Samsung, SK Hynix Sold Off on AI Fears. The Orders Haven't Followed Yet
The selloff reflects a market recalculating AI capex speed and memory earnings expectations, not confirmed cancellations. So far, the data show a gap between fear and reality.
Stocks fell first. Orders haven't, at least not yet
After Anthropic CEO Dario Amodei suggested slowing the pace of frontier AI model development in favor of independent safety reviews and more industry coordination, Sam Altman and Elon Musk both signaled some agreement on the need to pace things out. Markets translated that almost instantly into a new risk: AI server investment could slow too.
Here is the distinction that matters most: possibility is not the same as fact. What's confirmed so far is a safety debate over development speed, and a stock price correction. What is not confirmed is any large-scale cancellation of AI data center capex, any termination of long-term HBM contracts, or any order pullback from major customers.
The physical memory market is still tight
On September 16, Reuters reported that smartphone and laptop makers are bracing for a prolonged memory shortage. SK Hynix has said it views 2027 as a particularly difficult year in terms of supply capacity, and DRAM prices, while rising more slowly now, remain elevated.
SK Hynix's second-quarter earnings pointed the same direction. The company said it began mass shipments of HBM4 in the second quarter and is expanding long-term, multi-year contracts with roughly 10 core customers. Samsung Electronics started HBM4 mass production and commercial shipments in February, and shipped HBM4E samples in May.
None of this guarantees an endless boom. But at this point, there isn't enough evidence to conclude that HBM demand has already turned down just because an AI safety debate has surfaced.
Samsung and SK Hynix now need to be read differently
| SK Hynix | Samsung Electronics | |
|---|---|---|
| Core investment thesis | HBM leadership, direct beneficiary of AI server demand | HBM4 expansion plus recovery across commodity memory, foundry and mobile |
| Confirmed HBM progress | Mass shipment of HBM4 started, long-term contracts expanding | HBM4 mass production and commercial shipments started, HBM4E samples shipped |
| Sensitivity to an AI slowdown | Relatively high | Some cushion from business diversification |
| Key risks | High expectations, customer concentration, HBM pricing and yield shifts | Pace of HBM share gains, foundry profitability, mobile demand |
| Numbers to watch first | HBM shipment volumes, ASP, long-term contracts, AI capex | HBM mix, DRAM prices, DS division profit, foundry P&L |
The difference matters. SK Hynix is a purer bet on AI memory growth. Samsung is no longer just a company trying to catch up in HBM. It has entered a stage where the question is whether HBM4 mass production actually translates into market share gains and margin expansion.
The real risk isn't AI fear itself. It's the moment three numbers turn
1. Big Tech capex
If hyperscalers like Microsoft, Alphabet, Amazon and Meta start lowering their data center investment guidance, expectations for HBM demand could get a more realistic reset.
2. HBM orders and pricing
The key signal is whether long-term contract expansion stalls, or whether order delays, rising customer inventory and weakening pricing power show up at the same time.
3. Commodity DRAM prices
If HBM stays strong but commodity DRAM turns down, the profit leverage across the memory industry stays limited. If both markets stay strong together, the quality of earnings improves.
4. Interest rates
Rising long-term US rates can pressure valuations for chip stocks priced on future growth. That's one reason the stock can wobble even when earnings look fine.
Insight Times Editorial Desk
