Analysis
A lot of people have been asking me lately: after this memory run-up, this selloff, and then all the back-and-forth chopping, is the rally finally over?
The recent phase is this: the compute premium is gradually being passed down into the mid-layer of hardware—though some are hoping it will spread to software as well.
NVIDIA's earnings already told us this: massive storage demand from inference, large-model training, and supercomputing expansion is bound to exceed expectations. But the margin path—non-GAAP gross margin gradually declining from 75%—is expected to bottom in Q4'27 at 71%–72%, then recover to 72%–73% in FY2028, still below the 75% peak and not fully back to previous highs.
Market demand hasn't collapsed—what's squeezing the giants' margins is the sharp rise in memory costs (HBM/DRAM). NVIDIA management also explicitly attributed the margin pressure to rising memory prices on the earnings call.
NVIDIA gives, memory takes—this profit-shift cycle runs at least through the end of 2027.
So memory is still in a real earnings upcycle, and the boom is far from over.
Plenty of people treat memory as one monolithic group and mindlessly go all bullish or all bearish on it. That definitely isn’t the whole story.
✅ Let’s first clear up a common misconception: both SK Hynix and Micron have multi-year supply agreements (SKHY has also signed multi-year framework deals with NVIDIA and others). For both companies, these agreements provide revenue visibility but also pricing constraints—it’s not that SKHY has unlimited freedom while Micron is locked down.
The real difference is this:
1. SK Hynix: the HBM leader, capturing the AI windfall
Based on UBS market-share assumptions for 2027, HBM market share is SKHY≈60%, Samsung≈30%, Micron≈10%. SKHY is the core supplier in high-bandwidth memory; nearly all incremental AI training/supercomputing demand passes through it first.
Its business mix is tightly aligned with high-bandwidth AI demand. During pricing upcycles, the HBM premium shows up directly in earnings leverage.
On top of that, two more massive buffs:
✅ a $28.6B epic buyback, putting a real floor under the stock and boosting shareholder returns;
✅ on 2027 revenue expectations, its valuation is still significantly below Micron’s.
High growth + low valuation + massive buyback + HBM leadership: those four buffs are fully stacked.
2. Micron: low HBM share, the smallest slice of the AI windfall
Micron’s core business is DRAM + NAND, with HBM share of only about 10%–21%. It captures the smallest incremental share of the structural AI memory boom.
The current industry backdrop: DRAM prices continue to rise; NAND prices have stabilized but lack upward momentum. Because Micron has a high mix of traditional business, its natural pricing leverage is weaker than SKHY’s.
On top of that, China’s domestic ChangXin and Yangtze Memory are continuing to ramp capacity, putting pricing pressure on Samsung, SKHY, and Micron—and as the smaller player, Micron feels that pressure even more.
Let me show you a set of FY2027E valuation numbers. Note these are forward consensus midpoints, with wide sell-side dispersion—not realized results:
2027E revenue (consensus midpoint) ≈ $340B (range $330–365B)
Current market cap / EV ≈ $1.0T / $0.98T
Forward EV/Sales ≈ 2.4–2.8x
Micron
2027E revenue (consensus midpoint) ≈ $210B (range $190–240B)
Current market cap / EV ≈ $1.1T / $1.1T
Forward EV/Sales ≈ 4.1–5.1x
Even after this big rally, SK Hynix’s forward EV/Sales is still roughly 40%–45% lower than Micron’s.
The key point: the cheaper SK Hynix is actually growing faster, has a better end-market position, and offers more upside.
⚠️ But two caveats need to be clear:
Micron itself is the highest-quality memory name in U.S. equities. Its fundamentals are not weak, long-term agreements give it revenue visibility, and earnings elasticity is still there.
But in the current cycle, it is less well suited to the tape than SK Hynix:
Micron is a name for riding through cycles and defending the bottom;
SK Hynix is a name for capturing the upside, playing the beta, and harvesting the AI tailwind.
Different market phases favor different names.
Samsung grabbing HBM share: If Samsung challenges for the HBM leadership position in 2027 (UBS already assumes Samsung gets a 30% share), SK Hynix’s strongest thesis would loosen directly, and the rally would end in stages.
AI long-term demand weakening: If Nvidia’s large volume of “contractually locked-in order demand” for next year unwinds early, it could cause a phased stampede.
But for now, both are potential risks.
Key point: Micron earnings on September 30 (company confirmed: Q4 FY2026 earnings call on 2026/9/30 after market close).
In the second half of the AI trade, the excess returns are in memory and equipment, not in the compute giants;
Memory is a structural trade: the divergence of SK Hynix strong / Micron weak persists into 2027. The difference is in HBM share and product mix, not in “whether you have long-term agreements”;
Valuation gap, HBM track, buyback catalyst, and rising price cycle — these four logics support SKHY continuing to strengthen;
Near term, stay on the sidelines and wait for Micron’s earnings report; once it lands, reassess.