Morgan Stanley: Buy AI Dip, $3.4T Hyperscaler Capex to 2028

Morgan Stanley told clients on 9 November 2026 to buy the recent weakness in artificial intelligence stocks, arguing the market overreacted to US midterm election results and their read-through for data center construction. The bank kept its hyperscaler capital expenditure forecasts unchanged at $1.6 trillion for 2027 and $1.8 trillion for 2028, while flagging downside risk to the 2028 figure. Its preferred names, Amazon, Meta, Alphabet, Microsoft, CoreWeave and Nvidia, traded 2% to 8% higher on the session, with Nvidia the lone exception, down 6%.

Context — why the AI capex debate turned political

The report, titled "Buy the Drop: AI Investment Opportunities Post the US Midterms," frames the selloff as a political event rather than a demand event. Morgan Stanley expects 20% of US data center projects planned for 2028 to slip into 2029, equal to 5 gigawatts of capacity. That is the single largest revision in the note, and it lands almost entirely on the construction calendar rather than on the spending envelope.

The bank does not expect projects already under construction to be disrupted, which is why its 2027 forecasts hold. Instead, permitting and regulatory friction is expected to push some 2028 starts into the following year. The concern centers on Texas and Ohio, where newly elected governors could add requirements for data center construction.

The distinction matters because hyperscaler capex and data center capacity additions are not the same series. Capex can be deferred in timing while the aggregate multi-year total stays intact. Morgan Stanley's 2027 number is untouched, and its 2028 number is unchanged but carries an explicit downside caveat.

That leaves the AI trade priced against a political risk premium that the bank argues is too wide. The recent selloff, in its view, treated a scheduling problem as a demand problem. Investors scanning broader market structure can track sector-level moves on the Fazen Markets equities desk.

The macro backdrop for the call is a market that has already absorbed the midterm outcome. Morgan Stanley's argument is that the second-order effects on permitting, not the election result itself, are what the tape mispriced.

Data — what the numbers show

The capacity math is the core of the revision. US data center capacity additions are now expected to decline 4% in 2028 before jumping 38% in 2029. Globally, additions are forecast to grow 9% in 2028 and 20% in 2029. The US 2028 figure is the only negative print in the set, and it reverses the following year by a wide margin.

A shortage of more than 30% in powered data center capacity is expected to persist through 2028. That gap is the mechanism the bank uses to argue delays are not demand destruction: capacity stays scarce, so lease rates for existing compute hold up.

Series	2028	2029
US capacity additions	-4%	+38%
Global capacity additions	+9%	+20%
Hyperscaler capex	$1.8T	n/a

The bank also sees the potential for as much as 20 GW of additional data center development across Asia and EMEA, and a geographic shift within the US toward more accommodating states such as Tennessee and Louisiana.

On the equity side, Morgan Stanley rates Amazon (AMZN), Meta (META), Alphabet (GOOGL), Microsoft (MSFT), CoreWeave (CRWV) and Nvidia (NVDA) Overweight. It highlights Equinix (EQIX) and powered-shell providers WULF, RIOT, HUT and APLD as less exposed to Texas-related risk, and flags CIFR, GLXY and MARA as more vulnerable to delays or moratoriums. All of those names traded 4% to 9% higher except Cipher, which fell 2.2%.

Analysis — hyperscalers versus the supply chain

The report splits the AI complex in two, and the split is the actionable part. Hyperscalers carry the balance sheets, existing infrastructure and technical capability to work around power constraints. Suppliers to the construction boom do not.

If the bottleneck shifts from equipment availability to powered buildings, Morgan Stanley warns that suppliers could lose pricing power, accumulate inventory and see orders slow. It describes a possible traditional downcycle in 2028 for that group, including margin normalization and earnings downgrades, even with underlying AI demand intact.

Efficiency is the second lever. The bank points to advances in model architecture, distillation, training techniques and specialized chips that could let AI applications scale without matching increases in compute requirements. It cites Meta's target of a tenfold increase in compute efficiency in 2027. Nvidia's role is framed around improving AI output per watt of power, which is a different demand driver than raw accelerator count.

The financing channel reinforces the same hierarchy. Morgan Stanley expects AI-related debt issuance to rise in 2027, but says new data center developments and neocloud operators will face greater scrutiny from lenders. Access to capital, not access to demand, becomes the filter.

The counter-argument is straightforward: if the 2028 capex figure does carry downside risk, then the 2029 rebound is a forecast, not a backlog. Morgan Stanley itself flags that risk rather than dismissing it. Positioning reflects the tension, with flows favoring the largest platforms and powered-shell names over equipment suppliers.

Outlook — what to watch next

Three catalysts sit on the near-term calendar. CoreWeave reports on Tuesday, which the report flags as the next hard data point for the neocloud cohort. The Anthropic IPO is expected to proceed this month, following a Friday report to that effect. And the permitting regimes in Texas and Ohio will show up first in project announcements rather than in earnings.

On the data side, watch whether US 2028 capacity additions hold at the revised negative 4%, or whether further projects slip. The 2029 figures, 38% US and 20% global, are the numbers that need to survive intact for the bank's thesis to hold. The powered-capacity shortage, above 30% through 2028, is the variable that supports lease rates if delays persist.

For the supply chain, the tell is order books and inventory rather than headline AI demand. For the hyperscalers, it is capex guidance at the next reporting round. Neither is resolved by the midterm result alone.

Frequently Asked Questions

What does Morgan Stanley's call mean for retail investors holding AI stocks?

The report draws a line between hyperscalers and their suppliers rather than issuing a blanket buy. Amazon, Meta, Alphabet, Microsoft, CoreWeave and Nvidia carry Overweight ratings, while construction-linked names face a possible 2028 downcycle. Retail investors holding broad AI exposure may find that the two halves of the trade behave very differently if permitting delays materialize as forecast.

Why did Morgan Stanley cut its 2028 US data center forecast?

It did not cut capex. It shifted timing. The bank expects 20% of US projects planned for 2028, about 5 gigawatts, to move to 2029 because of permitting and regulatory friction in Texas and Ohio. That pushes US capacity additions to negative 4% in 2028 before a 38% jump the following year.

What happens to AI suppliers if data center construction slows?

Morgan Stanley warns that if the bottleneck moves from equipment availability to powered buildings, suppliers could see reduced pricing power, rising inventories and slowing orders. It describes a possible traditional downcycle in 2028 with margin normalization and earnings downgrades. Equinix and powered-shell names such as WULF, RIOT, HUT and APLD are flagged as less exposed to Texas risk than CIFR, GLXY and MARA.

Bottom Line

Morgan Stanley says the AI buildout is intact, but the winners are shifting from suppliers to hyperscalers.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. CFD trading carries high risk of capital loss.