利润暴增 557%，股价仍遭重挫：海力士的存储牛市还在么？

Profit surged 557% year-on-year, with an operating profit margin of 76%, both setting new historical records.

For any manufacturing company, these would be near-dreamlike figures, but SK Hynix achieved them: Q2 revenue reached 79.32 trillion KRW, up 51% quarter-on-quarter and 257% year-on-year; operating profit reached 60.54 trillion KRW, up 61% quarter-on-quarter and 557% year-on-year, both hitting new all-time highs.

However, after the earnings report was released, SK Hynix's stock price plummeted over 15% during intraday trading.

The reason is not complicated. The market had previously expected revenue of around 84 trillion KRW and operating profit of about 64 trillion KRW. The actual results were slightly below these expectations. In other words, Hynix did not deliver a poor earnings report, but rather one that was near-perfect, yet failed to meet the market's 'ultra-perfect' demands.

This is also a common challenge currently faced by AI hardware assets.

When the market already knows that revenue will grow, prices will rise, and profits will break records, simply showing 'good results' is no longer enough. Investors need confirmation of how quickly profits can continue to exceed expectations, and how long this memory bull market can last.

1. Numbers Still Strong, Below Expectations Doesn't Necessarily Mean Weakening Demand

First, look at the core business.

In the second quarter, SK Hynix's operating profit margin further improved from 72% in the first quarter to 76%, a sequential increase of 4 percentage points. This means the company not only sold more memory products but also saw improvements in product prices, sales mix, and unit profitability.

During the earnings call, it was disclosed that the average selling price of DRAM increased by about 30% quarter-on-quarter in Q2, NAND ASP rose by about 50% quarter-on-quarter, enterprise SSD revenue doubled quarter-on-quarter, and revenue from Solidigm's high-capacity enterprise SSDs (over 30TB) more than tripled quarter-on-quarter.

Simply put, the driver of Hynix's profitability is no longer just one single product.

HBM, AI server DRAM, enterprise SSDs, and price increases in traditional memory are all contributing simultaneously. The rising proportion of high-value products, combined with memory price increases and fixed cost amortization, has created extremely strong operational leverage— each additional unit of revenue is converting into a higher proportion of operating profit.

So, why was operating profit still below market expectations?

The company explained during the call that shipments of some high-value-added products were delayed to the second half of the year, and the product mix change also dragged down the blended average selling price. Therefore, as HBM4 shipments officially ramp up in the second half and 1c process (6th generation 10nm-class) DRAM gradually increases volume, these effects are expected to ease.

It's important to distinguish between two completely different scenarios:

• If the results fell short due to order cancellations, rising inventory, or a sudden weakening in end demand, it might signal a cyclical turning point.
• But if it's merely a deferral of revenue recognition for high-value products, then the 'lower earnings' this quarter could theoretically translate into incremental gains in the second half.

Based on the disclosed information, this miss is more likely due to a delayed product realization schedule, rather than a breakdown in the demand logic.

Furthermore, the net profit of 93.92 trillion KRW in Q2 cannot be directly equated to Hynix's core operating profitability.

Non-operating income for the quarter reached 62.2 trillion KRW, including approximately 63.3 trillion KRW in gains from the sale and valuation of investment assets. The market widely believes this includes gains from the disposal of investments related to Kioxia. Therefore, the true measure of the memory business's operational capability remains the operating profit of 60.54 trillion KRW, not the net profit which exceeds revenue.

Overall, excluding one-off investment gains, this remains an extremely strong earnings report. The only problem is that the market is no longer satisfied with just 'extremely strong'.

2. HBM4 Begins to Deliver, Memory Demand is Broadening

More noteworthy than exactly how much less Hynix earned in Q2 is how its growth engine is changing.

First, HBM4.

SK Hynix confirmed that HBM4 has started volume production and shipment in Q2, with plans to ramp up output in the second half. HBM4E also completed customer sample delivery in the first half.

The company stated that the current HBM4 product has achieved the operating speed required by customers, while also offering high energy efficiency, yield, and cost competitiveness. This indicates HBM4 is officially transitioning from customer validation, order expectations, and technical narratives into the revenue realization phase.

Although the deferral of some HBM4 shipments was one reason for the Q2 miss, it also creates clearer growth potential for the second half. The company expects DRAM shipments to increase by about 10% quarter-on-quarter in Q3, while NAND shipments are expected to show low single-digit growth. It is actively trying to secure long-term supply agreements (LTAs) to reduce the cyclical volatility typical of the traditional memory industry.

Data shows the company has completed LTA negotiations with about 10 customers and continues discussions with other major clients. These contracts are typically five-year agreements, involving long-term purchase commitments and potentially mechanisms like performance bonds. Pricing structures vary based on the client, product, and market conditions, aiming to balance price stability with market flexibility.

While LTAs don't guarantee memory prices will only go up, and might sacrifice some short-term upside during rapid price increases, they can provide more stable purchase commitments, capacity planning, and cash flow visibility. For a memory industry long plagued by the cycle of 'price rise → concentrated capacity expansion → inventory glut → price crash,' this change itself is part of an improvement in earnings quality.

More importantly, this demand wave is no longer limited to HBM around GPUs.

As we all know, market discussions about Hynix used to focus almost entirely on HBM. However, this earnings report shows growth is now spreading to traditional memory.

In Q2, Hynix's AI server DRAM, enterprise SSDs, and SOCAMM2 sales all grew simultaneously, and 1c process products began formal shipments. In NAND, 321-layer products have become the highest volume product, with the company planning to increase their share to about 50% of its domestic capacity by year-end.

The logic behind this isn't complicated.

HBM provides high-bandwidth data to GPUs, server DRAM handles larger memory capacity and agent operation demands, and enterprise SSDs handle long-term storage for model data, inference results, databases, KV Cache, and data lakes. As training scales up, more HBM is needed; as inference requests increase, it simultaneously boosts demand for both server DRAM and enterprise SSDs.

Hynix also disclosed that enterprise SSD revenue has doubled quarter-on-quarter, and it is developing new products for KV Cache offloading and near-GPU storage.

This means Hynix's investment thesis is transitioning from a single 'HBM leader' to a full-stack AI memory platform covering HBM, server DRAM, enterprise SSDs, and advanced NAND.

This is the most optimistic signal in this earnings report: growth is not reliant on just one star product but is spreading across a broader memory hierarchy.

Of course, traditional memory remains the biggest source of cyclical risk in this upswing.

HBM supply expansion is relatively slow, constrained by customer qualification, advanced packaging, and customization requirements. For commodity DRAM and NAND, once prices keep rising, Samsung, Hynix, Micron, and Chinese memory manufacturers all have incentives to ramp up production.

Therefore, while this earnings report confirms the current memory market is still tight, it alone cannot prove that an oversupply situation won't emerge after 2027.

3. Market Enters 'Hard Mode', But the Memory Bull Market Hasn't Peaked

Objectively, the market's biggest current fear is whether Hynix, with too many orders and too much cash, will be tempted to bet on another round of massive capacity expansion.

As of the end of Q2, SK Hynix held cash and short-term investments of 88 trillion KRW, an increase of 33.6 trillion KRW from Q1. Interest-bearing debt fell to 18.6 trillion KRW, resulting in net cash of 69.4 trillion KRW.

In short, Hynix has transformed from a company that needed to manage debt and cash flow during the last memory downturn into a net cash powerhouse with strong expansion capabilities.

The company estimates 2026 capital expenditure will reach 40 trillion KRW, up from 30.2 trillion KRW in 2025. In the short term, it will accelerate the mass production of M15X and prepare capacity for the Yongin Phase 1 clean room opening in early 2027. In the medium to long term, it will advance the P&T7 advanced packaging facility, the M17 NAND production base, and a new semiconductor industrial cluster.

These investments need to be understood in two parts.

Spending on HBM4, advanced packaging, leading-edge processes, and high-end enterprise SSDs is essentially filling existing supply bottlenecks. However, if capacity for commodity DRAM and NAND also expands rapidly during a period of high prices, it could sow the seeds for a future oversupply.

Therefore, high CapEx itself is not a negative. The real determinant of the cycle's direction is where the money is spent, when the capacity comes online, and whether long-term orders are in place to absorb it.

Based on current information, Hynix emphasizes that new capacity will be phased in based on confirmed customer demand and LTAs, rather than a one-off massive expansion. The company also believes its current capacity plans won't immediately lead to oversupply.

However, the market has reasons to remain cautious:

• On one hand, an operating margin of 76% is already at an extreme high. Even if profits continue to grow, the room for further margin expansion becomes increasingly limited.
• On the other hand, while LTAs can raise the floor for earnings, they might reduce the upside profit elasticity during periods of rapid price increases.
• Additionally, with cash accumulating rapidly, investors are demanding clearer shareholder return plans. Hynix stated that due to procedural restrictions related to its ADR issuance, it cannot currently disclose the specific form and scale of buybacks, dividends, or other capital return measures, but plans to announce further arrangements within the year.

In a nutshell, this earnings report did not prove the memory bull market is over, but it clearly shows that the trading logic for Hynix's stock has changed.

After all, HBM4 shipments have started, some high-value product revenue is deferred to H2, DRAM and NAND prices are still rising, enterprise SSD demand continues to expand, and about 10 customers have signed multi-year purchase agreements. These signals are hard to reconcile with a memory cycle entering a contraction phase.

It's just that the market has started to scrutinize how long this boom can last, using a higher bar.

Final Thoughts

Overall, this earnings report has not proven that the memory cycle has peaked.

It resembles more of a rhythm mismatch under high expectations – the market raised the passing grade from 100 to 110. Hynix delivered a near-perfect score, yet it was still deemed insufficient.

But falling short of expectations is not the same as a demand reversal.

With HBM4 entering mass production, about 10 customers signing LTAs, DRAM and NAND prices rising simultaneously, and demand for enterprise SSDs and AI server memory broadening, by all accounts, Hynix remains in a high-prosperity phase. Its growth engine is actually becoming more diversified, and its fundamentals are rock-solid.

However, the trading logic has changed.

The stock is transitioning from a 'profit growth trade' to a re-pricing cycle focusing on cycle length, earnings quality, and capital allocation.

Of course, the next record doesn't necessarily need to be more dramatic than this one. As long as Hynix can continue to prove that the 76% margin is not an isolated peak, but is being supported by longer order books, broader demand, and more disciplined capacity expansion.

Then the peak of summer is far from over.