Data Center Spending to Hit $2.1T by 2028 Says BofA

Corporate capital spending on digital computing infrastructure is outstripping historical records. For half a century, telecommunications networks and petroleum pipelines represented the most capital-intensive physical developments in the global economy. Companies spent billions laying undersea fiber and constructing refinery complexes over multi-year cycles. That physical scale has been eclipsed by server warehouses and computational clusters. In a research note published on Friday, October 23, 2026, Bank of America Global Research reported that total global data center spending will hit $1T this calendar year and more than double to $2.1T by 2028. The forecast, authored by senior equity analysts at the bank, illustrates the unprecedented scale of capital flowing into computer hardware, cooling facilities, electrical substations, and fiber networks. Hyperscale cloud operators and sovereign states are spending fortunes to secure computing capacity. The massive capital wave lands as power grids struggle to deliver promised electricity, an industrial bottleneck visible when America computing infrastructure expansion began to stall under utility delays.

The projections highlight a profound transformation in corporate balance sheets. Building modern computing parks is no longer an ordinary information technology expense; it has become an industrial construction enterprise requiring gigawatts of continuous power generation. Technology conglomerates including Microsoft, Amazon Web Services, Alphabet, and Meta have committed hundreds of billions in capital expenditures to purchase high-bandwidth processors, liquid-cooling manifolds, and private power generation equipment. BofA analysts noted that compute demand generated by machine reasoning models and generative software tools shows zero signs of cooling, forcing infrastructure builders to lock in multi-year equipment backlogs. Still, Wall Street asset managers are asking sharp questions regarding financial returns. Spending $2.1T on physical infrastructure by 2028 requires technology companies to generate massive software revenues to justify carrying heavy hardware depreciation charges on their ledgers. The mounting financial scrutiny matches warnings we analyzed when Wall Street set impossible profit targets on corporate technology earnings.

The Math Behind the Two Trillion Dollar Buildout

To evaluate BofA $2.1T projection, one must dissect where capital is being deployed inside modern server campuses. A decade ago, server racks were air-cooled cabinets pulling five to ten kilowatts of power, filled with generic central processors rented from standard hardware vendors. Building an enterprise data center cost tens of millions of dollars.

Modern server campuses behave like heavy industrial complexes. High-performance computing racks housing advanced accelerators pull up to one hundred kilowatts per cabinet. Dissipating that concentrated heat requires direct-to-chip liquid cooling loops, cooling towers, and closed-loop heat exchangers. BofA research team divided projected 2028 spending across four major capital categories:

• Accelerators and Memory Silicon: Specialized processing chips, high-bandwidth HBM4 memory stacks, and host processors account for roughly 48% of total expenditure.
• Power Generation and Electrical Grid Equipment: Dedicated natural gas peaker plants, high-voltage transformers, switchgear, and uninterruptible power supply batteries represent 22% of outlays.
• Thermal Cooling and Mechanical Systems: Liquid-cooling manifolds, chillers, pumps, and water recycling systems capture 16% of capital commitments.
• Physical Real Estate and Structural Construction: Reinforced concrete server halls, optical conduits, and security perimeters account for the remaining 14%.

The capital distribution demonstrates that spending has shifted outward from pure software code into heavy civil and electrical engineering. Managing cabinet power without triggering utility disputes was explored in our report on flexible data center power management.

The Electrical Power Wall: Substation Delays and Grid Congestion

The primary constraint threatening to slow down BofA $2.1T projection is not capital availability; it is the physical capacity of electrical grids. In traditional technology hubs like Northern Virginia, Silicon Valley, and Dublin, regional power utilities have run out of transmission capacity.

Commercial developers face interconnection queues stretching between four and seven years. Regional transmission operators like PJM Interconnection have warned that adding multi-gigawatt industrial data center loads threatens grid stability for residential ratepayers. These delays have forced cloud operators to abandon planned developments, a reality highlighted when Microsoft terminated its data center contract with Nscale over multi-year transmission connection delays. To bypass grid bottlenecks, tech operators are adopting behind-the-meter generation: purchasing on-site natural gas combustion turbines or signing direct deals with nuclear plants to supply dedicated baseload current. The political friction surrounding industrial energy consumption was examined when public pushback mounted against data center power consumption.

Comparing Capital Expenditure Projections Across Major Institutions