit output. TrendForce projects HBM's share of total DRAM wafer input rising from about 18% at the end of 2025 to 22% in 2026 and 30% in 2027, which by itself constrains how much total DRAM bit supply can grow.

There's also industry chatter, per TrendForce-linked reporting, about HBM suppliers pushing through steep 2027 price increases — with some estimates floating figures as high as a doubling of prices, though that remains speculative rather than confirmed.

NAND flash is expected to move in the opposite direction: TrendForce forecasts supply conditions easing in the second half of 2027 as new capacity comes online and consumer electronics demand stays soft, putting downward pressure on prices.

Server shipments grew 17% year-over-year in 2026, and TrendForce expects even stronger growth in 2027, driven by continued capital spending from North American cloud providers, new CPU platforms from Intel and AMD, and the commercialization of agentic AI. SK Hynix's own market outlook, published through its newsroom, describes an HBM-driven memory "supercycle" that it expects to continue, with HBM4 shortages deepening further in the second half of 2026.

What Does This Rivalry Actually Tell Us?

Step back from the quarterly numbers and a larger pattern emerges. The AI boom didn't just create demand for memory chips — it restructured what kind of business memory chip-making is. For decades, DRAM and NAND were commodity products defined by brutal boom-bust cycles and thin margins during the downturns. HBM has turned a slice of that business into something closer to a specialized, relationship-driven, high-margin product, with customers signing multi-year supply deals years in advance.

That shift is a big part of why Samsung and SK Hynix can justify investment numbers that would have looked reckless a decade ago — tens of billions of dollars in new fabs, committed years before the chips they produce even ship. It only makes sense if you believe, as both companies clearly do, that AI-driven demand for high-bandwidth memory isn't a temporary spike.

It's also worth sitting with how concentrated this part of the AI supply chain actually is. Two companies, both headquartered in the same country, currently produce the overwhelming majority of the memory that every major AI accelerator on the planet depends on. That's part of why the US export control regime around China's HBM ambitions has been so aggressive, and why Korea's government has leaned so heavily into supporting both companies' expansion plans — this isn't just a corporate rivalry, it's treated as a matter of national strategic advantage.

Whether Samsung's rapid HBM share gains continue to narrow SK Hynix's lead, or whether SK Hynix's early mover position and Nvidia relationship hold, is genuinely an open question. What's less in doubt is that for the next several years, the pace of the global AI buildout will be shaped, in no small part, by decisions made inside two corporate headquarters in Korea.