The Architecture of a New Era
Data centers of the future will be built, justified, approved, operated, and governed differently than current facilities, according to Henrique Cecci, a vice president analyst at Gartner. Speaking at the Gartner IT Symposium this week, Cecci urged chief information officers (CIOs) to make decisions flexible enough to adapt over a 20-year horizon.
The shift in construction and operational models is driven by rapidly evolving technical demands that are outpacing traditional infrastructure capabilities. Cecci noted that power density in data centers has historically risen from five to eight kilowatts per cabinet, reaching over 10–12 kilowatts today. He projected that future facilities will require 250 kilowatts per cabinet, a massive leap that necessitates entirely new engineering and cooling standards.
These technical changes are accompanied by significant economic pressures as the industry scales to meet demand. Gartner figures indicate that AI providers are preparing to spend nearly $1 trillion on AI infrastructure in 2026, with a further $6 trillion forecast between 2027 and 2030. To manage these escalating costs, Cecci argued that inference workloads are growing more than four or five times faster than training workloads. He suggested that modular, prefabricated options can reduce the cost of distributed capacity, offering a pathway to scale without prohibitive capital expenditure.
Barriers to Expansion and Cost Volatility
The path to building this new infrastructure is fraught with obstacles. Gartner research reported that $130 billion worth of projects were blocked or delayed in the first quarter of 2026, highlighting significant friction in the development pipeline. The firm also projected that data center electricity demand will climb toward 1,218 terawatt-hours (TWh) by 2030, straining existing grid capacities.
Financial planning for hardware must account for these disruptions and market volatility. Gartner’s strategic planning assumption holds that hardware costs for general-purpose data center servers will run 50% to 125% above mid-2025 levels through mid-2027, creating a challenging environment for budgeting and procurement.
Gartner identified four primary forces reshaping the industry: market economics, power and technology, regulation and social license, and geopolitics and sovereignty. Among these, power availability has become the decisive factor for facility location, surpassing network connectivity in importance, Cecci said. This shift means that sites are now chosen primarily for their energy resources rather than just their digital infrastructure.
Sovereignty and Geopolitical Realities
Geopolitical factors are increasingly influencing cloud adoption strategies and data residency requirements. Sixty percent of organizations expect geopolitical factors to increase their future use of cloud services, according to Cecci, reflecting a complex landscape where data movement is constrained by national borders.
Sovereignty concerns are pushing some organizations toward non-U.S.-owned facilities or the construction of their own data centers. This trend is driven by a desire for control and compliance with local regulations. Despite the trend toward cloud and distributed infrastructure, Cecci argued that on-premises data centers remain viable for governments, large banks, and large organizations seeking sovereignty and low-latency benefits.
Policy Frameworks and National Goals
Government bodies are beginning to formalize policies to address these infrastructure needs and manage the associated environmental and economic impacts. The UK government published the first draft National Policy Statement for data centers in October, aiming to streamline approvals while ensuring sustainable growth.
This policy framework supports a committed pipeline of more than 14 gigawatts (GW) of capacity, with a goal of at least 6 GW of AI-capable capacity by 2030. Such targets indicate a coordinated effort between public and private sectors to ensure that national infrastructure can support the next generation of computing power.