J.P. Morgan frontline investigation: Storage plummets, what is the core contradiction?

```

The Asian memory chip sector has undergone its most dramatic correction so far this year. Since the peak in June, major Asian memory stocks have collectively declined by about 30%, far exceeding the roughly 11% drop in the Philadelphia Semiconductor Index over the same period.

After JPMorgan recently interviewed more than 50 institutional investors in Hong Kong, it found that the market is not turning bearish on the memory industry, but is instead entering a new stage—from "AI infrastructure expansion" to "profit realization validation." The core of the current share price adjustment is not due to vanishing demand, but rather a clear divergence in market expectations regarding cloud service providers' capital expenditure, HBM prices, and future profit margins.

On July 14, JPMorgan analyst Jay Kwon released a report stating, about 70% of current market sentiment revolves around one variable—whether hyperscale cloud service providers (CSPs) can continue to significantly revise up their future capital expenditures. If the growth in capital expenditures cannot continue to exceed expectations, the memory sector may still face pressure in the short term.

What exactly is the market worried about?

JPMorgan believes that this round of memory stock corrections stems from three main factors.

First, the expectations for AI investment have run ahead of reality. Previously, the market continuously raised expectations for AI data center investment, and at the same time kept revising upwards the total addressable market (TAM) for the memory industry. But as stock prices surged, investors started to worry that sell-side forecasts for future demand were ahead of the actual capital expenditure plans of cloud providers.

Currently, most investors expect that over the next 3 to 6 months, global hyperscale cloud providers’ capital expenditures may further rise to $1 trillion to $1.5 trillion. If future earnings reports cannot validate this optimistic expectation, market sentiment may remain under pressure.

Second, the rise in DRAM prices has begun to slow down. The report notes that after consecutive price increases, year-over-year and quarter-over-quarter DRAM price growth both began to moderate after the second quarter of 2026, leading to cooled expectations for continued rapid profit expansion for the industry.

Third, Samsung Electronics' profit expectations have been revised downward ahead of time. Before the release of its Q2 financial report, the market had already lowered its forecast for Samsung’s operating profits, further dampening investor sentiment. JPMorgan believes the market’s focus has now shifted from “how fast the industry can grow” to “how long the current profit level can be maintained.”

Is LTA really a good thing?

Long-term agreements (LTA) have become one of the most-discussed topics in this round of roadshows.

Compared to a few months ago, investors' attitudes towards LTAs have improved significantly. The focus of discussion has gradually shifted from "whether there are LTAs" to "how manufacturers can use LTAs to bind core AI customers." However, market divergence still exists. More than half of surveyed investors remain cautious, mainly because the LTA coverage ratio of Korean manufacturers is still unclear and the quality of contracts is difficult to compare horizontally between companies.

JPMorgan expects that more than half of contract volumes will eventually fall under the LTA framework. More importantly, it believes that LTAs do not limit the space for future price increases. On one hand, some new orders can still be re-signed at higher prices in the future; on the other hand, the "take-or-pay" clause in contracts already provides certainty for orders, and prices for products not included in LTAs will still rise as supply and demand tightens.

Therefore, in JPMorgan’s view, an LTA is more of a tool for improving profit stability rather than a constraint on profit flexibility.

HBM prices: the biggest gap in expectations right now

If LTAs are about profit stability, HBM pricing directly determines profit flexibility.

JPMorgan says that currently, the biggest divergence between buyers and sellers surrounds HBM pricing. Most buy-side institutions expect that by 2027, the price per GB of HBM will double year-on-year, and extrapolate continued significant profit upgrades based on this. But JPMorgan is clearly more cautious. The firm estimates that the current average selling price of HBM in the industry is about $1.8 per GB, even slightly lower than some high-end server DRAM products.

More importantly, when memory manufacturers negotiate with cloud providers, HBM is not discussed in isolation, but rather pricing is determined based on comprehensive profitability from DRAM, NAND, and HBM as a whole. Therefore, HBM prices may not rise indefinitely. JPMorgan expects that a 25% to 30% year-over-year increase in HBM ASP by 2027 is more in line with industry reality.

However, the firm also points out, compared with the usual 3-to-5-year LTAs for traditional DRAM, HBM is basically repriced annually, which means manufacturers still have considerable room to raise prices later if AI demand continues to exceed expectations.

DRAM remains the tightest; enterprise SSDs become a new highlight

From a supply and demand perspective, JPMorgan still maintains a relatively positive view on the memory industry.

Among these, DRAM remains the product with the tightest supply/demand balance. According to the report, current DRAM supply is only able to meet about 50% to 60% of order demand, while the ratio for NAND is about 70% to 80%. Even if DRAM wafer capacity continues to expand over the next few years, the firm expects the supply-demand tension may last until 2027–2028.

Compared with weak consumer electronics demand, enterprise storage is becoming a new growth engine. JPMorgan notes that consumer NAND demand downgrades have exceeded expectations, but AI data centers are driving constant upward revisions in enterprise SSD demand, especially as demand for AI use cases such as KV Cache Offload is growing much faster than previously projected.

Industry chains currently predict that enterprise SSD shipments could reach nearly 500EB by 2027, a YoY increase of nearly 50%, with further upgrades still possible. Investors generally expect that large North American cloud service providers are willing to pay $0.5 to $0.55 per GB for enterprise SSDs, which will also continue to support NAND prices in the future.

Overall, JPMorgan believes that the current adjustments in the memory sector primarily reflect a market revaluation of expectations, rather than a reversal in industry fundamentals. The key to the sector's future performance is no longer just whether AI demand remains strong, but whether cloud providers’ capital expenditures can be sustained, and whether HBM profitability can match the optimism already priced into stocks.

Risk Warning and DisclaimerThe market carries risks and investments require caution. This article does not constitute individual investment advice, nor does it take into account specific investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article suit their own specific circumstances. Investing based on this information is at your own risk. ```