Summary of Key Points
SK Hynix reported explosive growth in its Q2 2026 results (operating profit up 557% year-on-year, net profit up 1242%), but due to both revenue and profit falling short of market expectations, coupled with concerns about the sustainability of capital spending in the upstream AI industry chain, the stock price plummeted after the financial report was released (over 6% decline in U.S. stocks and 12.97% decline in Korean stocks). This also caused a collective drop in the stock prices of other memory and AI hardware companies such as Micron and Sandisk. The market is beginning to question: Has the boom cycle for memory chips come to an end?
I. Outstanding Performance, but Still Being Sold Off: What's the Problem?
SK Hynix' profit growth seems impressive, but there are three reasons why the market isn't buying in:
1. Core indicators fell short of expectations: Operating profit was 60.54 trillion Korean won (about 280 billion RMB), 3.68 trillion less than expected; revenue was 79 trillion Korean won, also 5 trillion lower than anticipated. For investors, "not meeting expectations" is more significant than "fast growth."
2. Profit margin seems inflated: The net profit soared by 1242%, but a large portion of this came from the investment income from selling shares of铠hea (about 60.9 trillion Korean won), which is not money earned from core business operations. It's like your salary didn't increase much, but you suddenly made a big profit by selling a house—this kind of profit isn't sustainable.
3. The "bitter pill" of HBM: SK Hynix is the leader in HBM (high-end memory dedicated for AI), but HBM is subject to long-term supply agreements (LTAs) with fixed prices, so it can't benefit from rising spot market prices like traditional memory companies can. Meanwhile, its competitor Samsung, with a more balanced product portfolio, is actually making more money from the price increases in traditional memory, making SK Hynix' performance seem less impressive.
II. The Market Is Panicking: How Long Can AI Capital Spending Last?
The root of this decline lies in market doubts about the financial sustainability of the upstream AI industry chain:
- The trigger was Google's financial report: Google's Q2 capital spending doubled to $44.9 billion, but its free cash flow turned negative ($5.85 billion). In other words, it spent more money than it earned. The market is worried: Can big companies really continue to invest heavily in building AI data centers?
- CICC's analysis reveals the bubble: AI capital spending is shifting from using their own earnings to borrowing money, with an estimated $3.5 trillion in external financing needed over the next five years. To cover these costs, AI-related revenue would need to double annually while maintaining a 50% profit margin—a task as challenging as getting a rocket to re-use ten times.
- Valuation has already factored in too much optimism: SK Hynix' market value once exceeded $1.35 trillion, incorporating benefits such as HBM shortages, memory price increases, and AI expansion. Now, even slight underperformance leads to a "Davis Double Kill" (both valuation and performance decline).
III. SK Hynix' Dominant Position in HBM Is Weakening: Will the Market Share Be Shaken?
SK Hynix once held over 63% of the HBM market share, but it has now dropped to 56.4% due to competition:
- Samsung and Micron are accelerating their entry: Samsung began mass-producing HBM4 in February, and Micron skipped HBM3 to go straight to HBM3E and HBM4, increasing its market share from 5.8% to 23.1%. All three companies have passed NVIDIA's certification, which means NVIDIA is likely looking for more suppliers to reduce dependence on a single supplier.
- The double-edged sword of long-term agreements: Although SK Hynix has signed long-term supply agreements with ten customers, the prices in these agreements are fixed. If market prices drop, SK Hynix can't raise them; if competitors lower their prices to poach business, it has no choice but to accept the loss.
IV. Has the Memory Cycle Really Ended?
There's still no definitive answer, but the market's attitude has shifted from enthusiasm to rationality:
- Positive signals: SK Hynix expects a 10% increase in DRAM shipments in the third quarter, full production of HBM4 in the second half of the year, and long-term orders from ten customers. Global DRAM demand is expected to grow by about 20% this year, with NAND growing by 10%. The demand for memory will continue due to AI inference and the expansion of intelligent systems.
- Negative signals: Whether cloud companies can sustain their capital spending is crucial. If big companies cut back on investment due to cash flow issues, demand for memory chips will decline. Moreover, the market now values returns more and no longer blindly believes in the idea that "spending money will lead to success." Apple's market value surpassing NVIDIA again is a sign that investors prefer companies that can generate stable profits rather than those relying on heavy capital spending.
Conclusion
Whether the memory cycle has ended depends on the financial reports of cloud companies. If they continue to invest heavily in data centers, the cycle may extend; if capital spending begins to shrink, the good times for memory chips might be over.
Final Summary
SK Hynix' performance shows that memory chips are currently very profitable, but the market is concerned about how long this profitability will last. The story of AI infrastructure development is still unfolding, but investors are already starting to calculate whether the investments will pay off.
(End of article)