SK Hynix delivered record quarterly profit, yet the market reaction was harsh. The company posted second-quarter revenue of KRW 79.32 trillion, up 51% from the prior quarter and 257% from a year earlier. Operating profit reached KRW 60.54 trillion, up 61% quarter over quarter and 557% year over year, while operating margin rose to 76%. After the earnings release, SK Hynix shares fell more than 15% intraday.
The issue was not weak performance. It was that the results did not clear an even higher bar set by investors. Market expectations had been around KRW 84 trillion for revenue and KRW 64 trillion for operating profit. Actual results came in slightly below both marks. In a market where AI hardware names are already priced for exceptional growth, simply posting outstanding numbers no longer settles the debate. Investors now want proof that profit can keep beating expectations at speed and that the memory upcycle still has room to run.
Margins kept rising, and the miss does not automatically point to weaker demand
The underlying operating picture was still strong. SK Hynix lifted operating margin from 72% in the first quarter to 76% in the second, a gain of 4 percentage points. That kind of move suggests more than higher shipment volume. It also reflects better pricing, an improved product mix and stronger profitability per unit sold.
On the earnings call, the company said average selling prices in the second quarter rose about 30% quarter over quarter for DRAM and about 50% for NAND. Enterprise SSD revenue doubled from the prior quarter, while revenue from Solidigm’s enterprise SSD products above 30TB rose by more than twofold.
That matters because earnings are no longer being driven by one product line alone. HBM, AI server DRAM, enterprise SSDs and price gains in traditional memory all contributed. As higher-value products made up a larger share of sales, memory price increases and fixed-cost absorption combined to create powerful operating leverage. Each incremental unit of revenue translated into a larger share of operating profit.
So why did operating profit still miss expectations? The company said on the call that some shipments of high-value products were delayed into the second half, while changes in product mix weighed on blended average selling prices. It added that these factors should ease as HBM4 shipments expand in the second half and 1c process DRAM, described as the sixth-generation 10-nanometer-class node, ramps gradually.
The distinction is important. If a miss comes from order cancellations, rising inventories or a sudden drop in end demand, that may point to a cycle turning down. If the issue is delayed revenue recognition for higher-value products, then part of what was not booked this quarter could show up later in the year. Based on what the company disclosed, this result looks closer to a shift in revenue timing than a break in the demand story.
Net profit was KRW 93.92 trillion in the second quarter, but that figure does not directly represent the core earning power of the memory business. Non-operating income reached KRW 62.2 trillion during the quarter, including about KRW 63.3 trillion in gains from investment asset sales and valuation changes. The market broadly believes this included gains tied to the disposal of Kioxia-related investments. For that reason, operating profit of KRW 60.54 trillion remains the cleaner measure of core memory operations.
Strip out one-off investment gains, and the report was still extremely strong. The market’s problem was that extremely strong was no longer enough.
HBM4 is moving into revenue, and AI memory demand is spreading beyond one segment
The more revealing part of the report was not the size of the shortfall against expectations. It was the shift in the company’s growth engine.
Start with HBM4. SK Hynix said HBM4 began mass production shipments in the second quarter and that volumes will expand in the second half. HBM4E samples were also delivered in the first half. The company said its current HBM4 products meet customer requirements for operating speed and also show strong energy efficiency, yield and cost competitiveness. That means HBM4 is moving out of the validation-and-expectation stage and into actual revenue recognition.
Some HBM4 shipments were among the items pushed into the second half, which contributed to the earnings miss. But that same shift also leaves a clearer runway for later growth. The company expects third-quarter DRAM shipments to rise about 10% sequentially, while NAND shipments are expected to post low-single-digit growth. It is also trying to reduce the memory industry’s cyclical swings through long-term supply agreements.
According to the company, around 10 customers have completed negotiations on long-term deals, and talks continue with other major customers. These contracts are usually structured around five-year terms. They can include long-term purchase commitments and, in some cases, performance mechanisms such as deposits. Pricing structures vary by customer, product and market conditions so that contracts can preserve some balance between price stability and market flexibility.
Long-term agreements do not mean memory prices can only go up. They may also reduce some of the upside capture when spot prices spike quickly. Still, they offer clearer purchase commitments, better capacity planning and improved cash flow visibility. In an industry that has long been trapped in a repeated sequence of price increases, aggressive capacity expansion, inventory build and price collapse, that change itself can improve earnings quality.
Just as important, this demand wave is no longer limited to HBM linked closely to GPUs. For much of the recent rally, the SK Hynix story was mostly an HBM story. This earnings report suggests the expansion is spreading into broader layers of memory.
In the second quarter, sales of AI server DRAM, enterprise SSDs and SOCAMM2 all increased, while products based on the 1c process began official shipments. In NAND, 321-layer products became the company’s largest product by output share, and SK Hynix plans to lift that share to about 50% of domestic Korean capacity by year-end.
The logic is straightforward. HBM supplies high-bandwidth data to GPUs. Server DRAM supports larger memory capacity and agent workloads. Enterprise SSDs store model data, inference outputs, databases, KV Cache and data lakes over longer periods. If training scales up, more HBM is needed. If inference requests rise, demand should increase for both server DRAM and enterprise SSDs.
SK Hynix also said enterprise SSD revenue doubled sequentially and that it is developing new products aimed at KV Cache offloading and storage closer to GPUs. That shifts the investment case. The company is no longer framed only as an HBM leader. It is increasingly being viewed as a full-stack AI memory platform spanning HBM, server DRAM, enterprise SSDs and advanced NAND.
That is one of the more constructive signals in the report. Growth is not resting on a single flagship product. It is spreading across a wider set of memory layers.
Even so, traditional memory remains the biggest cyclical risk in the current run. HBM supply expands more slowly because it is constrained by customer qualification, advanced packaging and customization requirements. Standard DRAM and NAND are different. If prices stay high, Samsung, SK Hynix, Micron and Chinese memory makers all have an incentive to raise output. This report shows that the memory market is still tight now, but it does not by itself prove there will be no oversupply after 2027.
Demand remains strong, but investors are now pricing duration, quality and capital discipline
One of the market’s clearest concerns is whether SK Hynix, flush with orders and cash, might eventually lean too hard into another major capacity expansion cycle.
At the end of the second quarter, the company held KRW 88 trillion in cash and short-term investments, up KRW 33.6 trillion from the first quarter. Interest-bearing debt fell to KRW 18.6 trillion, leaving net cash at KRW 69.4 trillion. In simple terms, SK Hynix has moved from being a company focused on controlling debt and cash flow during the last memory downturn to one with the balance sheet to expand aggressively.
The company expects capital expenditure to reach KRW 40 trillion in 2026, above KRW 30.2 trillion in 2025. In the near term, it plans to bring forward mass production at M15X and prepare capacity for the first Yongin cleanroom, which is set to open in early 2027. Over a longer horizon, it will also move ahead with the P&T7 advanced packaging plant, the M17 NAND production base and a new semiconductor industry cluster.
Those investments need to be split into two categories. Spending tied to HBM4, advanced packaging, advanced process nodes and high-end enterprise SSDs is largely about relieving supply bottlenecks that already exist. But if conventional DRAM and NAND capacity also expands quickly while prices are high, that could plant the seeds for future oversupply.
That is why high capex, by itself, is not necessarily bearish. What matters is where the money is going, when the capacity comes online and whether it is backed by long-term orders. Based on current disclosures, SK Hynix said new capacity will be phased in according to confirmed customer demand and long-term agreements rather than through a one-shot full-scale expansion. The company also said its present capacity plan does not immediately lead to oversupply.
Investors still have reasons to stay cautious. A 76% operating margin is already at an extreme level, so even if profit keeps rising, the room for further margin expansion becomes narrower. Long-term contracts can raise the floor under earnings, but they may also reduce upside when prices jump fast. And with cash building so quickly, shareholders are looking for clearer capital return plans.
SK Hynix said that because of procedures related to ADR issuance, it cannot yet disclose the specific form or size of buybacks, dividends or other capital return measures. It said further details are planned within the year.
The earnings did not show the memory boom is over. They showed the stock’s playbook has changed
On fundamentals alone, this report did not make the case that the memory bull market has ended. HBM4 has started shipping. Some high-value revenue has shifted into the second half. DRAM and NAND prices are still rising. Enterprise SSD demand continues to expand. Around 10 customers have signed multi-year purchasing agreements. Those signals are hard to square with a memory cycle already in contraction.
What has changed is the market framework. SK Hynix is no longer being priced only on profit growth. The stock is now being judged on how long the cycle can last, how durable earnings quality is and whether management can deploy capital without recreating the industry’s old boom-bust pattern. In that setting, missing elevated expectations is not the same as a reversal in demand. It does mean investors want more evidence that a 76% operating margin is not a one-quarter peak, but a level supported by longer contracts, broader demand and tighter expansion discipline.
Based on the information disclosed in this earnings report, the memory upcycle has not topped out. What has risen is the standard the market is using to measure it.