Bank recommends buying AI dip, expects $3.4 trillion hyperscaler capex by 2028

Buying into the recent AI stock weakness is recommended by Morgan Stanley, which contends the market overreacted to the US midterm election results and their effects on data center construction.

Its report, 'Buy the Drop: AI Investment Opportunities Post the US Midterms,' states that the AI investment cycle is still solid, though political pushback against data centers could defer some projects to 2029.

The bank's projections for hyperscaler capital spending are unchanged at $1.6 trillion for 2027 and $1.8 trillion for 2028, though it notes some risk for the 2028 forecast.

Key points:

• Morgan Stanley expects 20% of US data center projects planned for 2028 to be delayed until 2029, amounting to 5 gigawatts of capacity.

• US data center capacity additions are forecast to drop 4% in 2028 before climbing 38% in 2029.

• Worldwide, data center additions are seen rising 9% in 2028 and 20% in 2029.

• A deficit of over 30% in powered data center capacity is anticipated through 2028.

• Among hyperscalers, Morgan Stanley prefers Amazon, Meta, Alphabet and Microsoft.

• The bank cautions that some AI supply chain segments could see decreased orders, inventory buildup and margin strains in 2028.

Anxiety is mostly focused on Texas and Ohio, where recently elected governors may impose extra rules for data center construction. Morgan Stanley does not foresee disruption to projects under construction, so its 2027 predictions stand.

Rather, it expects permitting and rule-related delays to move some 2028 projects into 2029. There is also a suspicion that power constraints were developing, which would have caused delays regardless.

Looking further ahead, Morgan Stanley sees a possible geographic shift, with hyperscalers moving some projects to more friendly states like Tennessee and Louisiana. The bank also sees scope for up to 20 GW of additional data center development in Asia and EMEA.

The bank contends the delays may eventually be positive for the AI investment cycle, helping to balance supply and demand, supporting lease rates for existing compute capacity and extending the investment period.

The key difference is between hyperscalers and their suppliers.

Morgan Stanley is especially positive on the biggest tech firms, which have the financial health, existing infrastructure and technical skills to handle power limitations.

Its favored stocks are:

• Amazon (AMZN), Meta (META) and Alphabet (GOOGL) receive an Overweight rating, with benefits in scale, capital access and AI development.

• Microsoft (MSFT) is rated Overweight, with comparable benefits.

• CoreWeave (CRWV) is also rated Overweight.

• Nvidia (NVDA) gets an Overweight rating, with an important role in boosting AI output per watt of power.

All these stocks are up 2-8% today except Nvidia, which is down 6% after the Jensen Huang honeypot allegations. CoreWeave, which reports on Tuesday, is up 8% and leads.

The bank points to the possibility of significant efficiency gains in AI models, such as Meta's goal of a tenfold improvement in compute efficiency in 2027.

Advances in model structure, distillation, training methods and specialized processors could let AI applications expand rapidly without a proportional rise in computing needs, according to the bank.

For firms that supply the data center construction boom, the outlook is more mixed.

If the bottleneck shifts from equipment availability to powered buildings, suppliers could face weaker pricing power, growing stockpiles and slowing orders. Morgan Stanley warns that the sector could see a traditional downcycle in 2028, including margin normalization and earnings downgrades, even as underlying AI demand remains strong.

Thus, stock selection becomes more important.

Among infrastructure-related opportunities, Morgan Stanley picks out Equinix (EQIX) and powered-shell providers such as WULF, RIOT, HUT and APLD, which it views as less vulnerable to Texas risks. It highlights CIFR, GLXY and MARA as possibly more vulnerable to construction delays or bans. All these names are up 4-9% except Cipher, which is down 2.2%.

The financing environment also benefits established players. Morgan Stanley forecasts a rise in AI-related debt issuance in 2027, but says new data center developments and neocloud operators will face greater lender scrutiny.

The broader message is that the AI buildup is not ending, but the beneficiaries may shift. The largest hyperscalers could gain from tighter capacity, better compute efficiency and weaker competition, while some equipment suppliers face a more difficult transition as the construction schedule adjusts.

The recent sell-off is seen as an opportunity by Morgan Stanley, but it is clearly not advising buying all AI stocks without discretion. Looking ahead, attention is on the Anthropic IPO, with a Friday report that it will go ahead this month.