April 30, 2026, Samsung Electronics released a quarterly earnings report that stunned the market: total revenue reached KRW 133.9 trillion, up 43% quarter-over-quarter and soaring 69% year-over-year, setting a new all-time record for single-quarter revenue. Operating profit surged to KRW 57.2 trillion, jumping 185% quarter-over-quarter and skyrocketing 756.1% year-over-year. The Device Solutions division, responsible for semiconductors, posted revenue of KRW 81.7 trillion and operating profit of KRW 53.7 trillion—both single-quarter highs. The DX division reported revenue of KRW 52.7 trillion and operating profit of just KRW 3 trillion, with semiconductors contributing 93.9% of the company’s total operating profit.
Almost simultaneously, SK Hynix delivered a financial report showing operating profit up 405% year-over-year: revenue of KRW 52.58 trillion, up 198% year-over-year; operating profit of KRW 37.61 trillion, with an operating margin of 72%—surpassing TSMC’s 58% and Micron’s 67.6%. SanDisk, which spun off from Western Digital just over a year ago, has seen its stock price soar roughly 3,640% since its IPO at $38.50 in February 2025, with another 460% gain so far in 2026.
Yet beneath these breathtaking numbers, three cracks are quietly emerging. They may not immediately reverse the cycle, but they serve as a reminder: even the most spectacular supercycle has its steep peaks.
Crack One: Samsung’s "Excess Profit" Dilemma—48,000 Employees Nearly Went on Strike
On May 21, 2026, a massive strike involving about 48,000 employees—originally expected to last 18 days—was narrowly averted at the last moment. With the mediation of Korea’s Minister of Labor, Samsung Electronics management and labor reached a preliminary wage agreement: 10.5% of operating profits from the semiconductor business will be allocated to a special bonus pool. Estimates show that the 28,000 employees in the memory chip division will receive an average bonus of KRW 600 million each this year, while employees in the mobile and TV divisions will get just KRW 6 million—a 100-fold gap.
These numbers reveal a deep internal divide. The semiconductor division contributes over 93% of Samsung’s operating profit, while the DX division—including mobile phones and home appliances—continues to struggle with profit margins. Mobile business head TM Roh has already warned internally that soaring memory chip procurement costs could lead to the division’s first-ever annual loss in 2026. Seoul National University professor Song Jae-ho summed up the issue: "When PhD-level employees in the DX division receive only KRW 6 million in bonuses, while high school-educated production workers in the semiconductor division get KRW 600 million, it’s hard for employees to see this structure as truly fair."
On May 27, the union voted with nearly 74% approval to accept the wage agreement, temporarily ending the strike threat. However, KB Securities analysts estimated that an 18-day strike could have reduced global DRAM supply by 3% to 4% and NAND supply by 2% to 3%. Even though the strike didn’t happen, the episode revealed a key fact: when a supercycle generates unprecedented excess profits, internal distribution tensions can become a black swan for the supply chain.
| Metric | Samsung Semiconductor Division | Samsung DX Division (Mobile/Home Appliances) |
|---|---|---|
| Q1 2026 Revenue | KRW 81.7 trillion | KRW 52.7 trillion |
| Q1 2026 Operating Profit | KRW 53.7 trillion | KRW 3 trillion |
| Average Annual Bonus per Employee | Approx. KRW 600 million (Memory Division) | Approx. KRW 6 million |
| Core Drivers | HBM4 mass production + AI server demand | Galaxy S26 flagship sales |
Crack Two: Traditional DRAM Profit Margins Surpass HBM—A Supply-Demand Mismatch
A severely underestimated signal comes from the production decisions of Samsung and SK Hynix.
On April 14, SK Hynix reduced its planned HBM4 shipments for 2026, cutting supply to Nvidia by about 20% to 30% from previous plans. The company clarified this was not due to weak demand, but rather an intentional optimization of product mix—shifting capacity toward higher-margin HBM3E and server-focused LPDDR products.
This aligns closely with Samsung’s messaging. During its Q1 earnings call, Samsung stated that HBM4 achieved global first mass production in February, but traditional DRAM products are now outperforming HBM in profit margins. Industry data shows that in Q1 2026, contract prices for standard DRAM rose from the initial forecast of 55%-60% to 90%-95%. In Q2, contract prices for generic DRAM continued to climb by 58%-63% quarter-over-quarter. Consumer Mobile DRAM prices jumped 80% in a single quarter. For example, DDR5 16Gb climbed from $5.524 to about $40 in one year—a 627% increase.
A typical comparison highlights the profit hierarchy mismatch: HBM chips have much larger die sizes than traditional DRAM, yielding fewer chips per wafer—producing 1GB of HBM consumes 3 to 4 times the wafer area of traditional DRAM. During price surges, traditional DRAM can actually deliver higher unit profits. Driven by profit maximization, Samsung and SK Hynix are reallocating capacity, disrupting the previously expected linear logic of "expanding HBM production to ease supply constraints."
The deeper issue is this "crowding effect" is bidirectional: HBM consumes massive wafer capacity (expected to reach 25% of industry front-end capacity by end-2026 and 31% by 2027), further tightening traditional DRAM supply and driving prices higher. At the same time, high profits from traditional DRAM incentivize manufacturers to invest more in legacy capacity, slowing HBM expansion. Both categories are simultaneously "short" and "profitable," temporarily breaking the supply-side self-correction mechanism.
Crack Three: Capital Expenditure Surges from $83 Billion to $144 Billion—Why Isn’t Capacity Keeping Up?
The market’s most perplexing paradox: capital expenditure is expanding rapidly, yet supply constraints are intensifying.
A JPMorgan report released April 22 showed its forecast for global memory industry capex in 2027 was raised from $83 billion in September 2025 to $144 billion—a 74% increase. Samsung’s 2026 capex plan exceeds KRW 110 trillion (about $73.3 billion), a historic high and up 21.7% from 2025. SK Hynix expects to spend about $20.5 billion, up 17% year-over-year.
However, projected bit supply for DRAM and NAND in the same period is only up about 12%-14%. The surge in capex hasn’t translated into proportional capacity growth—due to three main dampeners:
First, HBM’s die area consumption. Producing 1GB of HBM consumes 3 to 4 times the wafer area of traditional DRAM, with yields only 50%-60%. This structural loss can’t be offset by simply investing in more equipment. Second, extended equipment lead times. Samsung ordered about 20 EUV lithography machines from ASML, totaling over KRW 10 trillion, but delivery takes 18-24 months, and new production lines require at least five years from construction to full ramp-up. Third, manufacturers’ "strategic restraint." After suffering huge losses from misjudging demand in the previous cycle, Samsung and SK Hynix are cautious about expanding generic DRAM capacity.
| Supply Constraint Dimension | Specific Manifestation | Impact on Cycle |
|---|---|---|
| HBM Die Area Consumption | HBM uses 3-4x wafer area compared to traditional DRAM at same node | Reduces effective bit output by 7%-10% |
| Extended Equipment Lead Times | EUV lithography delivery 18-24 months | New capacity delayed until late 2027 or later |
| Manufacturer Strategic Restraint | New plants need at least five years from construction to full ramp | Limited incremental legacy capacity |
| China Capacity Variable | ChangXin Memory plans new plant for 2027 | Could become biggest supply-side wildcard |
At the same time, research firm Counterpoint’s data is even more aggressive: between 2026 and 2027, DRAM annual capacity needs to grow at 12% to ease shortages, but actual growth is only about 7.5%. By the end of 2027, global DRAM supply will meet only about 60% of market demand.
Three Narratives: Bulls, Bears, and the Middle Ground
The sustainability of the AI memory supercycle has sparked three distinct market narratives.
Bullish narrative: Structural shortages will persist at least through the end of 2027. UBS forecasts the DRAM supercycle will last until at least the end of 2027, with HBM-dedicated capacity rising from 25% to 31%, and "the oligopoly structure lacks competitive expansion incentives." JPMorgan believes the supercycle could extend to 2027 or 2028.
Bearish narrative: A cycle reversal may arrive in the second half of 2027. On May 18, former Samsung Semiconductor president Kyung Kye-hyun issued a major warning at the Korea National Academy of Engineering forum: Chinese companies are aggressively expanding capacity, and global supply structure may see a major inflection in the second half of 2027, with prices likely to reverse downward. He predicts the price drop effect will be fully apparent by the first half of 2028 at the latest. If AI investment returns disappoint, large tech firms may cut capex—"then not only prices, but demand itself could weaken."
Middle-ground narrative: The cycle will lengthen but won’t disappear; only the volatility pattern will change. SanDisk is pioneering a "new business model"—locking in future multi-year revenue with long-term contracts worth at least $42 billion and over $11 billion in financial guarantees for downside protection—changing the memory industry’s traditional quarterly pricing model. CEO David Goeckeler said, "The pain in this industry has always been the cyclical boom-and-bust." The middle-ground view is representative: the memory market remains cyclical, and while this cycle may be longer and stronger, valuations already price in much optimism—JPMorgan set SK Hynix’s target price at KRW 1 million, based on 2.7x price-to-book, a 30% premium to historical peaks.
Assessing Narrative Authenticity: What Risks Are Underestimated?
Within these mainstream narratives, three often-overlooked risks deserve attention.
First, the risk of "software and hardware substitutes." In March 2026, negative sentiment emerged amid concerns that hardware-level SRAM optimization and software-level data compression could reduce AI systems’ overall memory consumption. This controversy itself shows that demand-side stories aren’t set in stone.
Second, the ceiling for downstream resilience. In Q2 2026, the server DRAM market saw clear bargaining: upstream manufacturers insisted on price hikes, while downstream customers refused to accept high prices and turned to second-hand or recycled chips. When this substitution behavior hits a critical mass, it will materially suppress contract prices.
Third, the risk of valuation bubbles. SanDisk’s current price-to-sales ratio is about 16x, up from just 4.5x at the start of the year; its stock price reached about $1,410, up over 460% so far in 2026. Even with strong fundamentals, such valuation expansion means any news of order slowdowns could trigger sharp corrections. Samsung Electronics’ stock is up roughly 100% year-to-date, and SK Hynix’s shares have also soared—market pricing already embeds two to three years of sustained boom, which is itself a risk.
Industry Impact Analysis: Who Bears the Cost of the Supercycle?
This supercycle is not without cost, and the burden is being shouldered by the weakest links in the supply chain.
On the supply side, leading manufacturers have tightened supply through tiered allocation strategies, firmly controlling pricing power: from quota-based deliveries and NCNR agreements, to requiring advance payments for orders in April and May 2026, and locking in long-term contracts through 2027-2028. Small and mid-sized buyers face a dilemma: accept sky-high spot prices or simply get no supply. SK Hynix has made it clear: in 2026, no customer’s demand can be fully met, and suppliers hold absolute pricing power.
On the downstream consumer electronics side, DRAM and NAND procurement costs are projected to account for over one-third of entry-level smartphone total costs by mid-2026. PC makers are forced into panic buying, and some smaller brands are experiencing a "rationing era" in the supply chain—with some large orders now taking more than 20 to 40 weeks to fulfill. Even Samsung’s own mobile division faces the risk of its first-ever annual loss due to soaring memory chip costs.
In AI infrastructure, the shortage and high price of memory chips are becoming a physical bottleneck limiting the speed of AI compute deployment. While hyperscale cloud providers have secured most available capacity through long-term agreements, ongoing supply constraints mean not every AI project can get the hardware it needs on schedule—perhaps the most ironic twist of this cycle.
Conclusion
The AI memory supercycle is a genuine and profound structural shift—not a short-term hype-driven phenomenon. This should be acknowledged first. From Samsung to SK Hynix to SanDisk, the financials of these core players all point in the same direction: AI’s demand for memory chips is real, massive, and still accelerating.
But no supercycle offers a smooth ride. Samsung’s internal fractures, the profit hierarchy between traditional DRAM and HBM, and the lag between capital spending and capacity release—none of these cracks are fatal on their own, but together they form potential triggers for a cycle reversal. Especially with the uncertainties of China’s capacity expansion and AI investment returns unresolved, 2027 may become the true watershed for this supercycle.
What the market should really be wary of is not when the supercycle will end, but whether enough participants are prepared with answers before that endpoint arrives.
