SK Hynix posts record profits but its shares still drop 6.5%, here's why
The South Korean memory chip maker beat almost every measure of growth, yet Wall Street wanted even more. What that tells us about AI's voracious appetite for computer memory.
Key points
- SK Hynix revenue hit 79.32 trillion won ($54.55 billion) in Q2 2025, up 257% year-on-year.
- Operating profit soared 557% year-on-year to 60.54 trillion won, but missed analyst estimates of 64 trillion won.
- Shares fell 6.5% on Thursday despite the record results.
- First-half 2025 cumulative revenue exceeded 100 trillion won, a company first.
- SK Hynix has a multiyear supply deal with Nvidia worth over $500 billion.
Imagine acing an exam with 95% and your parents asking why it wasn't 100%. That, roughly, is what happened to SK Hynix this week.
The South Korean company makes memory chips, the components that let computers store and access data at speed, and supplies them to everything from smartphones to the giant data centres that run AI services. On Wednesday it reported second-quarter results so large they barely look real: revenue up 257% compared with the same period a year ago, and operating profit up nearly 557%. Yet its shares fell 6.5% on Thursday, because analysts had pencilled in even bigger numbers.
What did analysts actually expect?
Analysts were hoping for 84 trillion won in revenue and 64 trillion won in operating profit. SK Hynix delivered 79.32 trillion won ($54.55 billion) and 60.54 trillion won respectively. Close, but not close enough for a market running on maximum optimism.
| Metric | Reported | Expected |
|---|---|---|
| Revenue | 79.32 trillion won | 84 trillion won |
| Operating profit | 60.54 trillion won | 64 trillion won |
| Revenue growth (year-on-year) | +257% | , |
| Operating profit growth (year-on-year) | +557% | , |
| Gross margin | 83% | , |
Josh Gilbert, lead analyst for the Asia-Pacific region at trading platform eToro, pointed out that an 83% gross margin, meaning the company keeps 83 cents of profit for every dollar of revenue after manufacturing costs, is a sign that customers are scrambling for supply. "That doesn't exist in a market where demand is drying up," he said. "It exists in one where customers are fighting over supply."
Why is demand so high?
AI is the short answer. Training and running AI models requires enormous amounts of a specialised memory type called HBM (high-bandwidth memory), which stacks chips on top of each other to move data much faster than ordinary memory. SK Hynix is one of the leading makers of HBM, and Nvidia, the dominant seller of AI processors, is one of its biggest customers. The two companies recently signed a multiyear supply agreement worth over $500 billion.
SK Hynix shipped its latest generation chip, HBM4, in volume for the first time during Q2, and has already sent samples of the next step up, HBM4E, to customers for testing.
On the NAND side, a different type of memory used for longer-term storage in devices like SSDs (solid-state drives, the faster replacement for old spinning hard disks), the company says its most advanced 321-layer chips already make up the largest share of what it produces, with a target of 50% of domestic capacity by year-end.
What does this mean for ordinary people?
Most of us won't buy SK Hynix chips directly. But if you use ChatGPT, stream video, or own a recent smartphone or laptop, there's a reasonable chance SK Hynix memory is somewhere in the chain.
For investors, the story is a useful reminder: extraordinary growth can still disappoint if expectations ran too far ahead. First reported by CNBC Tech, the results are a snapshot of just how much money is flowing into AI infrastructure right now, and how hard it is for even the biggest beneficiaries to keep up with the hype.
Common questions
Why would a company's shares fall after record profits?
Share prices reflect what investors expected, not just what happened. When a company is seen as an AI darling, analysts set very high forecasts, and even strong results can fall short of those forecasts, pushing the price down.
What is HBM and why does AI need so much of it?
HBM, or high-bandwidth memory, is a type of chip that moves data extremely fast by stacking multiple memory layers into one compact unit. AI models need to process vast amounts of data simultaneously, and HBM is currently the best way to feed that data to the processors doing the work.
Should I worry about the share price drop?
A single-day drop after an earnings miss is normal market behaviour, not a sign of trouble. SK Hynix's underlying demand, its margins, and its pipeline all look strong. The drop reflects reset expectations more than any change in the company's fortunes.
