A curious situation is unfolding in the memory semiconductor market: despite securing long-term supply deals fueled by the AI boom, major players like Samsung and SK hynix are trading at surprisingly low valuations.
Some investors have started to argue that the business is transforming into a stable, subscription-like model, similar to SaaS (Software as a Service). The logic is that multi-year Long-Term Agreements (LTAs) with major cloud companies, supposedly running "to 2030," should guarantee predictable revenue and profits, justifying much higher stock price multiples.
However, the reality is more nuanced. While multi-year deals are indeed happening, they are typically for 3 to 5 years, not a decade. Both Samsung and Micron have confirmed this shift, but they've also clarified a crucial detail: pricing is not fixed. Instead, it operates within a band that is still anchored to fluctuating spot market prices. This provides companies with better volume visibility—they know how much they need to produce—but it doesn't give them complete control over pricing.
So, why does the market remain so cautious? The primary reason is that investors believe we are at the peak of a cycle. First, the current record-breaking earnings are driven by an extreme shortage. The AI frenzy pulled a vast amount of production capacity toward specialized HBM (High Bandwidth Memory), creating a crunch for conventional DRAM and NAND. This allowed suppliers to hike prices dramatically in late 2025. But the market anticipates that new factories and production lines will eventually come online, increasing supply and normalizing prices.
Second, there's significant buyer concentration risk. A handful of tech giants, like Nvidia and major hyperscalers, account for a massive portion of demand. While they are currently competing for limited supply, their immense scale gives them powerful leverage in negotiating future contracts. They can diversify suppliers or even develop their own chip designs, like Nvidia's 2027 logic-die strategy, to prevent any single memory maker from gaining too much power.
Finally, other factors keep a lid on valuations. The massive capital expenditure required to build new HBM facilities can strain free cash flow, even when profits are high. Lingering geopolitical risks related to US-China tech policies and operational issues like potential labor strikes also add to investor uncertainty.
In conclusion, while the shift to multi-year agreements is a positive step toward stability, it hasn't fundamentally erased the cyclical nature of the memory industry. The market is pricing these stocks based on this reality, acknowledging improved earnings visibility but stopping short of awarding them the premium valuation of a true SaaS business.
- Forward P/E (Price-to-Earnings) Ratio: A stock valuation metric that uses projected future earnings to determine how expensive a stock is. A low P/E can suggest a stock is undervalued or that investors expect earnings to decline.
- LTA (Long-Term Agreement): A contract between a supplier and customer that spans multiple years, designed to secure a stable supply of goods at predictable, though often flexible, prices.
- HBM (High Bandwidth Memory): A high-performance type of memory chip stacked vertically to provide faster data transfer speeds, essential for training and running large AI models.