SK hynix HBM Strategy Wins in AI Memory, but Its DRAM Share Is Sliding

SK hynix has strengthened its HBM lead despite losing 13 percentage points of global DRAM revenue share within a year. The SK hynix HBM strategy made the company a central supplier for AI accelerators. It also left the chipmaker less exposed to a sudden boom in conventional memory prices.

That tension matters more than a routine change in quarterly rankings. SK hynix reported record results while Samsung Electronics widened its DRAM lead and Micron nearly caught second place. The same commitments that secured SK hynix’s early advantage in AI memory limited its ability to capture a broader pricing surge.

The company is now accelerating production of newer 1c DRAM, its sixth-generation 10-nanometer-class process. That transition supports both conventional server memory and future HBM4E products. It is a test of whether SK hynix can defend its AI franchise without surrendering too much of the wider memory market.

Record Revenue Came With a Smaller DRAM Position

SK hynix grew rapidly during the second quarter, but Samsung and Micron grew even faster.

Counterpoint Research estimated that the SK hynix DRAM share fell from 39% in the second quarter of 2025 to 26% one year later. Samsung reached 39%, while Micron climbed to 25%. That left SK hynix only one percentage point ahead of Micron in the revenue ranking.

The decline did not represent collapsing sales. Counterpoint said SK hynix’s quarterly DRAM revenue increased 214% from the previous year. The company still lost share because competitors captured more of an unusually strong market expansion.

Separate DRAM revenue data from TrendForce showed the same competitive pattern. The research firm estimated that industry revenue increased 59.5% from the preceding quarter, reaching nearly $154.73 billion.

TrendForce placed Samsung first with 39.4% of second-quarter revenue. SK hynix followed with 24.9%, and Micron reached 23.3%. Differences between the two datasets reflect their methodologies, but both show the same directional result.

Samsung benefited from strong conventional DRAM pricing, higher bit shipments, and its early HBM4 production ramp. Its estimated revenue rose 63.4% from the first quarter. Micron’s revenue increased 65.5% as it prioritized higher-priced server memory.

SK hynix revenue rose 37.9% sequentially, a result that would look exceptional in most periods. However, its product mix limited the benefit from surging conventional DRAM prices. Its growth therefore appeared restrained beside two faster-moving rivals.

High-bandwidth memory, or HBM, stacks DRAM dies vertically to feed processors more data with lower energy use. It is essential for AI accelerators, where processors can otherwise sit idle while waiting for model data.

HBM had the highest share of bit shipments at SK hynix among the three largest suppliers, according to TrendForce. That specialization helped the company during the initial AI infrastructure boom. It became a disadvantage when conventional DRAM prices increased more sharply.

The SK hynix DRAM share decline therefore does not signal that customers abandoned its memory products. It shows how a changing price mix can reshape revenue rankings even when every major supplier is growing.

SK hynix’s position is still commercially significant. Yet the second-quarter result challenges the idea that HBM leadership automatically produces leadership across the whole DRAM category.

Why the SK hynix HBM Strategy Missed the Conventional Memory Windfall

SK hynix optimized early for scarce AI memory, while the market’s strongest pricing shifted toward less specialized products.

The company’s early HBM expansion was rational. AI accelerator vendors needed dependable volumes, and advanced stacks required close coordination across memory design, packaging, and system qualification. Long-term agreements gave SK hynix clearer demand and protected the investments needed for production.

Those arrangements also strengthened its relationship with Nvidia. The partnership helped SK hynix become the largest HBM supplier while competitors worked through product qualification and manufacturing issues.

However, long-term agreements can trade pricing flexibility for predictable demand. They commonly establish pricing formulas, ceilings, floors, or periodic negotiation mechanisms before all units ship. The exact terms vary by customer and product.

Counterpoint attributed part of SK hynix’s second-quarter performance to agreements signed before memory prices accelerated. Its market ranking analysis said earlier fixed pricing left less room to capture the latest market increases.

The issue became more visible during the transition from HBM3E to HBM4. Average prices for existing HBM3E products declined from a year earlier, while HBM4 launches arrived later than expected. Conventional DRAM prices rose during the same period.

This produced an unusual reversal. The premium category remained strategically central, but ordinary DRAM delivered more immediate pricing upside. SK hynix had already committed a larger proportion of its output to the segment facing temporary price pressure.

The company disputes any suggestion that its contracts are inflexible. SK hynix says agreement structures differ by customer and product. It also says those contracts can protect downside results while preserving access to favorable demand and pricing.

That explanation is credible, but the precise contract economics are not public. Investors cannot independently compare ceilings, adjustment schedules, or customer-specific margins. Market-share data offers an indirect view rather than a complete contract audit.

HBM also consumes production resources differently from conventional DRAM. It requires multiple memory dies, additional manufacturing steps, advanced packaging, and strict validation. One finished stack therefore represents more than a simple substitution between equivalent products.

Capacity allocation becomes a portfolio decision. Every wafer, packaging line, and engineering team assigned to HBM carries an opportunity cost. During early AI demand, that cost looked small beside HBM’s margins and strategic value.

The calculation changed when conventional DRAM supplies tightened. PC, smartphone, and server customers faced limited availability after manufacturers devoted more resources to advanced memory. Rising prices rewarded suppliers that retained greater flexibility across product lines.

Samsung captured both parts of that shift. It benefited from conventional DRAM while expanding HBM shipments. Micron also gained through its server-focused product mix and growing AI memory presence.

SK hynix still retained the strongest HBM position. Yet the quarter showed that specialization creates exposure as well as pricing power. A leader can dominate the market’s most discussed product and still lose share in the larger category.

Samsung and Micron Are Squeezing SK hynix From Both Sides

Samsung now combines scale with a stronger HBM position, while Micron is challenging SK hynix through rapid server-memory growth.

Samsung is the immediate primary opponent because it can compete across the entire memory portfolio. It leads conventional DRAM, is expanding HBM shipments, and can direct large investments toward newer manufacturing processes.

Counterpoint estimated Samsung’s second-quarter DRAM share at 39%, matching its level from 2024. That represented a sharp reversal from the previous year, when SK hynix briefly led the market.

Samsung’s advantage is not simply higher output. Its broader product mix allowed it to collect rising conventional DRAM prices while adding HBM4 shipments. That balance reduced the opportunity cost of betting on one segment.

HBM competition also tightened. Earlier first-quarter data placed SK hynix at 58% of HBM revenue, while Samsung and Micron each held 21%. One year earlier, SK hynix had 69%, indicating that its lead remained large but was narrowing.

Samsung has taken a more aggressive technology route for HBM4. It applied 1c DRAM to the generation and promoted operating speeds around 11.7 gigabits per second. SK hynix chose established 1b DRAM for HBM4 to prioritize stable mass production.

The SK hynix HBM strategy treats manufacturability as a competitive feature. A product with slightly lower headline specifications can still win if it qualifies sooner, yields consistently, and ships in dependable volumes.

Samsung’s approach places more weight on early process advancement. If its 1c yields and customer qualifications hold, Samsung can present higher specifications while expanding output. That would pressure SK hynix on performance and volume at once.

Micron creates a different risk. It is not the current HBM leader, but its overall DRAM revenue nearly matched SK hynix during the second quarter. Counterpoint estimated Micron’s share at 25%, only one point behind its Korean competitor.

Micron’s DRAM revenue increased fivefold from the previous year, according to Counterpoint. Its strategy emphasized server DRAM during a period when capacity remained limited and prices rose. That focus captured demand connected to both conventional servers and AI infrastructure.

The result is a two-front contest. Samsung is rebuilding leadership from above through scale and HBM4. Micron is closing from below through server products, customer diversification, and disciplined allocation.

SK hynix cannot answer those pressures by abandoning HBM. Doing so would weaken the franchise that differentiates it and connects the company to leading accelerator platforms. It must instead increase flexibility without compromising customer commitments.

The competitive problem extends beyond a single quarter. Long-term AI contracts can secure volumes for years, while process transitions require extensive capital and qualification work. Decisions made today shape what each supplier can deliver several product generations later.

This is also why the contest matters to South Korea. Samsung and SK hynix anchor the country’s memory industry, export performance, equipment demand, and technical workforce. Their combined strength does not eliminate competition between them.

Samsung gaining share can support Korea’s aggregate position. However, SK hynix losing flexibility creates company-specific risk and a broader strategic question. Korea needs leadership in premium AI memory without leaving profitable conventional segments open to overseas suppliers.

SK hynix HBM4E Depends on a Faster 1c DRAM Ramp

The company’s answer is not a retreat from HBM, but a process transition that serves premium and conventional memory together.

SK hynix is rapidly increasing the proportion of DRAM made with its 1c process. Industry estimates cited by Chosunbiz placed that share near 10% in the first quarter and 13% in the second.

The share was forecast to reach about 24% during the third quarter and 34% in the fourth. If that schedule holds, 1c would overtake the older 1b process during the first quarter of 2027.

A smaller manufacturing node can produce more bits from a wafer when yields remain stable. That improves cost efficiency and expands effective output without requiring wafer starts to increase at the same pace.

SK hynix expects the transition to support general-purpose products as well as future AI memory. Its 1c DRAM roadmap connects the process directly to HBM4E, the generation expected to follow HBM4.

This creates a bridge between the two sides of the portfolio. More productive 1c lines can supply DDR5 and server products while preparing core dies for SK hynix HBM4E. The company gains flexibility if both product families share a mature process foundation.

That strategy differs from its HBM4 decision. SK hynix used 1b DRAM for HBM4 and paired it with its established MR-MUF packaging method. MR-MUF fills spaces between stacked dies and helps manage heat while supporting mass production.

The conservative HBM4 choice reduced the number of simultaneous manufacturing changes. Introducing a newer DRAM node and an advanced package together can increase execution risk. SK hynix delayed that combination until HBM4E.

For HBM4E, however, stability alone will not be enough. Future accelerators will demand higher bandwidth and improved energy efficiency. Samsung and Micron will also refine their processes before those products reach large-scale deployment.

The current transition therefore carries two objectives. SK hynix must recover productivity in the wider DRAM market. It must also ensure that SK hynix HBM4E remains competitive with products built on similarly advanced processes.

Manufacturing yield is the central uncertainty. Yield measures the proportion of usable chips produced from each wafer. A new process improves theoretical density, but poor yields can erase its cost and capacity advantages.

Customer qualification creates another gate. AI accelerator vendors test memory for performance, thermal behavior, reliability, and integration with their systems. A manufacturer cannot convert a technical sample into substantial revenue until customers approve it.

The reported production mix remains an industry estimate rather than a company-published quarterly series. That makes the ramp direction more useful than any single percentage. Actual yields and customer schedules will determine whether the transition delivers its expected benefits.

SK hynix has also approved major domestic expansion. In August, its board authorized 35.2 trillion won for a second Yongin fab and 19.1 trillion won for the M17 facility in Cheongju.

The fab investment plan totals 54.3 trillion won. Yongin is intended as a next-generation DRAM production hub, including HBM, while Cheongju will expand NAND capacity.

New fabs will not solve immediate allocation problems. Semiconductor plants take years to construct, equip, qualify, and ramp. Near-term results depend more heavily on process migration and optimization inside existing facilities.

That timing makes the 1c ramp the most important operational response. It offers a route to additional bits before the largest new facilities reach useful scale.

Korea’s Strategy Faces a Commodity DRAM Blind Spot

Focusing on premium AI memory raises national value, but it also creates openings in products that fund scale and customer reach.

South Korea remains the center of advanced memory production through Samsung and SK hynix. Their HBM capabilities place Korean suppliers inside the most valuable AI accelerator supply chains.

However, leadership in premium products does not make conventional DRAM strategically irrelevant. Standard memory remains essential for CPUs, smartphones, personal computers, networking equipment, and general servers.

AI systems themselves require both categories. GPUs use HBM for extremely fast data access, while host processors rely on large quantities of conventional server DRAM. Growth in AI infrastructure can therefore tighten both markets.

The danger appears when established suppliers allocate too much attention away from mature segments. Customers still need those products, and price increases create an entry point for competitors willing to serve them.

China’s ChangXin Memory Technologies, or CXMT, illustrates that opening. Counterpoint estimated that CXMT’s second-quarter revenue increased 716% from the previous year. Its expansion centered on conventional DRAM and Chinese domestic demand.

A separate CXMT market analysis reported that the company raised 29.5 billion yuan through a Shanghai listing. It plans to direct part of those funds toward DRAM processes and research.

CXMT still faces meaningful constraints. Advanced equipment access, production yield, thermal performance, and overseas customer qualification can slow its progress. Mass-producing mature DRAM is also different from qualifying competitive HBM.

Those limits should not obscure the strategic pattern. A supply gap lets a newer producer build volume, improve manufacturing experience, and deepen customer relationships. Profit earned from conventional products can then finance more advanced development.

The same logic shaped earlier memory cycles. Suppliers accumulated process knowledge through high-volume production before moving into more demanding products. Market participation matters even when the initial products do not lead on performance.

Korea therefore confronts a portfolio problem rather than a simple choice between advanced and commodity chips. HBM provides higher strategic value, while conventional DRAM preserves scale and blocks openings for competitors.

SK hynix’s recent share decline makes that tradeoff visible. The company has not lost its technological position overnight. It has allowed competitors to capture more of the revenue growth in products outside its strongest franchise.

Samsung currently looks better balanced. It expanded HBM while retaining enough exposure to benefit from conventional pricing. That does not prove Samsung’s strategy will win every generation, but it strengthened the company during this market phase.

Government policy also has limited reach. Public support can accelerate infrastructure, workforce development, research, and domestic supply chains. It cannot determine which product mix customers will reward in each quarter.

Corporate execution remains decisive. Korean suppliers must qualify products on time, maintain yields, secure equipment, and negotiate contracts that preserve flexibility. Those tasks occur at the company level, even when their outcome has national implications.

The larger risk is complacency created by aggregate dominance. Samsung and SK hynix together still control a large portion of global DRAM. That combined position can hide divergent company trajectories and emerging competition.

A resilient Korean strategy needs both firms to compete effectively across cycles. One company cannot permanently compensate for lost opportunities at the other, especially as Micron and CXMT expand from different directions.

Three Signals Will Show Whether SK hynix Can Rebalance

The next test is measurable: 1c yields, HBM4 customer economics, and the gap between SK hynix and Micron.

The first signal is the 1c production ramp through the fourth quarter. Industry estimates expect 1c to reach about 34% of SK hynix DRAM production by year-end.

Hitting that level with stable yields would strengthen the company’s cost position and increase available bits. Missing it would weaken the planned bridge between conventional DRAM and SK hynix HBM4E.

Investors should watch company commentary about bit growth, process migration, and qualification rather than focusing only on the reported production share. Those operating indicators can reveal whether the new node is producing usable volume.

The second signal is the pricing and customer mix for HBM4. Shipments began increasing during the third quarter as newer Nvidia and AMD systems entered the market. That transition should reduce the effect of falling HBM3E prices.

Volume alone will not settle the question. The crucial issue is whether SK hynix can capture higher HBM4 value while keeping contracts responsive to market changes. Customer diversification will matter alongside Nvidia demand.

Samsung’s progress provides the comparison. If Samsung combines growing HBM4 volumes with strong conventional DRAM revenue, its balanced approach will gain further credibility. If its advanced process creates yield problems, SK hynix’s stability-first decision will look stronger.

The third signal is the revenue-share gap with Micron. Counterpoint placed only one percentage point between the companies in the second quarter. Micron overtaking SK hynix would be a visible warning about portfolio flexibility.

Such a change would not end SK hynix’s HBM leadership. It would show that a competitor with a smaller HBM position can produce comparable DRAM revenue through server products, pricing, and capacity allocation.

CXMT belongs on the longer watchlist, but its immediate meaning is different. Growth in its conventional DRAM share would show that established suppliers left a durable opening. Progress toward qualified HBM would raise the strategic stakes further.

The SK hynix HBM strategy remains one of the clearest successes of the AI hardware boom. It placed the company at the center of accelerator supply and transformed its competitive standing.

Now that success must evolve. SK hynix needs to preserve dependable HBM supply, improve contract flexibility, accelerate 1c yields, and protect its broader DRAM position. Doing all four is harder than simply producing more premium memory.

For developers and enterprise AI buyers, the outcome will influence accelerator availability, server configurations, and infrastructure costs. Memory supply can constrain deployments even when processors are available.

The next several months should answer a direct question. Can SK hynix turn its HBM specialization into a balanced memory advantage, or will Samsung and Micron keep capturing the growth around it?