AI infrastructure spending is creating an unusual situation in the semiconductor industry. Memory, traditionally viewed as a cyclical business vulnerable to oversupply, is becoming a critical part of the AI investment story.
SK Hynix’s reported third-quarter revenue of ₩104.6 trillion and operating profit of ₩83.3 trillion highlight the extraordinary profitability described in the latest market update. Seagate also delivered strong results, reporting $4.23 billion in revenue and $1.5 billion in free cash flow, while Teradyne exceeded expectations.
The rally is no longer just about GPUs. It increasingly includes the memory, storage and testing equipment required to support expanding AI infrastructure.
But record results create a difficult question for investors: Are these companies benefiting from a lasting structural shift, or are we approaching the most profitable stage of another semiconductor cycle?
1. HBM: The strongest AI connection, but expectations are high
High-bandwidth memory (HBM) has become an important component of advanced AI accelerators. As models grow and computing workloads become more demanding, processors need faster access to increasingly large amounts of data.
SK Hynix has established a strong position in this market, and expanding HBM4 shipments could support its next stage of growth.
Unlike conventional memory products, HBM involves advanced packaging and demanding technical requirements. These factors can create barriers to entry and help suppliers maintain pricing power when demand exceeds available capacity.
However, there is a risk in extrapolating today’s profitability too far into the future.
High margins attract competition. As manufacturers invest in additional production capacity and competitors improve their technology, supply conditions can change.
The question is whether SK Hynix can maintain its technological advantage while continuing to grow earnings as the market expands.
For investors, I would watch HBM demand, customer commitments, production yields and capital expenditure. These indicators may provide a better picture of future profitability than record quarterly earnings alone.
2. NAND: The recovery could offer a different opportunity
NAND flash memory serves a different market, including solid-state drives, enterprise storage and consumer electronics.
The upcoming SanDisk earnings report could provide useful evidence about whether improving conditions are spreading beyond specialised AI memory.
Enterprise SSD demand is worth watching because AI systems need fast access to data, not just powerful processors. As businesses expand AI infrastructure, storage performance and capacity become increasingly important.
Nevertheless, NAND is exposed to its own supply cycle.
When demand strengthens and manufacturers maintain production discipline, prices and margins can improve. But if producers expand capacity too aggressively, additional supply can eventually weaken pricing.
This makes NAND an interesting but different investment proposition from HBM.
The key question is whether improving demand can support sustainable profitability rather than a temporary rebound in memory prices.
3. Hard drives: The less obvious beneficiary of AI
Seagate’s results highlight another part of the infrastructure chain that can benefit from AI investment.
Training AI models requires enormous datasets, while businesses deploying AI applications generate and retain additional information. Much of this data needs to be stored economically over long periods.
High-capacity hard drives remain relevant because they can provide substantial storage capacity at a lower cost per unit of data than many alternatives.
This creates a different opportunity from HBM.
HBM supports high-speed computing, while hard drives help address the growing volume of information that data centres must retain.
Seagate’s reported $1.5 billion in free cash flow is particularly relevant because cash generation helps demonstrate whether strong operating performance is translating into financial flexibility.
However, storage demand is not immune to cyclical pressures. Investors still need to assess pricing, product demand, production costs and the sustainability of current earnings.
The less obvious AI beneficiaries may offer opportunities, but they still need to be evaluated on their own financial merits.
4. The biggest risk: Supply eventually catches up
Semiconductor memory has historically experienced cycles in which strong demand and rising prices encourage manufacturers to expand capacity.
Initially, this can produce exceptional earnings growth. But as new production comes online, the balance between supply and demand can change.
The consequences can be significant because memory manufacturers have substantial fixed costs. When selling prices weaken, profits can decline faster than revenue.
AI may change the industry’s demand profile, particularly for specialised HBM products, but it does not eliminate the risk of overinvestment.
Investors should therefore monitor whether capacity additions are growing faster than demand, whether inventory levels are increasing and whether management teams are becoming more aggressive with capital expenditure.
Record profits are encouraging, but the durability of those profits matters more when assessing a company’s long-term value.
5. Which part of the memory chain offers the clearest earnings support?
The three areas have different strengths.
HBM: Strong exposure to advanced AI computing, with technological barriers that may support margins. The main risks are competition, high expectations and the possibility that supply expands faster than demand.
NAND: Exposure to enterprise SSDs and broader flash-memory demand. A recovery in pricing could support earnings, but profitability remains sensitive to supply discipline.
High-capacity hard drives: Exposure to growing data-storage requirements, with free cash flow providing an important measure of financial performance. However, demand and pricing still need to remain supportive.
There is no guarantee that the strongest industry segment will produce the best stock returns. A company can deliver excellent results and still disappoint investors if its share price already reflects even better performance.
That is why valuation and forward guidance deserve as much attention as recent earnings.
My take
The most important development is that AI infrastructure is expanding demand across several parts of the semiconductor supply chain.
SK Hynix illustrates the potential of specialised memory. SanDisk’s upcoming results could help clarify the strength of the NAND market, while Seagate demonstrates why data storage deserves attention alongside computing power.
I would not assume that every memory-related stock will benefit equally. The businesses have different competitive advantages, supply conditions and profit drivers.
For the next phase of the rally, I would focus on three things: sustainable margins, disciplined capacity expansion and free cash flow.
The strongest companies may be those that can maintain pricing power and generate cash even when industry conditions become less favourable.
The real test of the AI memory boom is not whether companies can deliver record profits at the top of the cycle. It is whether they can protect their earnings when the cycle eventually turns.
For discussion purposes only. This is not financial advice.
Comments