A summary of Morgan Stanley Research
Morgan Stanley (8 November 2026): Buy the Drop – AI Investment Opportunities Following the US Midterms
Morgan Stanley believes the recent AI-related market sell-off is unwarranted and an overreaction to the US midterm election results. Its central argument is that AI investment isn't being cancelled; some data centre construction is simply being pushed into 2029.
The numbers are quite extraordinary.
Morgan Stanley's hyperscaler capital expenditure forecasts remain unchanged:
2025: $481bn
2026: $951bn (+98%)
2027: $1.64 trillion (+73%)
2028: $1.76 trillion (+7%)
That's approximately $3.4 trillion of investment in 2027–28 alone, with Google, Amazon, Microsoft and Meta accounting for the overwhelming majority.
Morgan Stanley expects approximately 20% of planned US data centre projects for 2028, representing 5GW of capacity, to shift into 2029 following political and regulatory changes. Importantly, that's only around 10% of global projects, and 2027 construction forecasts remain unchanged.
Its modelling suggests global data centre capacity additions will still grow 9% in 2028 and 20% in 2029. US additions are expected to decline 4% in 2028 before rebounding 38% in 2029.
Morgan Stanley also forecasts a greater than 30% shortfall in powered data centre capacity through 2028. In other words, demand for compute infrastructure continues to exceed available supply.
There are several important implications.
First, the AI investment cycle could actually last longer. Delaying some projects rather than cancelling them potentially extends the spending cycle, improves supply-demand balance and supports pricing for existing compute capacity.
Second, the largest technology companies benefit. Google, Amazon, Microsoft and Meta have the financial resources, infrastructure and flexibility to navigate regulatory delays, higher construction costs and power constraints. Smaller developers and heavily leveraged operators face greater challenges.
Third, compute efficiency becomes increasingly important. Meta is targeting a tenfold improvement in compute efficiency during 2027, while Google is delivering substantial improvements through its TPU hardware and software. More efficient models should make AI applications commercially viable across a much broader range of products and users.
Fourth, there are risks. Morgan Stanley acknowledges that a slowdown in 2028 construction could temporarily weaken pricing power, increase inventories and pressure margins across parts of the technology supply chain. It also warns that further delays could create downside risk to its 2028 capex forecast.
Nevertheless, the bank remains constructive on hyperscalers, selected semiconductor companies, AI infrastructure and software businesses, arguing that recent share-price weakness presents buying opportunities.
My takeaway: This is hardly the collapse of the AI investment cycle that some commentators have been predicting. Morgan Stanley is forecasting almost $3.4 trillion of hyperscaler investment over just two years, with compute demand continuing to exceed available capacity.
There may be some disruption to the timing of projects, particularly in 2028, but the underlying investment thesis remains firmly intact.
For companies supplying the essential computing and memory infrastructure, including NVIDIA and Micron, that is a substantially more encouraging outlook than recent market sentiment would suggest.
[image: 1794312596690-screenshot-2026-11-10-at-12.09.28.png]
