How AI turned memory chips into a gold mine
AI and the memory-chip shortage send Samsung and SK Hynix profits soaring
It is a historic first for SK Hynix. In 2025, the memory-chip maker became South Korea’s most profitable company, narrowly overtaking giant Samsung. The changing of the guard is driven by the breakneck expansion of infrastructure dedicated to generative artificial intelligence. This boom is fueling both an explosion in demand for high-bandwidth memory (HBM) — a segment largely dominated by the Korean group — and a broad shortage that is pushing selling prices sharply higher.
Last year, SK Hynix posted the highest operating profit in its history: 47 trillion won ($32.2 billion), twice as much as in 2024. Growth was even more spectacular than that of its revenue, which rose 47% over the same period. The favorable environment is also benefiting Samsung, its main rival in the memory-chip market, which reported 44 trillion won in operating profit. In the fourth quarter alone, the group even set records, buoyed by a surge in profits from its semiconductor division.
HBM sales multiplied by ten
The computing power required to train and run AI models does not rely solely on graphics cards sold at premium prices by Nvidia. These must be paired with memory, particularly the latest-generation HBM chips, which save space while reducing energy consumption. Since 2023, this market has expanded more than tenfold, reaching $35 billion last year, according to estimates from Bank of America. It is expected to grow by a further 58% this year, to $55 billion.
Only three players manufacture HBM chips. According to consultancy Counterpoint, SK Hynix captured 57% of sales in the third quarter of 2025. In just three years, its revenue has more than tripled, smashing the peaks reached in 2021 during the previous chip shortage. Behind it, Samsung is beginning to close its technological gap. Its market share has climbed back to 22%, edging just ahead of US rival Micron. The South Korean conglomerate is banking heavily on the launch of the fourth generation of HBM to sustain its commercial momentum.
Other chips squeezed
Faced with strong demand for AI memory, the three leading manufacturers have redirected part of their production lines toward these more lucrative, higher-margin components. As a result, profit growth is far outpacing revenue growth. This reallocation is occurring at the expense of other types of chips, notably DRAM and NAND memory used in smartphones, computers and game consoles. According to consultancy TrendForce, 70% of global production is now devoted to AI.
Demand for AI is “absorbing such a large share of available capacity that it is creating an unprecedented shortage for other segments of the industry,” a Micron executive told Bloomberg. Order books at SK Hynix, Samsung and Micron are already full for 2026. The consequence: prices for other components are soaring. In the fourth quarter of 2025, DRAM prices jumped by 45% to 50%, and are expected to rise by a further 55% to 60% in the first quarter. Enough to further boost manufacturers’ profits.
A “disciplined” investment strategy
SK Hynix, Samsung and Micron did not anticipate demand on this scale. Worse still, all three companies underinvested in recent years following the severe overproduction crisis that hit the sector from late 2022. While they now plan to increase capacity, new production lines will not come on stream before 2027 or 2028. They are also remaining cautious in their expansion strategies. Even as it speaks of a “significant” increase in investment, SK Hynix insists its policy will remain “disciplined.”
The South Korean group intends to keep capital expenditure at 30% of revenue. Its rival Samsung also promises to invest more, but without significantly expanding production capacity, in order to “prioritize long-term profitability”. In an industry accustomed to cycles — regularly swinging between shortages and oversupply — the memory-chip giants fear a market reversal if enthusiasm for AI fades just as their new factories come online.