To The Moon

Asian semiconductor stocks suffered a sharp selloff today. SK hynix dropped roughly 10%, Samsung Electronics fell 7.4%, South Korea’s KOSPI declined more than 5%, and Japan’s Kioxia also saw double-digit losses. What makes this interesting is that AI memory demand has not suddenly disappeared. Instead, investors are reacting to a potential shift in the global memory supply landscape, driven partly by a technological breakthrough from China.

On October 12, Chinese memory manufacturer CXMT (ChangXin Memory Technologies) disclosed progress in its next-generation 4F² DRAM cell architecture, with plans to introduce server memory products using the technology by the end of this year. Compared with the industry’s mainstream 6F² structure, the new architecture could potentially increase memory density and reduce cost per bit. Samsung and SK hynix are also developing similar technologies, but CXMT’s progress has prompted investors to reconsider the competitive outlook for global DRAM.

The real issue is not whether CXMT can suddenly take market share away from Samsung or SK hynix tomorrow.

It is that one of the most important assumptions supporting the memory bull market — limited supply — may eventually face a challenge.

The investment thesis behind the AI memory rally has been straightforward. AI servers require more HBM and server DRAM, advanced manufacturing capacity remains constrained, and hyperscalers are locking in supply through long-term agreements. Strong demand combined with disciplined supply has given memory manufacturers exceptional pricing power, driving earnings sharply higher and supporting premium valuations.

But CXMT’s latest development challenges a key part of that equation: if new competitors can eventually produce higher-density memory at lower costs, how long can today’s elevated margins and pricing power last?

This is closely connected to the recent debate around Michael Burry’s bearish position on Micron and Toshiba’s plans to expand HDD capacity. AI demand can remain strong, but high industry profits naturally encourage technological improvements and new capacity investment. In a cyclical industry like memory, rising demand does not guarantee permanently rising prices. What ultimately matters is whether demand growth can continue outpacing effective supply growth.

Still, there is an important distinction investors should not ignore.

A technological breakthrough is not the same as successful mass production. CXMT’s reported progress does not yet establish commercial yields, actual manufacturing costs, product reliability or hyperscaler qualification. Moving from an advanced architecture to stable, large-scale production requires significant execution, particularly for server memory products where reliability and performance are critical.

In fact, the near-term memory market remains relatively tight. TrendForce previously projected that conventional DRAM contract prices could rise another 10%–15% quarter over quarter in Q4 2026, while NAND Flash contract prices could increase 15%–20%. AI servers continue to absorb substantial memory capacity, and CXMT’s announcement does not immediately change that supply-demand balance.

That suggests today's selloff may be more about repricing future competition and earnings risk than confirming an immediate end to the memory upcycle.

There is another development worth connecting to this story. Broadcom CEO Hock Tan recently suggested that as wafer and memory supply bottlenecks gradually ease, capital could become the biggest constraint on continued AI expansion. Together, these developments point toward a potential change in the AI investment narrative.

Previously, the market's concern was:

AI demand is so strong that there aren't enough chips or memory modules.

Now investors are beginning to ask:

What happens if supply catches up, while financing the next wave of AI infrastructure becomes increasingly expensive?

Tiger View

Tiger thinks today’s selloff in SK hynix and Samsung does not automatically mean AI memory demand has peaked. But it does highlight a changing competitive landscape.

Memory competition may increasingly shift from who controls scarce capacity to who can deliver more usable capacity at a lower cost.

AI has fundamentally expanded the scale of memory demand, but it has not eliminated the industry's cyclical nature. Even if HBM and server DRAM demand continue growing, new manufacturing technologies, additional competitors and capacity expansion could eventually weaken the pricing environment.

Tiger would watch three signals next: whether CXMT can move its 4F² architecture into stable mass production and secure server customer qualifications, whether DRAM spot and contract prices begin showing sustained weakness, and whether Samsung, SK hynix and Micron start adjusting their long-term order or margin expectations in response to new competition.

If the next phase looks like:

Technology advances → production ramps slowly → AI demand absorbs new supply → DRAM pricing stays firm,

then today's correction may prove to be primarily a valuation adjustment after a powerful rally.

But if the sequence becomes:

Technology advances → effective capacity expands rapidly → customer bargaining power improves → DRAM prices and margins decline,

then the supply risks investors are worried about today could begin translating into actual earnings pressure.

So Tiger’s key question is not simply:

“Can CXMT catch up with Samsung and SK hynix?”

It is:

“If AI memory becomes less scarce, how much of a premium will investors still pay for memory stocks?”

Related Stocks

DRAM / HBM: $Micron Technology(MU)$

Watch: whether new competitors and manufacturing technologies challenge future DRAM pricing power, and whether long-term HBM contracts can continue supporting earnings.

Korean Memory Leaders: SK hynix, Samsung Electronics

Watch: server DRAM and HBM product mix, next-generation manufacturing technology, and the sustainability of elevated margins.

AI Compute: $NVIDIA(NVDA)$

Watch: how changing memory costs affect overall AI server economics and whether hyperscaler capital spending continues expanding.

Custom AI Silicon: $Broadcom(AVGO)$

Watch: AI infrastructure financing constraints, ASIC order execution and long-term returns on hyperscaler investment.

Semiconductors: $iShares Semiconductor ETF(SOXX)$

Watch: whether valuation pressure in memory stocks spreads to the broader AI chip and semiconductor equipment sectors.

Today’s Poll

SK hynix plunges 10% as CXMT advances its memory technology. What's your take on the AI memory cycle?

① AI demand remains strong — this is mostly a sentiment-driven correction

② CXMT and other competitors are catching up, threatening memory pricing power

③ Earnings are still strong, but the market is already pricing in a cycle reversal

④ Wait for actual production, pricing and order data before making a call

For market discussion only. This is not investment advice. Markets involve risk, and investment decisions should be made carefully.

Sources: Reuters (October 12–13, 2026), South China Morning Post (October 12), and TrendForce (September 30). Stock moves refer to the October 13 Asian trading session.