Sinyal kontraintuitif dalam siklus super memori AI Samsung: Mengapa DRAM lebih menguntungkan daripada HBM?

29 April 2026, Samsung Electronics (005930.KS) delivered a quarterly financial report capable of rewriting the history of Korean corporations.

Consolidated revenue of 133.87 trillion won (approximately $901 billion), up 69.2% year-on-year, up 42.7% quarter-on-quarter, setting a new record for single-quarter revenue. Operating profit of 57.23 trillion won (approximately $384 billion), surged 756.1% year-on-year, soaring 185.1% quarter-on-quarter, not only exceeding internal company expectations but also outperforming the market consensus of 55.28 trillion won.

A more intuitive comparison: in all of 2025, Samsung Electronics' total operating profit was 43.6 trillion won. This means that just one quarter of 2026's profit has already surpassed the total for the entire previous year.

In the profit structure, the Device Solutions (DS) division responsible for semiconductor business has become the absolute profit engine. This division's quarterly operating profit reached 53.7 trillion won, accounting for about 93.9% of the company's overall operating profit, whereas in the same period last year it was only about 1.1 trillion won. The DS division's operating profit margin is approximately 65.7%, which is the core driver behind Samsung's overall operating profit margin rising to about 42.8%.

After the earnings release, multiple international investment banks quickly raised their target prices. Goldman Sachs raised the target price of 005930 from 205,000 won to 260,000 won, maintaining a "buy" rating; Citibank on May 12 raised the target from 300,000 won to 460,000 won and initiated a 90-day upward catalyst watch; Nomura on May 18 increased the target from 340,000 won to 590,000 won.

However, what truly makes the market ponder is not these financially fully priced figures, but a remark casually made by management during the earnings call: at the current point in time, the profitability of traditional DRAM is actually higher than HBM.

This statement directly conflicts with the mainstream market perception. In the narrative of the "AI memory supercycle," HBM is regarded as the most profitable and most barrier-rich track. Why would management release such an "counterintuitive" signal in the best quarterly report in history? The underlying business logic behind this is far more complex than the financial data itself.

An "Counterintuitive" Profit Structure

The judgment that traditional DRAM is more profitable than HBM is not an isolated statement from Samsung, but is supported by clear industry data.

According to the latest report released by semiconductor research firm TrendForce on May 27, 2026, influenced by annual bargaining mechanisms, the wafer output value and profit margins of HBM chips were surpassed by DDR5 RDIMM in Q1 2026. TrendForce had previously pointed out in January that server DRAM prices and profit margins had already overtaken HBM, becoming the most profitable item.

Specifically, the pricing cycle of HBM differs structurally from standardized DRAM. HBM contracts are usually negotiated on an annual basis, with longer lock-in periods, and price adjustments lag behind the spot market; whereas traditional DRAM (such as DDR5) has a more responsive price discovery mechanism, reflecting supply and demand gaps more quickly and allowing for higher price increases.

From the price data perspective, this contrast is even more intuitive. In Q1 2026, general-type DRAM contract prices increased by 90% to 95% quarter-over-quarter, while NAND Flash rose by 55% to 60%. Against the backdrop of large-scale HBM capacity being squeezed, supply of traditional DRAM shrank even faster than demand growth, creating a positive spiral of "scarcer and more expensive, more expensive and more profitable."

The overall performance of the global DRAM market also confirms this trend. According to data released by Counterpoint Research on May 27, 2026, global DRAM revenue in Q1 surged 80% quarter-over-quarter and 260% year-over-year, reaching a record $97 billion, just a step away from a single-quarter billion-dollar revenue. Samsung maintained its leading position with a 38% market share in DRAM revenue.

Behind this profit margin inversion, there is also a deeper capacity logic: capacity squeeze effect. HBM chips are significantly larger than standard DRAM of the same generation, consuming far more wafer capacity. As HBM shipments increase, a large portion of advanced process capacity is occupied, causing effective supply of traditional DRAM to continue shrinking. The imbalance of supply and demand, combined with more flexible pricing mechanisms, ultimately drives traditional DRAM to surpass the relatively rigid pricing of HBM in marginal profit.

HBM Track: Competition Under the Aura

Understanding the profit inversion between traditional DRAM and HBM also requires examining the competitive landscape of the HBM market.

In the key generation HBM4, Samsung achieved the world's first mass production and shipment of HBM4 in February 2026, delivering commercially to customers. The company expects HBM's overall sales in 2026 to exceed three times last year's, with HBM4 revenue surpassing half of total HBM revenue from Q3.

In terms of technological iteration, Samsung is also accelerating the development of HBM4E. At the GTC conference in March 2026, Samsung publicly showcased HBM4E samples, with single-pin speeds reaching 16 Gbps and total bandwidth exceeding 4.0 TB/s, about 21% higher than HBM4 bandwidth. The first batch of HBM4E engineering samples started production in Q2, prioritized for key customers for system-level validation.

However, market share data reveals a different picture. According to TrendForce, in 2026 SK Hynix's share of the HBM market is expected to be about 50%, with Micron increasing to 28%. The global DRAM revenue ranking from Counterpoint Research shows Samsung leading with 38% market share, followed by SK Hynix at 29% and Micron at 22%.

While Samsung has an early advantage in mass production of HBM4, SK Hynix, with a longer-standing partnership with Nvidia on HBM, still enjoys an early lead in supply allocation. This "pioneer barrier" is not solely determined by technical parameters but is built on long-term certification cycles, joint validation history, and supply chain trust. Whether Samsung's catch-up strategy can achieve a share breakthrough in HBM4 and subsequent generations depends on whether it can convert its production lead into substantial customer supply share while maintaining technological pace.

Pricing Power Game: Profit Lag Under Annual Agreements

TrendForce's May 2026 report reveals the core contradiction in HBM pricing mechanisms: HBM contracts for 2026 adopt an annual pricing framework, with prices locked in during the second half of 2025. When the overall DRAM market entered a severe shortage in Q1 due to AI demand, HBM contract prices remained "fixed" at relatively low levels, unable to rise with the standard DRAM market.

Additionally, the HBM market in 2026 saw the entry of new suppliers, increasing supply and providing some support to buyers in contract negotiations, further suppressing short-term upward price elasticity of HBM.

This lagging pricing framework has already triggered strong reactions from suppliers. TrendForce points out that as HBM4 supply negotiations for 2027 begin in Q2 2026, manufacturers will significantly raise HBM prices to reflect supply-demand imbalance and new manufacturing costs. The higher profits of traditional DRAM give suppliers more bargaining power—if HBM's long-term profits remain below those of traditional products, manufacturers lack economic incentives to expand HBM capacity, creating a game that pushes buyers to accept higher HBM contract prices.

SK Hynix's strategy differs. According to TrendForce, amid the price reversal of RDIMM over HBM, SK Hynix has explicitly stated "not pursuing short-term profits, but aiming for long-term balance between HBM and conventional DRAM." This differentiated stance suggests that Samsung and SK Hynix may be diverging in capacity allocation strategies: Samsung prefers to maximize short-term gains through high-margin traditional DRAM, while SK Hynix focuses on consolidating long-term HBM supply position.

Labor Strike in May 2026: Supply Risks in the Supercycle

In an environment of full capacity, Samsung Electronics faced its largest labor conflict since its founding in May 2026. The nationwide Samsung union announced an 18-day general strike. The union demanded that 15% of annual operating profit be allocated to employee bonuses, while management only agreed to 10% and refused to institutionalize the bonus system.

On May 18, the Suwon District Court partially approved Samsung's strike ban request, requiring during the strike period to maintain safety-related personnel and obligations, and prohibiting raw material deterioration, with a daily fine of 300 million won for violations. On May 20, labor-management mediation ultimately broke down, but a temporary salary agreement was reached in late May. After a vote, 73.7% of union members approved the agreement, avoiding the planned large-scale strike.

The deep root of this strike lies in the imbalance of profit distribution during the AI cycle. In Q1, Samsung's operating profit surged 756%, but chip employees felt they had not received commensurate returns from this AI-driven profit explosion. Meanwhile, SK Hynix had already reached an agreement with unions in September 2025, removing the bonus cap (originally capped at 1,000% of basic salary) and including 10% of annual operating profit into the employee bonus pool for ten years. It is estimated that in 2026, SK Hynix employees could receive bonuses of 600 million to 700 million won per person, while Samsung's chip employees' bonuses are less than one-third of that. This pay gap has led to about 200 core Samsung engineers defecting to SK Hynix in recent months.

From an industry perspective, if subsequent strikes impact production rhythm, the consequences will quickly propagate through the global semiconductor supply chain. JPMorgan estimates that a large-scale internal strike at Samsung could reduce 2026 operating profit by about 2.1 trillion to 3.5 trillion won and decrease semiconductor revenue by about 1% to 2%. Korea's KB Securities predicts that if only about 30-40% of union members participate in the strike, it could disrupt about 3-4% of global DRAM supply. In the current environment of ongoing supply-demand gaps, any substantial capacity loss will further push up the storage chip price center.

The Three Layers of the Supercycle

Integrating the above information, the current AI memory supercycle can be clearly understood as a three-layer structure:

First layer: Structural demand-driven. Unlike past short-term recoveries driven by consumer electronics replacement cycles, this storage rally is driven by the structural demand re-evaluation from AI training, inference, and large data center expansion. Nomura Securities in a research report stated: as AI semiconductor demand shifts from training to inference workloads, memory demand is entering an exponential expansion period. In inference scenarios, the demand for KV-cache memory is a product of multiple variables: user base, participation duration, AI task complexity, and inference token consumption, meaning memory demand could grow thousands of times over the next five years.

Second layer: Capacity squeeze and supply rigidity. Large-scale expansion of HBM has occupied a significant amount of advanced process wafer capacity, directly squeezing the effective supply of traditional DRAM. After experiencing a downturn, storage manufacturers have become more cautious in capital expenditure, with new capacity release cycles typically taking 12-18 months. IDC China Vice President Zhou Zhengang predicts this storage price increase cycle will span beyond 2026, possibly extending into 2027 or 2028.

Third layer: Reallocation of pricing power. As cloud service providers sign long-term supply agreements locking in low prices, overall DRAM prices in 2026 show a "hard to fall, easy to rise" structure. Meanwhile, the annual pricing framework for HBM faces a fundamental reset, with 2027 contract negotiations expected to see price increases far exceeding previous market expectations. TrendForce explicitly states that manufacturers will dynamically balance capacity allocation based on profit comparisons between HBM and traditional DRAM.

The interaction among these three layers determines the unique nature of this supercycle: it is not only a boom driven unilaterally by demand but also a "disciplined prosperity" constrained by supply rigidity—manufacturers, aiming to maximize profits, are motivated to maintain rather than break the current supply-demand structure. This makes the duration and stability of this cycle markedly different from the traditional storage cycle of "boom—oversupply—crash."

Conclusion: Internal Tensions of the Supercycle

Samsung's Q1 2026 financial report appears as a perfect footnote to the "AI memory supercycle"—record-high revenue and profit, exponential growth rate, and tripling of HBM shipments. But beneath these dazzling numbers, the remark about "traditional DRAM profitability exceeding HBM" during the call reveals an overlooked internal tension within the supercycle narrative.

This tension manifests on multiple levels: the inverted profit structures between traditional DRAM and HBM, the lagging annual pricing mechanism of HBM and upcoming contract price resets, the ongoing market share and technological race between Samsung and SK Hynix, and the internal game between AI profit explosion and profit distribution. These tensions are not negating the supercycle but are key clues to understanding its unique nature.

From an investment perspective, Samsung's current valuation narrative is at an important turning point. After the market fully digests the explosive growth in profit volume, the next phase of pricing focus will shift to the health and sustainability of profit structures—whether HBM contract prices can catch up in 2027, whether labor risks can be institutionalized, and whether HBM market share can achieve breakthrough growth. The evolution of these variables will determine whether Samsung Electronics can transition from a "cycle beneficiary" to a "structural leader" in the AI memory supercycle.