US Data Centre Capacity Slips to 2029
Morgan Stanley moves 5 GW of projects after midterm votes, a shift it says will reshape the 2028 to 2029 growth curve.
Updated on Nov. 10, 2026 in Data Centers
Live Poll
Do you trust big tech companies to keep investing in AI as political rules change?
Morgan Stanley shifted 5 GW of US data centre capacity from 2028 to 2029 after Democratic midterm victories, and recommended a selective buy-the-dip approach to AI stocks across hyperscalers, semiconductors and infrastructure. The shift is a forecast revision by the brokerage, not a change to projects already under construction.
Why it matters
Democratic victories in the US midterm elections raised concerns that a Democrat-controlled Congress would curb excessive AI spending on data centres, and tech stocks fell. Newly elected Democratic governors in Texas and Ohio are working with legislators on new requirements for data centre developers, which is why some 2028 projects there are expected to slip.
The brokerage's figures are capacity-addition forecasts measured in gigawatts, projecting year-on-year changes for the US and globally across 2028 and 2029. Morgan Stanley also estimates 20 GW of potential spillover capacity across Asia and EMEA, though the report's underlying methodology is not described in the source.
The players
Morgan Stanley
A US investment bank whose research division publishes sector forecasts and equity recommendations covering the AI infrastructure supply chain.
Amazon
A hyperscale cloud operator whose AWS division runs some of the largest data centre networks in the world and is included in Morgan Stanley's buy recommendations.
Meta
An operator of large AI training clusters and data centres that is included in Morgan Stanley's recommended hyperscaler list.
A hyperscale cloud and AI operator running its own data centre fleet and custom AI accelerators, included in Morgan Stanley's hyperscaler recommendations.
Microsoft
A hyperscale cloud operator whose Azure platform builds large AI data centres, included in Morgan Stanley's hyperscaler recommendations.
The details
Data centre capacity is measured in gigawatts (GW) of power the facilities can draw, and each project's timing determines when that capacity enters service. Morgan Stanley's analysis moves 5 GW of planned US capacity from a 2028 completion to 2029, which the brokerage attributes to political changes at the state level. Because the shift affects roughly a fifth of US projects, it changes the year-on-year growth profile for the sector even if total planned capacity is unchanged.
Timeline
2027: Morgan Stanley expects higher AI-related debt issuance led by hyperscalers.
2028: Some data centre projects shift from 2028 to 2029 according to Morgan Stanley.
2029: Shifted data centre projects are expected to come online.
The Tech Race
Morgan Stanley's forecast places the AI infrastructure build-out in a race where US and global capacity timing both depend on policy decisions by state governors and national legislators. The projected 38% US rise in 2029 positions the US capacity wave later than the 2028 schedule the brokerage had previously assumed.
Users of cloud and AI services from hyperscalers may see capacity constraints ease later than previously expected, since 5 GW of US capacity moves into 2029. Investors in the named hyperscalers, semiconductor and infrastructure companies face a recommendation to buy selectively during the sell-off.
The takeaway
Watch whether the 2029 capacity wave arrives on Morgan Stanley's revised schedule as Texas and Ohio finalise developer requirements. The brokerage's 20 GW Asia and EMEA spillover estimate is the next figure to test against regional build-out announcements.
What happens next
Morgan Stanley expects higher AI-related debt issuance led by hyperscalers in 2027, and shifted projects are scheduled to come online in 2029.
Further reading
For background on how capacity planning shapes this build-out, see our Data Centers coverage.
Live Poll
Do you trust big tech companies to keep investing in AI as political rules change?
