BofA Global Research expects AI demand to drive a sharp expansion in data center spending: $1 trillion globally in 2026 and $2.1 trillion by 2028. Its longer-range projection puts the market at as much as $2.93 trillion by 2030, but those forecasts describe the scale of the buildout—not the returns investors can expect from it.
18
BofA’s data center forecast, at a glance
BofA estimates that the data center market, valued at $275 billion in 2022, could reach $2.93 trillion by 2030. The brokerage also projects a 37% compound annual growth rate from 2025 to 2030. That growth rate applies to the stated 2025–2030 period; it should not be read as the annual rate between the 2022 and 2030 figures.
18
The near-term spending estimates are especially striking: $1 trillion in 2026, followed by $2.1 trillion in 2028. BofA attributes the expansion to rising demand for AI services and their development.
18
What the cost breakdown says about the buildout
Data centers require substantial spending on both computing equipment and the infrastructure around it. BofA puts conventional data center construction costs at about $49 million per megawatt, with IT-related expenses accounting for 78% of costs. Servers alone account for an estimated $31.4 million per megawatt, according to reporting on the forecast.
12
22
AI-focused “neocloud” facilities are estimated to cost about $57 million per megawatt. That is roughly 16% more per megawatt than the $49 million conventional-facility estimate. Greater space efficiency does not, by itself, mean lower cost per unit of power capacity.
8
12
The figures suggest that computing equipment is a major part of the investment, while construction and supporting infrastructure remain important too. The forecast does not, on its own, identify which individual companies will capture the most profit.
Who may benefit—and what could constrain growth
The buildout creates demand for more than servers. Data center spending has become a growing source of revenue for industrial companies making generators, construction equipment, switchgear and chillers, Reuters reports.
6 That points to opportunities across the supply chain, but a spending boom should not be confused with a forecast of any particular company’s earnings or stock performance.
Power availability is also a practical issue. Reuters reports that U.S. hyperscalers are turning to fuel cells as they seek quick, reliable electricity amid grid constraints.
21 That illustrates how delivering power can be part of the infrastructure challenge alongside building facilities and installing computing equipment.
Bigger spending forecasts are not the same as guaranteed returns
BofA strategist Sebastian Raedler has highlighted how sharply expectations for hyperscaler spending have risen: a source summarizing his October 16 remarks puts the estimate at $1 trillion over the next 12 months, up from $330 billion in early 2025.
36 That jump helps explain why the scale of AI infrastructure investment is attracting attention.
But a spending forecast answers how much money may be put into infrastructure; it does not establish how much revenue, profit or investor return that spending will produce. The supplied reporting supports the size and direction of BofA’s forecast, but does not independently verify every claimed hyperscaler spending estimate or a specific AI revenue break-even figure. Treat those as separate claims, not as conclusions that follow from the data center forecast.