Key Takeaways
Amazon shares jumped as the company ramped up investment in artificial intelligence (AI) infrastructure, while Apple fell the same day on concerns that it lacks AI exposure. What matters to investors isn't the two stocks' moves themselves, but which components — and how much of them — go into each data center Amazon builds. The answer leads straight to Samsung Electronics and SK Hynix.
What Happened
In early trading on U.S. markets on the 31st, Amazon and Apple shares diverged. Amazon posted a sharp gain (surge) on signs that its AI infrastructure expansion is translating into earnings, while Apple lagged relatively due to a business structure with limited AI exposure. The split can be read as the difference between a company that has already paved the "AI highway" and one that hasn't yet.
What stands out is the path behind Amazon's earnings improvement. As its cloud division (AWS) invests more in servers and storage to handle AI workloads, Amazon rents out that capacity and charges for it, collecting revenue that works like a toll. Amazon is the one building the road — the data centers — but the materials consumed with every mile of road laid, DRAM, HBM, and NAND, are supplied by someone else: Samsung Electronics and SK Hynix.
Background and Context
A single AI server carries far more DRAM and HBM than a conventional server. Each accelerator chip contains multiple layers of stacked HBM, and the CPUs and memory slots that support it are paired with high-capacity DRAM as well. News that a hyperscaler is raising its capex (capital expenditure) is itself a leading indicator that pushes up the memory demand curve. Amazon's expanded data center investment ultimately flows upstream into order books at memory suppliers.
Impact on the Market and Stocks
- Samsung Electronics - With supply exposure to hyperscalers in both commodity DRAM and HBM, expanded capex from cloud companies including Amazon puts upward pressure on memory utilization rates and average selling prices.
- SK Hynix - With a high proportion of revenue coming from HBM, it has the greatest sensitivity to the benefits of AI server capacity expansion. However, as HBM stacking layers increase, yield management also becomes more difficult, so capacity expansion doesn't translate directly into profit.
- Apple - Its earnings are driven more by on-device AI features and hardware sales cycles than by AI infrastructure investment, leaving it relatively left out of this rally.
- Domestic semiconductor equipment and materials stocks - As memory makers' capacity expansion translates into actual orders, the benefits spread sequentially to upstream equipment and materials suppliers.
Investor Checkpoints
- Watch next quarter's capex guidance from hyperscalers (Amazon, Microsoft, Google, Meta) to see whether the pace of AI infrastructure investment holds up.
- Yield rates for SK Hynix's and Samsung Electronics' 12-layer-and-above HBM products, and the timeline for customer qualification.
- Track DRAM spot and contract price trends to see whether the capex expansion is actually converting into higher prices.
- The launch schedule for Apple's new products featuring on-device AI features is another variable that could change the narrative of being left behind.
Outlook
The optimistic scenario is one where hyperscalers' capex expansion continues for several quarters, lifting memory utilization rates and prices together. In that case, the earnings improvement in Samsung Electronics' and SK Hynix's memory divisions could last longer than expected. The risk to watch on the other side is valuation. With expectations for AI infrastructure benefits already largely priced in, any slowdown in capex guidance — even a slight one — could trigger a sharp pullback. There is also a chance that HBM yields fall short of expectations, preventing the effects of capacity expansion from translating well into margins.
This article was automatically summarized and analyzed based on the original news report. View original (Maeil Business Newspaper, Securities)
