Fear is the biggest short! Morgan Stanley calls for bottom-fishing in semiconductors: land and electricity shortages are not a problem, approved projects can't meet demand

The semiconductor sector has recently faced a sell-off, but Morgan Stanley believes that market panic itself is the greatest source of risk, rather than industry fundamentals.

According to Trading Desk sources, Morgan Stanley analyst Joseph Moore clearly stated in a report that business visibility for AI infrastructure construction remains “exceptionally strong,” and communication with semiconductor manufacturers' clients indicates that this trend is sustainable. While constraints such as land, power, and shell (Land-Power-Shell, LPS) have raised market concerns, he pointed out that it is precisely these bottlenecks that make industry growth more stable and long-lasting. In terms of coverage, the bank noted that its S&P median stock has fallen 53% from its peak, and analysts believe the current sell-off has been excessive.

Morgan Stanley admits that policy uncertainty after the midterm elections and the potential impact of capital market sentiment on spending willingness are short-term suppressing factors, so the sector may exhibit a range-bound trend. However, the bank emphasizes that AI spenders remain resolute, with many cash-rich companies already locking in multi-year spending plans. Even emerging cloud service providers with higher financing costs are facing a shortage of computing power. There is a “never seen before” immense disconnect between the panic on Wall Street and the optimism in Silicon Valley, the bank notes.

Morgan Stanley gets straight to the point: “Our greatest fear is fear itself.”

The report highlights that orders from downstream semiconductor customers are not only not weakening but are actually accelerating—each day, a large number of new orders are flowing into already fully allocated backlogged capacity. Communication with spenders shows that multi-year spending intentions from cash-rich companies remain fully intact. Even emerging cloud service providers troubled by high interest rates are currently facing the core issue of not being able to acquire enough computing power resources within already approved spaces, rather than a lack of spending drive.

The bank also points out that the market's concern of “everything has to wait until 2028” is essentially no different from the past three years. At the end of 2023, as investors gradually adapted to 2024, the reason for stock pressure became “it’s all about 2025.” This cyclical anxiety about the next 12 to 24 months is typical in the semiconductor sector and does not indicate genuine deterioration in fundamentals.

From a demand perspective, Morgan Stanley cites data showing that total AI computing power token volume continues its steep growth, with total token spending rebounding to historical highs, and mainstream GPU rental prices remaining relatively strong.

Market concern regarding Land, Power, and Shell (LPS) constraints has risen since the midterm elections. Morgan Stanley acknowledges that deployment slowdowns in areas like Ohio and Texas do constitute localized disturbances, but sees clear market misinterpretation of the issue.

Communication with buyers reveals that LPS will not be the biggest challenge in 2027—the core reason being that there simply aren't enough semiconductors to fill the currently approved spaces. In other words, even if LPS issues are completely resolved, semiconductor supply itself will remain the limiting factor.

Meanwhile, the industry is quickly seeking LPS alternatives outside the US or in more policy-friendly states, and Morgan Stanley regards industry adaptability as a key variable. Their supply-demand model indicates that even by 2028, semiconductor supply will be the bottleneck for the entire ecosystem. Morgan Stanley also emphasizes that the existence of LPS constraints is actually a profound reason why AI infrastructure growth can remain continuous and stable.

Morgan Stanley ranked its covered semiconductor subsectors by priority:

First, GPU/XPU computing power, with NVIDIA and Broadcom at the core. Both are currently valued at about 9 times forecasted 2028 P/E, combining robust demand with the Fabless business model's profit buffer during demand downturns. NVIDIA remains the top holding; its broader customer base grants it greater flexibility in computing power deployment, sovereign customer business provides additional upside, and the next-generation Feynman architecture could raise per GW revenue from about $40 billion to over $50 billion.

Second, Storage, with a particular focus on Western Digital and Micron Technology. Morgan Stanley believes valuation compression in storage stocks has created significant upside; substantial free cash flow accumulated this cycle is being used for buybacks. For example, with Western Digital’s current share price at $835 and a buyback size at about 50% of revenue, the company could repurchase over 50% of its outstanding shares by the end of 2028. The bank is equally positive on both Western Digital and Micron Technology.

Third, Semiconductor Production Equipment (SPE), where shareholder returns will be the core theme in 2027. Lam Research (LRCX) has newly authorized a $25 billion buyback, bringing its total to about $28 billion—about 11% of market capitalization; Applied Materials and KLA each have outstanding buyback authorizations at about 5% of market cap.

Fourth, Analog Chips, with a positive outlook on NXP Semiconductors (NXPI) and Analog Devices (ADI). The analog cycle remains strong, foundry mature-process capacity utilization has reached nearly 90%, but Morgan Stanley considers valuation elasticity for analog stocks relatively limited, making them less attractive than computing power and storage.

Fifth, Networking, fundamentally strong but relatively more expensive than peers and more crowded. SiTime (SITM), Astera (ALAB), Semtech (SMTC), and Marvell Technology (MRVL) all show positive risk-return profiles, but the bank overall favors other segments.

Sixth, CPU, which has excellent fundamentals recently, but the shortage cycle is expected to end earlier than for storage, and supply-side growth (AMD guiding for 100% growth next year, Intel unit volumes expected to rebound sharply) is significantly outpacing demand. Valuations are relatively high. Morgan Stanley prefers AMD versus Intel, but overall ranks CPU last.

Altogether, Morgan Stanley’s core conclusion is: The semiconductor sector will experience range-bound movement, with upside limited by capital market sentiment and LPS resource management, while downside is supported by strong recent business and multiple long-term catalysts.

The bank notes that using peak capital market optimism as an investment signal—buying at euphoria, selling at despair—is clearly not an alpha-generating strategy. With the sector now generally pulling back over 50% from its peak, the current position is closer to the lower end of the range, and the risk-return reward is significantly improving.

The bank admits that the biggest tail risk is if the current movement is not range-bound but rather the early phase of a cyclical peak, in which case current valuations could become a “value trap.” However, given broad input from the supply chain indicating persistent demand, Morgan Stanley maintains a range-bound base case, and suggests investors maintain an active buying stance when the sector is near the bottom of its range.