Morgan Stanley's latest research shows that political headwinds triggered by the U.S. midterm elections are materially altering the pace of AI data center construction, with roughly 20% of new U.S. capacity originally slated to come online in 2028 now pushed to 2029, and about 40% of chips deployed in the U.S. between 2026 and 2028 facing the risk of being unable to receive power in time. This assessment will have knock-on effects across the technology supply chain, the data center financing market, and the overall AI capital expenditure cycle.
According to Morgan Stanley's November 9 report, the firm estimates that about 5 GW of U.S. data center projects will be deferred from 2028 to 2029. After adjustments, annual new U.S. data center capacity will decline 4% year over year in 2028 (while global capacity still grows 9%), before rebounding 38% in 2029 (global growth of 20%). The 116 GW of power demand corresponding to U.S. chip shipments from 2026 to 2028 carries a net potential power gap of 44 GW, or roughly 38% to 40%, driven mainly by insufficient grid capacity, projects still under construction, and midterm election-related power delays.
The report makes clear that the policy tools capable of actually constraining project advancement rest primarily with state and local governments, while the federal political landscape has limited direct influence on specific projects.
Morgan Stanley believes newly elected governors are unlikely to interfere with projects already under construction, so 2027 project expectations remain unchanged; however, some projects in Texas and Ohio originally scheduled for 2028 could be delayed as governors and legislatures negotiate new requirements. The four main consequences are: more dispersed data center siting (including a shift overseas), rising premiums on existing capacity, the risk of a temporary gap in the 2028 technology supply chain, and an overall lengthening of the AI capital expenditure cycle.
Political Headwinds Reshape Construction Pace, 2028 Emerges as a Dividing Line
Morgan Stanley's report notes that opposition to AI and data center expansion was a clearly polarizing issue among voters in this midterm election, with the firm's August 2026 survey showing that 72% of likely voters considered candidates' AI positions "very important" or "somewhat important," a figure reaching as high as 76% among those aged 18 to 34.
However, the report emphasizes that whether campaign rhetoric can translate into actual project restrictions remains constrained by procedural factors. Although Democrats scored important gubernatorial victories in key states such as Texas and Ohio, control of state legislatures remains split between the two parties, making it difficult for a governor's unilateral position to bypass the legislative process and take direct effect.
Disruption to projects under construction is expected to be limited. The report judges that new governors are more likely to exert influence by working with legislatures to set new approval requirements for data center developers, rather than directly halting construction. This logic leads the report to maintain its 2027 construction expectations while concentrating the pace adjustment on new projects breaking ground in 2028.
The Power Gap Is the Deeper Constraint Variable
Beyond the political dimension, the report identifies insufficient power supply as the core constraint on chip deployment from 2026 to 2028.
Under Morgan Stanley's waterfall model, total power demand corresponding to U.S. chip shipments from 2026 to 2028 is 116 GW; after deducting the 25 GW that can be covered by data centers under construction and 21 GW of available grid capacity, and adding the 6 GW gap from midterm election-related power delays, the potential power gap reaches 76 GW.
On this basis, alternative power solutions such as turbines, Bloom Energy fuel cells, and nuclear power can bridge 27 GW, while shifting deployment to Asia can reduce the gap by another 5 GW, ultimately narrowing the net potential gap to 44 GW, or about 40% of the 116 GW total demand.
The U.S. Remains the Largest Market, but Overseas Demand Is Rising
Morgan Stanley estimates that the U.S. market's leading position will not be fundamentally altered by the outcome of this midterm election. Using 2025 as a baseline, the U.S. is expected to add about 103 GW by 2030, China about 55 GW, and Europe about 13 GW.
But the midterm election results may further push U.S. developers to look overseas. Morgan Stanley believes Southeast Asia (+6 GW), India (+7 GW), Europe (+12 GW), Australia (over +4 GW), Japan (+2 GW), and South Korea (+2 GW) are viable regions that can reasonably absorb U.S. data center capacity. The report also notes that new European capacity is expected to mainly serve local European demand.
When selecting overseas build sites, construction cost and electricity price are the two most important considerations. The report estimates that data center construction costs are rising at more than 10% per year, driven by factors including the energy crisis, rising interest rates, grid investment needs, and rising wage costs. GPUs account for roughly half of total capital expenditure per GW of AI data center capacity, making them the single highest-weighted item in capital allocation.
Hyperscaler Capex Forecasts Maintained, Credit Market Shows Quality Divergence
Morgan Stanley maintains its capital expenditure forecasts for hyperscale technology companies including Google parent GOOGL, Amazon, Microsoft, and Meta.
By its estimates, combined capital expenditure for these firms will grow from about $481 billion in 2025 to about $951 billion in 2026, about $1.64 trillion in 2027, and further to about $1.76 trillion in 2028, with growth slowing sharply from 98% in 2026 and 73% in 2027 to 7% in 2028. The report also notes that if the 2028 construction delays materialize, there is a possibility of downward revisions to that year's capital expenditure.
At the credit market level, the report observes clear quality divergence within AI financing. Bond issuance by high-quality hyperscale technology companies has recently picked up, but their credit spreads have widened to the widest level of the year, above the broad investment-grade bond index, despite their generally higher credit ratings. Meanwhile, issuance of more interest rate-sensitive data center construction debt has slowed markedly, with related bonds continuing to underperform same-rated indices since the summer.
Morgan Stanley expects chip financing transactions and shorter-duration hyperscaler debt issuance to increase. As the market reassesses the risk premium on data center construction, data center bonds are expected to remain volatile. If the construction pace slows, pressure from new financing supply may be somewhat alleviated.
Comments