Data center electricity consumption is expected to quadruple by 2035: the power crisis and infrastructure reconstruction driven by AI computing power

Introduction

The explosive growth in AI computing demand is reshaping the energy landscape of global data centers. According to the latest BloombergNEF report, by 2035, electricity consumption by U.S. data centers will quadruple, reaching 20% of the nation's total power generation. This is not an isolated forecast—institutions such as EPRI and S&P Global are also continuously raising their projections. For enterprise IT decision-makers, the scarcity and rising cost of electricity supply will become key variables affecting data center site selection, architecture design, and long-term costs.

Event Background

On July 21, 2026, BloombergNEF released a report stating that the expansion of AI computing power will drive U.S. data center capacity to nearly 200 gigawatts over the next decade, with nearly half used for AI training and inference. The United States will continue to dominate global AI chip deployment, hosting 64% of the world's AI chip electricity demand by 2033. The report also shows that this forecast is 83% higher than BloombergNEF's estimate from December 2025, reflecting an acceleration in AI investment enthusiasm beyond expectations.

Technical Analysis: Why Is AI So Power-Hungry?

The core of AI computing is GPU clusters. Taking the NVIDIA H100 as an example, a single GPU can consume up to 700 watts, while a typical AI training cluster contains thousands or even tens of thousands of GPUs, with total power consumption reaching tens of megawatts. Unlike traditional cloud computing workloads, AI training requires long-duration, high-load operation, causing data center power density to surge from the conventional 6–10 kW per rack to over 40 kW or even higher. Liquid cooling technology becomes a necessity but also increases infrastructure complexity. Although inference consumes less power per instance, large-scale deployment still drives up total electricity demand. Additionally, the continuous iteration of AI models (e.g., doubling model size) further intensifies the hunger for computing power.

Analysis of Enterprise Impact

Cost Impact: CAPEX and OPEX Both Rise

The share of electricity costs in the total cost of ownership (TCO) of data centers will increase from the current 20–30% to over 50%. Enterprises need to reassess site selection for new data centers, prioritizing locations near cheap renewable energy sources (e.g., the U.S. Midwest and Southwest). Meanwhile, power capacity expansion for existing data centers may face grid connection bottlenecks, leading to longer construction cycles.

Deployment and Operations Impact

Grid bottlenecks force enterprises to adopt distributed strategies: deploying smaller data centers across multiple regions or signing long-term power purchase agreements with utility companies. The widespread adoption of liquid cooling requires operations teams to acquire new skills, while the energy consumption of cooling systems also needs optimization.

Security and Compliance

Unstable power supply may lead to service interruptions, requiring enterprises to establish more robust backup power and disaster recovery mechanisms. In addition, multiple regions (e.g., the EU and some U.S. states) are tightening energy efficiency regulations for data centers, so enterprises need to plan in advance for carbon footprint reporting and renewable energy certificate procurement.

Market Competition Analysis

Cloud Vendor Competition## Market Competition Analysis

Cloud Provider Competition

Hyperscale cloud providers (AWS, Azure, Google Cloud) are making massive investments in building their own data centers and renewable energy. By locking in long-term power purchase agreements (e.g., partnerships with nuclear and wind power projects), they gain cost advantages. Meanwhile, small and medium-sized cloud providers and colocation providers may face dual pressures of rising electricity costs and supply shortages, forcing them to pivot toward edge computing or regional markets.

Data Center Operators

Operators such as Equinix and Digital Realty are accelerating power capacity expansion, but grid interconnection bottlenecks have become a constraint. Operators with pre-secured power capacity will gain a competitive edge. For example, the PJM grid has a backlog of interconnection requests, causing new data center construction lead times to extend to 3-5 years.

AI Chips and Hardware

Chipmakers like NVIDIA, AMD, and Intel are alleviating power pressure by improving energy efficiency (e.g., power consumption improvements in H200 and MI300X), but demand growth still outpaces efficiency gains. Liquid cooling solution providers (e.g., CoolIT, Vertiv) will see market opportunities.

Industry Trend Observations

From Centralized to Distributed: Data Center Layout Evolution Driven by Power

Traditionally, data centers were concentrated in Virginia (PJM region), but today's power shortages are pushing new construction toward Texas (ERCOT), the Midwest, and the West. Enterprises are beginning to adopt a "build where power is available" strategy, prioritizing power accessibility and pricing when choosing data center locations.

Renewable Energy and Nuclear Revival

AI giants are collaborating with utility companies to build dedicated solar, wind, and nuclear power plants. Microsoft has signed power purchase agreements for small modular reactors (SMRs), while Google and Amazon are also investing in nuclear energy. This signals that the data center industry will evolve from a pure consumer to a co-builder of energy infrastructure.

Grid Modernization Is Urgent

Grid upgrades require massive investment, and approval processes are slow. The plight of PJM serves as a warning to the industry: power infrastructure could become the biggest bottleneck for AI development. Governments and enterprises need to collaborate to streamline interconnection processes and accelerate transmission line construction.

CloudTechDaily Insight

The explosive growth in data center power demand is not just an energy industry issue; it is a core variable in enterprise IT strategy. Over the next five years, electricity costs will replace hardware costs as the single largest factor in total cost of ownership for data centers. Enterprises must prioritize power accessibility, renewable energy availability, and grid stability in their data center site selection, alongside traditional models that rely on latency or bandwidth.We believe that the competition in AI infrastructure is shifting from pure computing performance to a comprehensive game of "computing power + electricity." Companies that can secure cheap, stable, and green electricity first will gain a significant cost advantage in the next decade. At the same time, technological innovations such as liquid cooling and smart power distribution will accelerate deployment, but the most critical factor is the modernization of grid infrastructure—which requires coordinated action from policymakers, utilities, and tech giants.

For enterprise CIOs and CTOs, now is the time to start evaluating existing data center power contracts, planning next-phase capacity, and exploring distributed deployment strategies. The potential of AI is beyond doubt, but its sustainability depends on whether we can solve the fundamental issue of power supply.