Nomura believes the so-called 'two major headwinds for memory' are in fact a false narrative. South Korean investments will take several years to materialize, and HBM production is crowding out general-purpose capacity; the core global memory market remains severely supply-constrained. Meta’s move aims to enhance capital returns and reduce computing costs, which would instead stimulate even greater incremental AI demand. With AI demand yet to peak, unwarranted market pessimism has created a window for sector revaluation.
Nomura believes that misplaced market sentiment has obscured structural opportunities, and the two perceived headwinds—Korean memory makers’ capacity expansion and Meta leasing out idle computing power—are in fact non-issues.
Recently, the global memory chip market has been clouded by two alleged bearish rumors: first, major Korean memory manufacturers have announced massive capacity expansion plans, triggering deep concerns about future oversupply; second, Meta’s announcement to lease out idle computing power has been interpreted by some investors as a warning sign that demand for AI hardware may have peaked.
However, according to ZuiFeng Trading Desk, Nomura Securities stated outright in its July 2 research note that market concerns have been significantly exaggerated. The reality is as follows:
South Korea’s investment plan, totaling KRW 480 trillion, will take at least 5 to 10 years to translate into actual production capacity—a distant solution incapable of addressing near-term shortages—and the diversion of capacity toward high-margin HBM (High Bandwidth Memory) is exacerbating a severe supply shortfall in mainstream memory products;
Meanwhile, Meta leasing out computing power will not dampen hardware demand; on the contrary, by lowering token costs, it will trigger the ‘Jevons Paradox,’ thereby stimulating even greater incremental AI demand.
Overall, Nomura maintains that the core issue facing the global memory industry remains acute supply shortages, and AI-driven structural demand growth has yet to peak. While investor concerns about oversupply are understandable, they are clearly overblown, and the market’s excessive reaction may present a window to reassess valuations in the memory sector.
South Korea’s Massive Capacity Expansion Plan: A Distant Solution That Cannot Alleviate Near-Term Shortages; Oversupply Fears Are Significantly Overstated
Recently, South Korean memory manufacturers, their affiliates, and the government jointly announced a large-scale, medium- to long-term investment initiative with no clear timeline, totaling KRW 4.8 quadrillion (of which KRW 3,700 trillion is directly related to memory). This staggering figure quickly intensified investor fears of an oversupply in memory chips.
However, Nomura pointed out that conspiracy theories about 'global memory companies colluding to manipulate prices through supply control' and fears of overcapacity are entirely unfounded.
First and foremost, the market is currently facing an acute supply shortage—not oversupply.
Faced with unprecedentedly strong demand from the AI sector, memory manufacturers have no choice but to prioritize production of high-margin HBM chips. This shift in capacity allocation has directly slowed the growth of general-purpose memory chip output.
Since the second half of 2025, robust growth in demand for general-purpose memory has already triggered severe supply shortages. Despite memory makers aggressively expanding capacity at a pace far exceeding expectations, they still cannot meet the enormous market demand.
Second, the lead time for semiconductor investments to translate into actual production capacity is extremely long.
The South Korean government’s intervention stems primarily from the fact that existing corporate production clusters are nearing their capacity limits in terms of land, power, and water resources, necessitating government support for the development of new, medium- to long-term clusters beyond 2035.
Take the ‘Yongin Semiconductor Cluster,’ a mega-project launched nine years ago, as an example: its first cleanroom is not expected to be completed until February 2027, with limited-scale production only commencing by year-end—meaning it effectively takes more than a decade from investment to actual production.
Nomura estimates that the newly announced investment plan will not have a material impact on the market for at least another five to ten years.
Finally, the industry’s risk-mitigation mechanisms have undergone structural changes.
In the past, cyclical fluctuations in the memory industry typically resulted from underinvestment during downturns or record-breaking investments during demand surges. Today, however, companies not only use long-term agreements (LTAs) as hedging instruments but also benefit from the structurally stable growth outlook driven by AI.
Moreover, employee bonus schemes tied to profitability now serve as a new buffer against risks of overcapacity and declining margins. Companies will absolutely not make unnecessary, blind investments solely due to government directives.
Meta Leasing Idle Computing Power: Not a Sign of Peaking Demand, but Rather Emulating AWS to Enhance Capital Returns
The market’s second concern stems from Meta’s decision to sell its excess computing capacity to external customers, which some view as an early signal of weakening demand for AI memory and hardware.
Nomura explicitly refutes this, arguing that it is merely a natural evolution of a maturing business model—logically akin to Amazon’s creation of AWS cloud services years ago to monetize its underutilized data centers.
First, leasing computing capacity is an inevitable solution to 'peak redundancy.'
Data center construction inherently needs to align with 'peak computing' demand, meaning substantial computing capacity remains idle during off-peak hours and seasons. Meta’s core businesses—social networking and advertising—experience significant fluctuations in computing utilization across different times of day.
Among cloud service providers (CSPs) that use their data centers for both internal and external purposes, Meta is the only company that has not yet entered the cloud business. With economies of scale now established, following xAI’s example by selling surplus computing capacity externally is a highly logical move to enhance Meta’s return on invested capital (ROIC). Failing to do so would represent a significant waste of resources as capacity continues to expand.
Second, the released computing capacity will nourish a broader AI ecosystem.
Meta’s externally offered computing capacity will become a critical resource for companies like Anthropic and OpenAI, which lack their own data centers but are eager for computing power to deliver enterprise-grade AI services.
Third, it triggers the 'Jevons Paradox,' generating additional incremental demand.
Nomura emphasizes that Meta’s decision is by no means a turning point indicating reduced demand for AI-related hardware. On the contrary, given current computing shortages driving up per-token prices, Meta’s entry into the market is expected to help stabilize token prices downward.
According to the 'Jevons Paradox'—which posits that technological advances reducing the cost of resource usage ultimately increase total consumption of that resource—the reduction in usage costs will spur entirely new, large-scale AI demand, thereby further solidifying underlying demand for storage and computing hardware over the long term.
Looking to select stocks or analyze your holdings? Want to understand the opportunities and risks in your portfolio? For any investment-related questions,just ask Futubull AI!
Editor/KOKO