[Industry Insights] Why Morgan Stanley Warned to Reduce Memory Exposure Despite DRAM Prices Soaring

Recently, the global semiconductor market is enjoying what appears to be an “unprecedented boom” on the surface. Not only general-purpose DRAM (DDR4), but also the prices of next-generation DRAM (DDR5) and high-bandwidth memory (HBM) used in artificial intelligence (AI) servers have been rising day after day. That is also reflected by the fact that Samsung Electronics, a leading domestic semiconductor company, recorded an earnings surprise this year—breaking through 89 trillion won in operating profit in the second quarter.

Amid the recent market excitement, global investment bank Morgan Stanley released a report that poured cold water on it. It recommended reducing (Underweight) the weighting of the three global memory semiconductor companies—Samsung Electronics, SK hynix, and Micron. With DRAM prices climbing and earnings reaching their peak, this article provides an in-depth analysis of the background and key reasons behind why Morgan Stanley sounded such a contrarian warning.

Spot prices are an illusion… Fear of a “Peak-out” in earnings momentum

The most fundamental reason Morgan Stanley recommended cutting the weight of memory semiconductor stocks is “the slowing pace of upward revisions to earnings estimates”—that is, concerns about passing the peak (Peak-out) of momentum.

Stock prices respond more sensitively to future rates of change than to current earnings. Over the past 1 to 2 years, memory semiconductor companies’ stock prices could rise sharply because earnings forecast outlooks were revised upward explosively every month and every quarter. But recently, the upward trajectory of these earnings estimates has begun to flatten rapidly.

Although DRAM prices remain at high levels and may rise a little further, the analysis is that the “scale of price increases” itself is slowing. In addition, because last year already delivered such a steep earnings rebound, it has been pointed out that the hurdle for generating additional earnings surprises going forward has become too high.

In the end, Morgan Stanley interpreted today’s solid DRAM prices not as the “continuation” of the boom, but, from the perspective of the stock market, as the “last spark” or a signal of a peak.

A shift in leading stocks: from hardware (HW) manufacturers to hyperscalers (Cloud)

In this report, Morgan Stanley diagnosed that the narrow rally centered on semiconductors is entering its final phase, and 전망ed that the center of gravity of asset allocation should move from memory semiconductors to “big-tech hyperscalers (Hyper-scaler·large-scale data center operators).”

It is argued that the leading companies that previously saw margins explode by supplying essential inputs in the early stage of building AI infrastructure—such as Samsung Electronics, SK hynix, and Micron (hardware infrastructure)—will be replaced by new leading companies that have entered the stage of service enhancement and monetization using infrastructure: cloud and service companies such as Microsoft, Amazon (AWS), and Alphabet.

In other words, as the AI paradigm evolves from the “infrastructure-building stage” to a stage where it “makes money through the infrastructure that has been built,” the main beneficiaries will change. The logic is that big-tech companies such as Amazon and Microsoft, which previously bought Nvidia chips and DRAM to build data centers, are now at the timing to create genuine AI value through their own cloud services and recover margins. By contrast, it is believed that infrastructure hardware suppliers may be the first to be exposed to oversupply risk.

The theory of AI investment pace adjustment and the risk of oversupply

The key driving force behind the rise in DRAM prices, including HBM, was demand for AI servers. However, a so-called “pace adjustment theory” is gaining attention—suggesting that global big-tech companies’ AI investments may start to catch their breath.

Even though big-tech firms poured astronomical amounts of money into building AI data centers, if a clear revenue model (ROI—return on investment) does not become visible, the speed of investment execution inevitably has to slow down due to pressure from shareholders.

At the same time, the production capacity of memory semiconductor manufacturers is expanding rapidly. Samsung Electronics, SK hynix, and Micron have carried out record-breaking levels of capital expenditure to expand their HBM and high-performance DDR5 production capacity (CAPA), and they are also setting up astronomical investment plans. Morgan Stanley’s warning is that while demand growth may dip even slightly, once supply volumes begin to pour into the market in earnest, the market could quickly flip into an oversupply (Over-supply) phase.

The trauma of the past “Memory Winter” report and the market’s viewpoint

Morgan Stanley has a precedent: in August 2021, it released a report titled “Memory, Winter is Coming,” which had the effect of sharply driving down domestic semiconductor stocks. At that time, DRAM prices were also solid, but as Morgan Stanley predicted, the semiconductor downcycle became reality a few months later, giving their analysis significant influence in the market.

Of course, among domestic brokerage firms and experts, there are also no small number of arguments that Morgan Stanley’s warning is somewhat excessive. First, they point to HBM’s unique structure. Unlike past general-purpose DRAM cycles, current HBM is said to be closer to a “made-to-order product,” because customers sign supply contracts with it 1 to 2 years in advance, making a sudden surge in oversupply difficult to come by. Next, there are analyses that demand for high-performance SSDs based on PCIe 6.0 is newly taking off in the enterprise market, and that DRAM also has relatively strong downside rigidity because generation transitions are being carried out tightly.

In conclusion, the background behind Morgan Stanley’s call to “reduce memory” even amid the rise in DRAM prices appears to have originated from a view that goes beyond the illusion created by current earnings indicators and that needs to proactively respond to a slowdown in future growth rates and a shift in the leading-stock paradigm.

The future direction of the semiconductor market depends on whether big-tech firms succeed in monetizing their AI service revenues and on the memory industry’s disciplined supply management. For investors, it is time to keep a close eye not only on the current overwhelming operating profit numbers in the tens of trillions of won, but also on changes in demand in downstream industries and the broader flow of global capital.