AI’s Power Crunch: Is There Still an Opportunity in Memory Stocks?

1️⃣ Can Power Constraints Change the Memory Cycle?

Data centers need power before their servers can go to work. If electricity constraints and permitting delays slow construction, customers may also slow their purchases of AI hardware. Memory, as an essential server component, is exposed to that shift.

Yet Morgan Stanley argued on November 13 that memory supply would become even tighter in 2027. DBS Bank, meanwhile, warned on November 12 that new memory capacity could arrive in 2028 just as data center construction faces delays. [1][2]

The question for investors is whether power constraints will push memory from shortage to surplus, or simply stretch demand over a longer period. That distinction determines whether the industry outlook still deserves a bullish view today.

2️⃣ The Institutional Case: How Do Buildout Delays Affect Memory Supply and Demand?

Step One: Identify Which Years Are Most Exposed to Construction Delays

Morgan Stanley’s Stephen Byrd believes that local permitting and power constraints in parts of the US will push some data centers originally scheduled to come online in 2028 into 2029. He still sees a path to completing those projects. Overseas development and on-site power generation can provide alternatives, but his current assessment is that they cannot fully close the US power shortfall. [1]

DBS Bank also considers a longer, more expensive buildout more likely than widespread cancellations. Drawing on the project data it cites, the bank notes that a substantial portion of the 2027 pipeline is already under construction, providing relatively strong visibility into near-term deliveries. Less advanced projects in the 2028 pipeline are more exposed. [2]

The two institutions share a key assessment: the initial impact is on when new capacity comes online. Hardware purchases across every project and every year do not all stop at once.

Step Two: Why Could Supply Still Be Tight in 2027?

Morgan Stanley’s Joe Moore says that, in recent conversations with hyperscalers, some customers identified memory supply as their most pressing component bottleneck. He expects supply to tighten further in 2027, rather than move into surplus. [1]

He also notes that when some customers lack deployment capacity, other customers with suitable facilities may be able to absorb demand for broadly usable compute hardware. That needs to be assessed project by project, however. It does not mean every memory order can be redirected without disruption. [1]

Together with DBS Bank’s assessment of projects already under construction, the case for a strong 2027 cycle is that customers still need to procure hardware on schedule while component supply remains insufficient. Greater power risk does not, by itself, mean near-term purchases have already fallen below the memory supply vendors can deliver. [1][2]

Step Three: Why Is the 2028 Risk Different?

This is where a meaningful institutional disagreement emerges.

Moore remains relatively optimistic about 2028. He believes demand visibility could hold up, although that needs to be validated quarter by quarter. DBS Bank is more cautious about memory supply: previously initiated capacity expansions are coming online, creating greater supply risk, particularly for HBM. If data center delays also temper demand growth, memory suppliers’ pricing power could come under pressure. [1][2]

The question shifts from “Do customers still want AI?” to “When new memory supply arrives, can customers buy and deploy hardware according to their original plans?”

Even if long-term demand for AI remains intact, deferred purchases can change the supply-demand balance in a particular year. Both deferred demand and permanently canceled orders can pressure suppliers’ business conditions in that year, but they have different implications for the overall investment cycle.

3️⃣ Our View: Positive on 2027, but Reassess the 2028 Supply-Demand Balance

Our answer is that power constraints do limit the realization of memory demand from new data centers. However, the institutional evidence available so far is not enough to overturn the case for tight data center memory supply in 2027. We therefore retain a positive view of the 2027 industry cycle, while avoiding an unconditional extrapolation of the original construction plans into 2028 and beyond.

The key is to compare demand with supply. Even if customer purchases fall short of the original plan, shortages can persist as long as those reduced purchases still exceed what suppliers can deliver. The cycle may change direction only when demand slows enough for additional supply to catch up.

This also explains why project delays can both extend the AI investment cycle and weaken memory suppliers’ pricing power for a period. A longer cycle means demand is realized over more time; pricing power depends on the relationship between purchases and supply in a particular period. One cannot substitute for the other.

For memory, we see support in the near-term supply-demand balance. The main risk is that capacity expansion and deferred purchases coincide in 2028.

HBM is closely tied to AI accelerators, but it also depends on customers deploying hardware on time. DBS Bank specifically identifies HBM as an area facing supply risk in 2028. Strong exposure to AI alone therefore does not establish that it is the most resilient to power constraints. [2] The available evidence is also insufficient to establish a definitive investment ranking among HBM, server DRAM, and data center NAND.

4️⃣ What Would Make Us Change This View?

The next step is to track construction, procurement, and supply, rather than focus solely on the total capital spending announced by hyperscalers.

First, watch whether delays spread to near-term projects. Focus on data centers already under construction and originally scheduled to come online in 2027. Are power connections and delivery dates slipping too? A meaningful deterioration in near-term delivery visibility would weaken the foundation of our positive 2027 view.

Second, watch whether project delays actually become procurement delays. Look for customers postponing hardware deliveries or revising memory purchasing plans, alongside changes in suppliers’ disclosed delivery schedules and orders. Local permitting headlines signal risk; changes in procurement plans provide more direct evidence that the risk is reaching memory demand.

Third, watch whether additional supply catches up with actual purchases. Track capacity coming online, actual shipments, customer and supplier inventories, and contract prices for HBM and other data center memory products. Price changes in one product cannot represent the entire memory market.

If near-term projects keep moving forward, purchases are not materially deferred, and deliverable supply remains tight, we retain our positive 2027 view. If purchases keep shifting out while new supply arrives on schedule, inventories build, and pricing comes under pressure, we should turn more cautious.

5️⃣ Your Turn: Do You Give More Weight to 2027 Shortages or 2028 Supply Risk?

Tigers! With these two time horizons in mind, would you rather pursue the near-term opportunity in tight data center memory supply, or first price in the risks of future capacity expansion and deployment delays while waiting for more evidence?

Share your take in the comments~ [Cool]

Sources

[1] Morgan Stanley: Can AI Chip Demand Outrun the Power Crunch?, November 13, 2026; recorded at 10 a.m. in New York. The full official public transcript featuring Joe Moore and Stephen Byrd was reviewed.

[2] DBS Bank CIO / Yeang Cheng Ling: Global Tech: Lengthier Build-out; Not Broken Thesis, November 12, 2026. The public analysis on the official website was reviewed; the full downloadable PDF and the underlying SemiAnalysis project model it cites were not reviewed.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment(6)

  • Top
  • Latest
  • Kentzw
    ·11-16 17:30
    I would give more weight to the 2027 shortage opportunity, but keep the 2028 supply risk firmly in mind. Tight memory supply could support pricing power and earnings growth in the near term. However, if manufacturers expand capacity too aggressively, the market could shift from shortage to oversupply. I would focus on companies with strong order visibility, pricing power and disciplined capacity expansion rather than simply chasing the current AI demand trend.
    Reply
    Report
  • D1ane
    ·11-16 17:04
    I would focus more on the 2027 shortage opportunity, but I wouldn’t ignore the risks beyond that. Tight memory supply could support pricing power and strong earnings for companies supplying data centre memory in the near term. However, the market may already be pricing in some of that growth. By 2028, new capacity could ease shortages, while construction costs and deployment delays could create another challenge. I would favour companies with strong demand visibility, pricing power and solid financials rather than chasing the biggest short-term gains.
    Reply
    Report
  • HENG8
    ·11-16 14:05
    @Bullish Memory supply constraints are likely to persist through 2027, but should gradually ease from 2028 onward as new production capacity comes online and beyond.
    Reply
    Report
  • mr bean
    ·11-16 13:24
    I’d give more weight to 2027 shortages for now, but keep one eye firmly on 2028. Memory demand is real, but pricing power depends on supply discipline. The danger isn’t just new capacity arriving — it’s arriving just as data-centre buildouts slow. That’s when the cycle gets interesting! 👀
    Reply
    Report
  • 吉3186
    ·11-16 12:55
    AI内存股:2027年仍看好,2028年要小心!

    这篇文章的核心观点是:电力不足可能延迟数据中心建设,但不代表AI内存需求消失。2027年供应仍可能紧张,2028年则要提防供应增加、需求延后的风险。

    一、为什么电力不足会影响内存股?

    AI数据中心需要大量电力才能运行。如果电力供应和审批延迟,数据中心就无法按计划投入使用,服务器和内存采购也可能跟着延后。

    简单来说:没有足够电力,就算买了AI芯片和内存,也未必能马上使用。

    二、为什么2027年仍然看好?

    摩根士丹利认为,部分客户仍面临内存供应不足的问题。

    而且不少2027年数据中心项目已经动工,短期采购需求仍有一定支撑。

    因此,内存供应紧张的情况可能持续到2027年。

    三、为什么2028年要小心?

    主要有两个风险:

    1. 数据中心建设延迟,客户推迟购买内存。

    2. 内存厂商扩充的产能陆续投产,市场供应增加。

    如果需求放缓,同时供应增加,就可能从缺货转向供过于求,内存价格和厂商利润率便可能承压。

    四、我的投资观点

    时间 看法

    2026年底至2027年 相对乐观
    2028年 保持谨慎
    长期AI需求 仍有增长潜力

    投资者可以重点关注SK海力士、三星、美光等内存厂商,但不能只看AI需求,还要观察库存、售价、订单和产能扩张。

    总结:2027年看短缺,2028年防过剩。

    我认为目前不必因为电力问题就看空整个AI内存行业,但也不能假设缺货会一直持续。

    真正决定内存股股价的,不只是AI需求有多强,而是未来内存供应能否跟上需求。

    记住:行业需求增长,不代表股票价格一定上涨;买入价格和市场预期同样重要。

    Reply
    Report
  • Jerry Lam
    ·11-16 12:50
    🐯 **我的选择:看好2027年的内存短缺机会,但会提前防范2028年的供需逆转风险。**

    **① 2027年:供给紧张的逻辑仍然成立。** AI服务器对HBM、DRAM的需求持续增长,而内存扩产需要时间。即使部分数据中心延期,也不代表短期需求会突然消失。

    **② 2028年:我反而更加谨慎。** 如果新增产能集中释放,恰好遇上电力不足导致服务器部署延期,就可能出现“需求还在增长,但供应增长更快”的局面,进而压低内存价格和企业利润率。

    **③ 投资上,我更关注MU等存储厂商的实际订单、合同价格、库存变化和资本开支,而不是单纯追逐AI概念。**

    🚀 **我的策略是参与2027年的景气周期,但不会无限期持有。** 一旦出现库存上升、交货周期缩短、价格涨幅放缓等信号,就会考虑逐步降低仓位。

    **内存行业最危险的时刻,往往不是需求消失,而是市场仍然相信长期短缺时,供给已经开始追上需求!**

    Reply
    Report