Demand for data centers is rising sharply, driven by the rapid growth of AI and 
cloud computing

However, existing infrastructure cannot keep pace. Goldman Sachs estimates that U.S. data centers already face a capacity shortfall of more than 11 gigawatts (GW), with the cumulative gap expected to exceed 40 GW by 2028.

The expansion of AI-focused facilities is accelerating at an unprecedented rate

Global commercial real estate services and investment management company Jones Lang LaSalle projects that North American data center capacity will increase eightfold, from 5.6 GW in 2024 to 44 GW by 2030, and Deloitte forecasts that U.S. AI data center power demand could soar from 4 GW in 2024 to 123 GW by 2035. Together, these projections highlight the massive scaling of infrastructure needed to support the next wave of AI adoption. 

At the same time, the industry’s energy footprint is expanding rapidly

Bain Capital forecasts that data centers’ share of global electricity consumption will more than double to 9% by 2030, and McKinsey expects AI-related infrastructure to represent 71% of total North American data center capacity by that year.

Despite advances in chip efficiency, the increasing complexity of AI workloads continues to drive sharp growth in overall power consumption.

We evaluate U.S. data center supply and demand dynamics to identify imbalances. The chart highlights the current and projected shortfall in available capacity relative to projected demand.

“AI has the potential to be more transformative than electricity or fire.”

— Sundar Pichai, CEO of Google and Alphabet 

AI infrastructure spending

AI infrastructure spending is entering a hyper-growth phase, propelled by surging compute demand. Analysts project that the AI infrastructure market could reach between $3 trillion and $5 trillion by the end of the decade, with the majority of investments directed toward building capacity to support AI workloads. 

Meeting this unprecedented level of demand will require immense and sustained capital investment

Bain’s research shows that maintaining sustainable investment levels in the cloud sector would require around $500 billion in annual capital spending to generate roughly $2 trillion in revenue, highlighting the enormous scale of capital intensity in AI and cloud infrastructure. Even with full-scale cloud adoption and efficiency gains from AI-driven optimization, the sector could still face an annual investment shortfall of around $800 billion, underscoring the magnitude of the challenge ahead.

With this in mind, as we track AI throughout 2026, we will emphasize capacity building rather than broad AI deployment

Of course we’ll want to continue to see demand take shape, but the key question going into the next 12 months is whether the AI suppliers can build the required capacity to meet demand projections in future years. Without the build, demand cannot emerge.

We analyze the relationship between hyperscalers’ projected cloud revenues and capital expenditures, anticipating that 2026 will remain a significant year for continued infrastructure buildout. Hyperscaler cloud revenue reflects combined sales from Azure (Microsoft), AWS (Amazon), Google Cloud (Alphabet), and Oracle Cloud Services and Applications.

We see two key capex risk factors to monitor



First, will the short reinvestment cycle for certain AI investments — servers and semiconductors have a life of approximately three to five years — create a capex hole that demand is just unable to fill? 



Second, another “DeepSeek moment” still looms: What if models increase in efficiency enough to call into question the wild levels of capex spending we've already seen? 

“AI is likely to be either the best or worst thing to happen to humanity.”

— Elon Musk, CEO of SpaceX and Tesla

Data center building 

Building a data center is a complex and time-intensive process, typically requiring 18 to 30 months from initial concept to commissioning. The timeline varies based on project scale, site conditions, equipment availability, and how early developers procure key components.

However, global expansion efforts are increasingly constrained by grid connection delays as utilities in high-demand regions struggle to manage capacity backlogs. In many cases, developers face long wait times or even temporary moratoriums on new projects while grid operators work to expand transmission capacity.

According to the International Energy Agency (IEA), grid connection queues now range from one to three years across the U.S., extending up to seven years in certain states such as Virginia. Similar challenges are seen globally, with five-to-seven-year waits in the U.K., up to seven years in Germany, and more than five years in Japan’s Kanto region.

Rising operational costs also reflect the strain on power networks

Between 2019 and 2022, congestion-management expenses tripled in Germany, the U.S., and Great Britain and increased sixfold in the Netherlands, underscoring the mounting pressure on energy infrastructure worldwide.

The existing power constraints could slow the pace of capacity expansion, potentially resulting in an even larger backlog by 2026 or 2027.

The AI economy

In the first half of 2025, personal consumption expenditures contributed 2.11 percentage points to U.S. real GDP growth of 3.18%, lower than 3.81% in 1H 2024 and 4.04% in 1H 2023.

Technology-related spending (information-processing equipment, software, and R&D excluding software) contributed 2.28 percentage points in 1H 2025, surpassing consumer spending for the first time in recent years.

In comparison, technology-related spending contributed only 0.56 percentage points to 4.43% GDP growth in 1H 2024 and 0.38 percentage points to 5.47% in 1H 2023, showing a sharp increase in tech-related investment.

“The future of AI is in our hands.”

— Tim Cook, CEO of Apple

Overview I Economy I Equity market I Bond market I AI I Housing market I Global debt