SEOUL, July 30 (AJP) - South Korea's two memory giants confirmed their supremacy in back-to-back second-quarter earnings releases, generating a combined operating profit of about 150 trillion won ($108 billion), equivalent to roughly 6 percent of the country's entire 2025 nominal GDP, with margins in the 70 percent range that few manufacturers could dream of.
Samsung Electronics and SK hynix delivered the same assurance: the extraordinary combination of tight supply, explosive artificial-intelligence demand and multiyear customer commitments could last for at least two more years.
Investors remained unconvinced.
SK hynix ended Thursday at 1,298,000 won, down about 25 percent this week and more than 56 percent from its June 26 peak of 2,987,000 won. Samsung Electronics closed at 207,000 won, about 20 percent lower on the week and nearly 45 percent below its June 19 high of 374,500 won.
The selloff revealed the central contradiction of Korea's AI memory boom. The chipmakers are earning profits at a scale rarely seen in global manufacturing, but markets are no longer rewarding the records. Investors are instead asking how long margins near 70 percent can survive before rivals expand supply, customers resist higher prices or the AI investment cycle begins to slow.
Samsung Electronics on Thursday reported a record operating profit of 89.2 trillion won from its chipmaking operations in the second quarter, translating into an operating margin of about 70 percent.
The result came one day after SK hynix posted an industry-leading margin of 76.1 percent, the highest among major global chipmakers and ahead of Taiwan Semiconductor Manufacturing Co. and Micron Technology.
Together, the earnings underscored South Korea's dominance in AI memory, where surging demand for high-bandwidth memory has transformed its two largest semiconductor manufacturers into the industry's biggest profit generators.
Samsung's Device Solutions division posted operating profit of 89.2 trillion won, compared with 8.1 trillion won a year earlier, while revenue surged to 127.5 trillion won on record demand for DRAM and NAND products.
Companywide, Samsung confirmed quarterly revenue of 171.5 trillion won and operating profit of 89.5 trillion won, both the highest in its history.
A stronger U.S. dollar added about 3.1 trillion won to operating profit from the previous quarter, mainly benefiting component businesses. Research and development spending reached a record 16 trillion won as the company continued to invest in advanced memory, foundry processes and packaging technologies.
Samsung said its memory business delivered another record quarter by responding to strong server demand driven by the spread of agentic AI and the continued expansion of hyperscale data centers.
The company increased shipments of HBM4 and became the first in the industry to ship HBM4E samples, strengthening its claim to technological leadership after trailing SK hynix during the initial stages of the HBM boom.
Its System LSI business posted record first-half revenue despite softer overall demand, supported by higher sales of mobile processors and image sensors. The foundry business also improved as demand increased for HBM base dies and products from U.S. customers.
Yet the broader industry picture remained even more striking.
SK hynix's 76.1 percent operating margin reflected the premium pricing and supply discipline generated by its early lead in HBM, the fastest-growing and most profitable segment of the memory industry.
Samsung, meanwhile, is using its broader scale across DRAM, NAND, foundry and advanced packaging to narrow the gap and secure a larger share of the AI infrastructure market.
The contrasting share-price reaction showed how dramatically investor expectations have shifted.
Both companies argued that the current cycle differs fundamentally from previous memory booms.
During its earnings conference call, Samsung said explosive investment in AI infrastructure, agentic AI and sovereign AI projects would keep demand ahead of supply through at least 2028, even as chipmakers accelerate capacity expansion.
The company said the long lead times required to build fabrication plants, install advanced equipment and expand packaging capacity would prevent supply from catching up quickly with demand.
That outlook suggests a structural shift rather than another conventional one- or two-year memory cycle.
Samsung said customers are increasingly signing multiyear supply agreements and making advance payments to secure future capacity, practices that were uncommon in earlier upcycles when new factories eventually flooded the market and drove prices lower.
The company has secured long-term supply agreements with five global hyperscale customers and is in the final stages of talks with another five AI-related clients.
Once those negotiations are completed, Samsung expects multiyear contracts to cover 60 to 70 percent of its planned medium-term production capacity.
The shift could weaken the memory industry's traditional boom-and-bust pattern.
For decades, DRAM and NAND prices swung sharply between scarcity and oversupply because manufacturers expanded production aggressively during booms, only to face collapsing prices once new capacity entered the market.
Longer customer commitments, advance payments and years-long construction timelines could instead lock in demand and keep capacity structurally tight through much of the AI infrastructure expansion.
For Samsung Electronics and SK hynix, that would improve earnings visibility and reinforce South Korea's central role in the global AI supply chain.
Samsung expects HBM4 shipments to expand significantly in the second half, while its foundry business should benefit from growing demand for advanced processing and HBM base dies.
SK hynix is also seeking to protect its early lead through next-generation HBM products and deeper collaboration with major AI chip customers.
The durability of their dominance, however, will depend on how quickly rivals catch up in HBM, advanced packaging and memory production, as well as whether hyperscale technology companies can continue financing AI infrastructure at the current pace.
Copyright ⓒ Aju Press All rights reserved.