Korea expands memory production, Meta rents out computing power—Nomura discusses "two major negatives for memory."

Nomura believes that panic-driven sell-offs due to emotional misjudgment are concealing structural opportunities, and the two major "negative factors"—South Korea's memory expansion and Meta renting out computing power—are actually pseudo-problems. Recently, the global memory chip market has been shrouded in two major "negative rumors": first, South Korean memory giants announced massive expansion plans, sparking deep market fears of future overcapacity; second, Meta declared it would rent out idle computing power externally, which some investors interpreted as a warning signal that AI hardware demand may have peaked. However, according to Wind Trading Desk, Nomura Securities stated in a July 2nd research report that market concerns have been seriously exaggerated. The truth is: - Korea’s huge 48 trillion won investment plan will require at least 5 to 10 years to convert into actual capacity, meaning relief is far off, and high-profit HBM (High Bandwidth Memory) is squeezing out general memory capacity, causing the market to face severe supply shortages; - Meanwhile, Meta renting out computing power will not lessen hardware demand—instead, by lowering Token costs, it will trigger the “Jevons Paradox,” stimulating even greater incremental AI demand. Overall, Nomura believes that the key contradiction in the global memory industry remains a severe supply shortage, with AI-driven structural demand growth yet to peak; investors’ concerns about overcapacity are understandable but clearly excessive, and the market’s overreaction may provide a window to reassess valuations in the memory sector. South Korea's Massive Expansion Plan: Relief Is Distant, Overcapacity Fears Are Overblown Recently, Korean memory companies and their affiliates, in conjunction with the government, announced a large-scale mid-to-long-term investment plan with no clear schedule, totaling as much as 4.8 trillion won (of which 3.7 trillion won is directly related to memory). This huge figure quickly intensified investor concerns about oversupply in the memory chip market. But Nomura points out that the so-called "global memory companies colluding to control supply and manipulate prices" conspiracy and worries about overcapacity are baseless. - First, the current situation is an extremely severe supply shortage, not surplus. Faced with unprecedentedly strong demand from the AI industry, memory companies have no choice but to prioritize producing high-profit HBM chips. This capacity tilt directly causes a slowdown in growth of general memory chip production. Since the second half of 2025, strong growth in general memory demand has triggered a serious supply shortage. Despite memory companies aggressively expanding capacity beyond expectations, they still cannot meet the massive market demand. - Second, converting semiconductor investment into actual capacity is an extremely lengthy process. Government intervention is mainly due to the fact that companies’ existing production clusters are about to exhaust their carrying capacity (land, electricity, water), and the government needs to support the construction of new clusters for the mid-to-long term after 2035. For example, the “Yongin Semiconductor Cluster” mega-project that started nine years ago is not expected to complete its first cleanroom until February 2027, with small-scale production beginning at the end of that year—meaning it actually takes more than ten years from investment to production. Nomura expects that the newly announced investment plan will have its first substantial impact on the market in no less than five to ten years. - Finally, the industry's risk-resistance mechanism has undergone structural change. In the past, cyclical fluctuations in the memory industry often stemmed from insufficient investment during downturns or record-breaking investment during demand booms. Now, companies not only have Long-Term Agreements (LTAs) as hedging tools, but also enjoy structurally stable growth expectations driven by AI. Additionally, employee bonuses tied to profits serve as new buffers against risks of overcapacity and profit decline. Companies will never engage in unnecessary blind investment just because of government requirements. Meta Renting Idle Computing Power: Not a Peak in Demand, but Raising Capital Returns à la AWS The second market concern comes from Meta’s decision to sell its excess computing power to external customers, which some see as a precursor to weakened demand for AI memory and hardware. Nomura clearly refutes this, seeing it as merely a natural evolution toward a mature business model, similar to how Amazon once launched AWS cloud services to monetize idle datacenter capacity. - First, renting out computing power is the inevitable solution to "peak redundancy." Datacenter construction is inherently designed to match "peak computing power" demand, meaning there is a lot of idle capacity during off-peak periods and seasons. Meta’s main businesses (social networking and advertising) experience huge fluctuations in computing power utilization at different times. Among cloud service providers (CSP) who use datacenters for both internal and external purposes, Meta is the only company yet to get involved in cloud business. As economies of scale develop, following xAI’s example of selling surplus computing power externally is an extremely natural decision for improving Meta’s ROIC. Otherwise, as capacity expands, it would be a huge waste of resources. - Second, the released computing power will nurture a larger AI ecosystem. The computing power sold externally by Meta will become a valuable resource for companies like Anthropic and OpenAI, which lack their own datacenters but urgently need computing power to provide enterprise-level AI services. - Third, triggering "Jevons Paradox" and creating more incremental demand. Nomura emphasizes that Meta’s move is definitely not a turning point for declining AI-related hardware demand. On the contrary, due to current computing power shortages leading to rising single-Token prices, Meta's entry into the market will likely stabilize Token prices downward. According to the "Jevons Paradox" (technological advancement reduces resource usage costs, but ends up increasing overall resource consumption), lower usage costs will create entirely new, larger-scale AI demand, thereby further solidifying underlying demand for memory and computing hardware in the long run. Risk Warning and Disclaimer The market carries risks and investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users' unique investment objectives, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their specific circumstances. Invest accordingly at your own risk.