Recent volatility has weighed on AI infrastructure stocks as investors question whether years of heavy spending on data centres and computing power can continue at the same pace. Into that uncertainty, Morgan Stanley published a 27 July report, “Playing the AI Infrastructure Dip,” arguing that the market has become overly cautious and that the recent pullback has been driven more by technical factors than by any deterioration in underlying fundamentals. The bank's bottom line: it remains bullish on what it calls the “Intelligence Superhighway,” even as it flags real speed bumps ahead.
Morgan Stanley's Core Investment Thesis
The bottleneck on AI growth is shifting from chip supply to power supply. Hyperscalers are committing up to $1.4 trillion in capex and racing to secure 30–120 GW of compute by 2028, but the US grid can't keep pace, leaving a projected shortfall of up to 38 GW through 2028. Because “powered shell” data-centre infrastructure earns far higher returns than traditional power projects, Morgan Stanley's highest-conviction call is to own the companies that control power access, not just the companies that make chips.
The Scale of AI Infrastructure Investment
Morgan Stanley's report quantifies the scale of the infrastructure buildout underpinning this thesis:
These figures illustrate why Morgan Stanley believes electricity, not semiconductors, is becoming the defining constraint on AI expansion. As compute capacity accelerates, companies that can deliver power, grid connectivity and data-centre infrastructure are expected to capture a growing share of AI-related investment.
Closing the AI Power Gap
Morgan Stanley quantifies the 2026–2028 US data-centre power shortfall from first principles: roughly 68 GW of total power demand, less than about 15 GW already under construction and 15 GW of contracted grid capacity, leaves a starting gap of 38 GW.
The report then walks through the solutions, narrowing that gap with natural gas turbines, Bloom Energy-style fuel cells, siting data centres at existing operational nuclear plants, and converting former Bitcoin-mining sites (which already hold grid interconnection rights) into data-centre capacity. Weighting each solution by its probability of success, Morgan Stanley estimates these bring the net shortfall through 2028 down to somewhere between roughly 1 GW and 11 GW a gap that's shrinking, but not yet closed. The report singles out repurposed Bitcoin-site power and fast-to-deploy on-site generation as the two quickest paths to closing it, since both can offer a one- to three-year time-to-power advantage over waiting on standard grid interconnection.
Why Capital is Rotating Toward “Powered Shells”
A second pillar of the report's argument is economic rather than structural. Morgan Stanley compares the returns available from traditional renewable power-purchase agreements (PPAs) against those available from “powered shell” data-centre leases pre-built facilities with guaranteed power access leased directly to hyperscalers:
The gap is stark: despite a far higher cost per watt to build, powered shell leases generate materially higher free cash flow yields and more than fifteen times the net value creation per gigawatt of a comparable renewable PPA. Morgan Stanley argues these economics explain a real shift already underway in where capital is being directed away from pure electricity generation for its own sake, and toward the infrastructure that packages power together with the physical real estate hyperscalers need to deploy it.
Morgan Stanley's Five AI Infrastructure Themes
Rather than viewing AI infrastructure as a single investment theme, Morgan Stanley organises its investment case around five broad categories:
- AI infrastructure bottlenecks: Companies addressing labour shortages, time-to-power solutions, energy storage, power developers and data-centre REITs.
- Compute manufacturing: The semiconductor and hardware ecosystem, where chip supply continues to lag AI demand.
- Chinese AI solution providers: Leading Chinese AI companies that Morgan Stanley believes remain competitively positioned despite valuation concerns.
- Energy security: Businesses supporting reliable electricity supply, with energy storage identified as a key long-term opportunity.
- Hyperscalers: Large cloud providers capable of converting substantial AI capital expenditure into scalable, long-term returns.
Taken together, the report's core argument- a widening power shortfall and the superior economics of infrastructure that helps close it- supports the first and broadest of these categories: the companies positioned between committed AI capital expenditure and delivered computing capacity.
While Morgan Stanley remains constructive on AI infrastructure, it also acknowledges several risks that have contributed to the recent selloff. The report addresses each of these concerns and explains why it believes they are unlikely to derail the long-term investment case.
Three Concerns, and Morgan Stanley's Response
Morgan Stanley frames the recent selloff around three specific investor concerns, and addresses each with supporting data.
Concern 1: “Tokenmaxxing” limits.
The worry is that enterprises will cap employee token spending, capping AI-linked revenue growth. Morgan Stanley pushes back on the underlying economics: it estimates that the current median enterprise employee spends less than $11 a month on tokens, while a typical enterprise AI use case saves the business around $55, against a token cost of roughly $2–5 to deliver. On that math, enterprises adopting AI more aggressively is a matter of when, not if. The report adds that margins on token sales should stay attractive for the LLM developers themselves as GPU generations improve: its model projects data-centre token-sale margins rising from around 60% on current-generation Blackwell chips toward 80–90% on the upcoming Rubin and Feynman generations.
Concern 2: Chinese open-weight model competition.
Morgan Stanley reserves judgment on the raw capability of recent Chinese LLMs, but argues the competitive pressure cuts in its favour either way: efficiency gains from Chinese and American developers alike tend to lower the cost of AI use, which historically expands total demand rather than shrinking it (a dynamic the report ties to Jevons' Paradox, the idea that cheaper access to a resource raises overall consumption of it, not just per-user use).
Concern 3: People, Power and Politics.
Morgan Stanley calls this the most valid of the three concerns, though still a “speed bump” rather than a structural barrier. It breaks the challenge into three parts: a shortage of skilled data-centre construction labour (electricians, welders, pipefitters); increasingly constrained grid access, with interconnection queues in some regions stretching five to seven years; and a rising wave of political pushback, including several US states rolling back data-centre tax incentives and a proposed federal Ratepayer Protection Act that would require data centres to help fund the grid upgrades their demand creates.
How AIPOWR Aligns with Morgan Stanley's AI Infrastructure Thesis
While Morgan Stanley does not recommend AIPOWR specifically, the ETF closely reflects many of the report's preferred AI infrastructure themes. Listed on the Abu Dhabi Securities Exchange, AIPOWR provides GCC investors with exposure to utilities, electrical equipment manufacturers, data-centre infrastructure and AI compute providers rather than concentrating primarily on semiconductor companies. Mapping the portfolio against Morgan Stanley's five investment themes shows that the ETF is most heavily aligned with the firm's highest-conviction idea, AI infrastructure bottlenecks, while also providing selective exposure to hyperscalers and the compute manufacturing ecosystem.
Fund Snapshot
The fund's sector positioning reflects the report's thesis directly: Industrials (34.60%) and Utilities (33.03%) together account for nearly 68% of the portfolio, complemented by Information Technology (15.08%), Real Estate (6.94%), Consumer Discretionary (4.94%) and Communication Services (4.83%). Top holdings include Microsoft, Siemens, NextEra Energy, Iberdrola, Schneider Electric, Broadcom, Eaton, Amazon, ABB and Alphabet alongside further positions in Vertiv, Equinix, Quanta Services, Enel, Constellation Energy, National Grid and Digital Realty.
As with any fund tied to a fast-evolving structural theme, AIPOWR carries risks worth weighing: a small asset base (~$17.6mn AUM) that limits trading liquidity, a limited operating history with no one-year performance record yet, ~68% combined sector concentration in industrials and utilities, an elevated index weighted average P/E of 35.38x, exposure to any slowdown or de-speccing of hyperscaler capex, and a structure that is not Shariah-compliant a relevant constraint for GCC investors operating under Shariah mandates.
Bottom Line
Morgan Stanley's research suggests the AI investment opportunity is broadening beyond semiconductor manufacturers to the companies enabling electricity generation, grid expansion and data-centre development. Even under the firm's more optimistic "time-to-power" scenarios, a meaningful supply gap is expected to persist through 2028, reinforcing its preference for AI infrastructure. For GCC investors looking to express that investment view through a locally listed vehicle, AIPOWR offers diversified exposure to many of the companies positioned to benefit from this long-term structural trend.
