BofA Forecasts AI Data Center Spending to Reach $2.1 Trillion

Key Takeaways

• Global data center spending is forecast to reach $1 trillion in 2026 and $2.1 trillion by 2028.
• Servers, networking, power and cooling account for much of the investment needed to support AI services.
• Electricity availability, construction capacity and financial discipline could shape how quickly planned facilities enter service.

BofA Global Research, as reported by Reuters, forecasts that global data center spending will reach $1 trillion in 2026 before climbing to $2.1 trillion in 2028. AI-service demand is expected to provide the main impetus, pushing cloud providers and their suppliers into a period of unusually heavy infrastructure investment.

The longer-range projection is even larger. BofA estimates that the data center market could expand from $275 billion in 2022 to $2.93 trillion by 2030, implying a compound annual growth rate of 37% between 2025 and 2030. The forecast includes more than buildings. It encompasses the computing systems, networking equipment, electrical infrastructure and cooling capacity needed to operate increasingly dense AI clusters.

An AI data center is expensive before it processes a single prompt. BofA estimates construction costs at $49 million per megawatt, with IT equipment representing 78% of total cost. Servers account for about $31.4 million per megawatt, while networking adds another $4.8 million per megawatt.

That cost profile helps explain why NVIDIA, Microsoft Azure and Amazon Web Services occupy prominent positions in the expansion. NVIDIA supplies much of the accelerated computing technology behind AI workloads, while Microsoft Azure and Amazon Web Services are building the cloud environments through which many businesses consume that capacity. Demand ripples outward to memory, storage, networking, cooling, backup power and construction providers.

Capital budgets are responding. The International Energy Agency reported that capital expenditure by five major technology companies exceeded $400 billion in 2025 and is expected to increase another 75% in 2026. That scale suggests AI infrastructure is becoming a major component of technology-sector investment rather than a narrow experimental category.

The constraint may increasingly shift from access to chips toward access to power. Global data center electricity consumption is projected to rise from 485 terawatt-hours in 2025 to about 950 terawatt-hours in 2030, according to the IEA. Consumption at AI-focused facilities is expected to triple over the same period.

Can utilities, grid operators and equipment manufacturers keep pace? In some regions, obtaining grid connections and high-voltage equipment can take longer than installing servers. Developers may therefore favor markets with available generation, predictable permitting and credible plans for transmission upgrades. Power procurement could also influence where cloud capacity is located and how quickly customers receive it.

The spending mix is changing as well. IDC forecasts that annual service-provider procurement of AI infrastructure will rise from $401 billion in 2025 to $1.45 trillion by 2030. Over that period, data center facilities capital expenditure is projected to increase from $276 billion to $655 billion.

Those figures point to two overlapping build cycles. One involves computing hardware that can be refreshed relatively quickly as accelerators improve. The other involves long-lived physical assets, including substations, cooling plants, generators and data halls. Operators have to coordinate both cycles without committing too much capital to facilities that could be poorly matched to later hardware designs.

That said, higher spending does not automatically translate into proportional revenue. Providers still need sufficient utilization, attractive service pricing and customers willing to move AI systems from pilots into sustained production. Rapid hardware advances can also shorten the economic life of expensive equipment, raising depreciation pressure and complicating return calculations.

For enterprise buyers, the expansion could increase access to AI computing, but procurement decisions are likely to remain nuanced. Capacity location, network latency, energy costs, data governance and service portability can matter alongside raw processor availability. Some organizations may use public cloud services, while others pursue dedicated or shared infrastructure for sensitive and predictable workloads.

The $2.1 trillion forecast ultimately describes more than a construction boom. It reflects a broad attempt to industrialize AI delivery at global scale. Whether every planned facility gets built is less certain. Still, spending on servers, networks, power systems and cooling is moving AI from the software budget into the heart of corporate infrastructure planning.