SK hynix reported record second-quarter results on Wednesday as strong demand for high-bandwidth memory (HBM) chips used in AI servers continued to reshape the semiconductor industry.

The South Korean chipmaker said revenue reached 79.3187 trillion won and operating profit came to 60.5426 trillion won, increases of 257% and 557%, respectively, from a year earlier. First-half revenue also rose above 100 trillion won for the first time in the company’s history.

Despite the record earnings, SK hynix shares fell after the announcement. Investors had expected even stronger results as AI-related expectations had been building for months, underscoring how elevated market expectations have become across the AI chip sector.

That reaction highlights a broader shift in the chip industry: companies tied to AI infrastructure are now being judged not just on growth, but on whether they can keep pace with demand that has become both concentrated and difficult to supply.

Why memory is at the center of the AI trade

High Bandwidth Memory, or HBM, has become a core component of AI infrastructure. By vertically stacking memory chips, HBM delivers far greater bandwidth and better power efficiency than standard DRAM, allowing Nvidia’s AI accelerators and other chipmakers’ AI systems to handle heavy workloads more efficiently.

As Nvidia’s leading HBM supplier, SK hynix is widely viewed as one of the clearest indicators of demand for AI infrastructure.

In its earnings announcement, SK hynix said demand for its products continues to grow as AI moves toward more advanced “agentic” systems capable of performing higher-level functions.

To prepare for future demand, the company has signed long-term contracts with about 10 major clients, giving them the right to reserve production capacity years in advance.

The same trend is visible in the broader memory market. TrendForce expects contract prices for DRAM to rise 58% to 63% in the third quarter, while NAND flash prices could increase about 70% to 75%, as manufacturers prioritize HBM and server memory over consumer products. The price increases are being driven by limited memory supply, investment in hyperscale cloud infrastructure, and long-term contracts.

Samsung and Micron are benefiting too

SK hynix is not the only memory company benefiting from AI demand. Samsung Electronics recently said it expects second-quarter operating profit of 89.4 trillion won and revenue of about 171 trillion won, citing strong demand for AI memory, though questions remain over whether the industry’s rapid growth can continue.

Samsung has also taken an aggressive position in HBM4. TrendForce said the company became the first to make HBM4 commercially available in large quantities after starting mass production in February. TrendForce also reported that Samsung has generated more than $1 billion in HBM4 revenue within four months of the product’s release.

Samsung said its commercial HBM4 offers data transfer speeds of 11.7 Gbps per pin, above the JEDEC minimum of 8 Gbps, along with better power and thermal efficiency.

Micron Technology has also reported record quarterly results. Chief Executive Officer Sanjay Mehrotra called memory “a strategic asset” in the age of artificial intelligence and said hyperscale customers are signing more multi-year supply contracts to secure future supply.

The shortage is spreading to phones and PCs

The AI boom is also creating shortages in other parts of the market.

IDC has warned of an “unprecedented memory chip shortage” that could last until as late as 2027 because more production is being diverted to HBM for AI accelerators rather than to PCs and smartphones.

The research firm said global DRAM supply is expected to rise by roughly 16% and NAND supply by about 17%, both below historical averages.

As supply tightens, smartphone makers may need to absorb higher component costs, reduce device specifications, or pass price increases on to consumers. In IDC’s downside scenario, global smartphone shipments could fall by as much as 5.2%.

HBM4 competition is intensifying

Competition in next-generation AI memory is also heating up. SK hynix began mass shipments of HBM4 during the quarter and plans to expand production in the second half.

The company said its latest chips meet customer performance requirements while improving power efficiency, and that it has already completed sample shipments of HBM4E for future AI accelerators.

Samsung, however, moved first in commercial HBM4 production and has gained a meaningful competitive edge. TrendForce estimates that Samsung will generate more than $1.2 billion from HBM4 sales this year, making it the first supplier to surpass $1 billion in HBM4 revenue.

SK hynix has taken a different approach, prioritizing the HBM3E market before expanding HBM4 production. The company expects to ramp up HBM4 output from the third quarter onward.

Looking ahead, SK hynix said it intends to expand advanced chip packaging while improving NAND production under strict budget discipline.

The rise of AI has transformed the memory business, shifting the focus of chipmakers from simple output toward bringing faster, more advanced AI memory products to market.