AI data center energy demand could soar to 123GW by 2035, a thirtyfold increase, according to a report from Deloitte.

Much of this growth is the result of the higher energy demand required for AI applications. The report notes that a five-acre data center equipped with CPUs and GPUs might see its energy usage increase from 5MW to 50MW.

The report was based on an April 2025 survey of 120 US-based power companies and data center executives, with 79 percent of respondents believing that AI will significantly increase power demand through 2035 due to widespread adoption.

The report highlighted seven gaps to powering the AI boom, including grid constraints, supply chain disruptions, security concerns, difficulty in securing permits, workforce constraints, and gas delivery.

Grid constraints were pinpointed by respondents as the primary challenge to sustained growth, with 72 percent of respondents viewing power capacity as very or extremely challenging, especially about mismatches between data center and grid buildout timelines.

The primary issue cited was the fact that power capacity development usually takes significantly longer than data center buildouts. This is especially notable within the natural gas sector, which has accounted for much of the new capacity needed by AI data centers.

However, it is noted that gas generation growth is being impacted by supply chain disruptions, worsened by the tariff regime imposed by the Trump administration. This has led to fears that gas power plants won’t come into service at a quicker timeframe than it would take to secure a grid connection. For example, gas power plant projects that have not already contracted equipment are not expected to become available until the 2030s.

Consequently, the report says that in the same timeframe, renewables and battery storage can account for 92 percent of all planned generating capacity additions to the grid in 2025. However, the transmission that is required to bring this new capacity to the grid could scupper this, often taking more than a decade to build.

Concerns over securing permits and their varying timelines were also cited, with elongated permitting times impacting project schedules and causing price inflation. According to Deliotte, over the past year, contested projects have grown by 29 percent, and local restrictions have grown by 73 percent.

Growing risks of a skill shortage were also noted, especially within the data center sector, with 63 percent of respondents viewing it as the top challenge.

The report advises several strategies in closing the AI infrastructure gap. According to survey respondents, the most important strategies to overcome these challenges are technological innovation, regulatory changes, and more funding.

Innovations include advances in cooling technology, transforming power delivery in chips to reduce power losses, and transmitting power onto the grid in a different way, such as the utilization of grid-enhancing technologies.

Regulatory changes highlighted include system integration measures that would allow for increased transparency across the interconnection network, remove speculative projects from the interconnection queue, and advance priorities. In addition, the report argues that greater stakeholder cooperation between power companies, data centers, and grid operators could support the regulatory process and allow for data centers to meet the demand for their operations more efficiently.