Memory stocks have come under pressure in recent weeks, caught in a broader semiconductor sell-off driven by profit-taking, macro uncertainty, and a fresh wave of concern over Chinese DRAM competition. Micron has pulled back 18% since October 1st, SanDisk is off by 23.62% over the same period, whist Samsung Electronics are down 19%, and SK Hynix 29%. It has been a tough few weeks, but analysts continue to look through the volatility.
Nomura has reiterated Buy ratings on Samsung Electronics and SK Hynix with price targets implying more than 200% upside, arguing that markets are significantly underestimating both the durability of AI-driven memory demand and the earnings power it will generate through at least 2028.
Samsung Electronics shares trade at KRW 223,500, well below Nomura’s KRW 670,000 target, while SK Hynix has similarly lagged, with both names seeing estimated forward price-to-earnings multiples compress to around 3x on 2027 consensus earnings, a level Nomura describes as historically cheap for companies at the centre of one of the most powerful structural demand cycles in the industry’s history. The target on SK Hynix at KRW 4,700,000 points to a perceived upside of 263% from here.
Why So Bullish on AI?
At the heart of Nomura’s thesis is a straightforward supply-demand argument. AI workloads are shifting rapidly from model training to inference at scale, and that transition is far more memory-intensive than markets currently appreciate. Data centre capacity is projected to expand 41% over the next year alone, rising from 22.3 gigawatts to 31.5 gigawatts globally, and the memory required to service that buildout cannot be manufactured quickly enough to close the supply gap before 2028 at the earliest. That structural tightness, Nomura argues, will sustain pricing power and drive extraordinary free cash flow generation across the memory value chain for years.
The numbers already bear this out. Net cash across nine major semiconductor manufacturers has surged from US$27 billion in the first quarter of 2024 to US$282 billion in the second quarter of 2026, and Nomura projects that figure will approach US$2 trillion by the end of 2028. That accumulation reflects both the pricing environment and the operating leverage inherent in leading-edge memory production at scale.
Top Australian Brokers
- Pepperstone - Top Australian broker - Read our review
- eToro - Invest in ASX and international shares - Read our review
On the demand side, Nomura’s estimates for the two largest frontier AI laboratories reinforce the case. Combined revenue for OpenAI and Anthropic is estimated to have reached US$27.3 billion in the third quarter of 2026, representing year-on-year growth of 507%. A temporary slowdown in the sequential growth rate has been attributed to partner accounting adjustments, enterprise customer migration patterns, and summer seasonality rather than any structural weakening in end demand. Anthropic’s achievement of adjusted operating profitability in the second quarter of 2026 adds further weight to the argument that frontier AI business models are maturing faster than the bears anticipated.
Cloud hyperscalers, meanwhile, reported healthy operating margins in the June quarter: 39% at Amazon Web Services, 41% at Microsoft Intelligent Cloud, and 36% at Google Cloud. These figures suggest the infrastructure layer of the AI economy remains financially robust, even as third-party data centre operators and pure-play AI providers accumulate debt at an accelerating pace.
What Lay Ahead
The most recent catalyst for weakness in memory stocks was an announcement from China’s ChangXin Memory Technologies, known as CXMT, regarding progress on a 4F-squared cell architecture and plans to release DDR5 RDIMMs. The news rattled markets across South Korea and the West, reviving fears that Chinese domestic DRAM production could erode pricing and market share for incumbents.
Nomura pushed back firmly on that interpretation, characterising the reaction as excessive. The gap between announcing architectural progress and achieving volume manufacturing at commercially competitive yields is substantial, and critically, CXMT has no credible path into High-Bandwidth Memory, the segment where Samsung and SK Hynix generate their most valuable margins and where the AI compute buildout is most acutely concentrated.
Whether markets are willing to re-rate memory stocks ahead of confirmed earnings delivery in the fourth quarter remains the central question. Nomura’s position is clear: the discount is unwarranted, the supply constraints are structural, and the earnings power being built inside Samsung and SK Hynix is not yet reflected in their share prices.
Bull Case:
- Memory supply constraints are structural and will persist through at least 2028.
- Samsung and SK Hynix trade at historically cheap 3x forward earnings multiples.
- AI lab revenue surged 507% year-on-year, confirming real enterprise demand.
- Chipmaker net cash is on track to reach US$2 trillion by end-2028.
- CXMT poses no credible threat to High-Bandwidth Memory profitability.
Bear Case:
- OpenAI and Anthropic need 61% sequential revenue growth in Q4 to meet consensus.
- Rising leverage among data centre operators risks slowing infrastructure buildout.
- CXMT’s architectural progress could eventually pressure commodity DRAM pricing.
- Macro uncertainty and profit-taking continue to suppress near-term sentiment.
- Any AI monetisation shortfall could trigger renewed capital expenditure scrutiny.
