AI Data Centers Face Backlash as Power Costs Jump 36% Since 2020

AI data centers are drawing scrutiny over electricity and water use, even as operators argue they support jobs, tax revenue and grid investment. The debate matters for utilities, hyperscalers and investors tracking AI infrastructure spending.

AI data centers have become a flashpoint in the broader buildout of artificial intelligence infrastructure, with critics tying new projects to rising power bills and water consumption. The political resistance is notable because it cuts across ideological lines, from conservative communities to progressive activists.

The core tension is simple: demand for computing is accelerating, but local grids and water systems were not built for a surge in hyperscale facilities. Since 2020, residential electricity prices have risen by more than 36%, helping fuel concerns that households could end up subsidizing the next wave of AI investment.

At the same time, supporters argue that AI data centers are an essential part of the modern digital economy, generating tax revenue, supporting high-paying jobs and, in some cases, helping finance long-delayed grid upgrades. For investors, the issue is no longer whether data center demand is growing, but how quickly infrastructure, regulation and public acceptance can keep pace.

Key Facts

  • Residential electricity prices have increased by more than 36% since 2020, intensifying scrutiny of large power users such as AI data centers.
  • Microsoft tested an underwater data center near Scotland’s Orkney Islands in 2018 with 864 servers connected to shore by an undersea cable.
  • After two years, Microsoft found the underwater servers failed at about one-eighth the rate of comparable land-based systems.
  • Large hyperscale operators in this market include Meta, Google and Amazon Web Services, alongside enterprise users such as JPMorgan Chase.
  • Microsoft ended its underwater data center experiment in 2024, despite the project’s positive reliability results.

AI Data Centers

Data centers are the physical backbone of cloud computing, mobile applications and enterprise software. They house servers, networking equipment and storage systems that enable everything from messaging and digital payments to video streaming and AI model training. What has changed is the scale: AI workloads require far more computing power than traditional internet services, increasing the need for larger and more energy-intensive facilities.

That shift has put hyperscalers under pressure. Operators such as Meta, Google and Amazon Web Services are expanding capacity to serve both private-sector and public-sector clients, while companies building AI models need increasingly dense computing clusters. The result is a new category of local economic debate, where residents weigh the promise of jobs and tax revenue against concerns over utility bills, land use and water availability.

The issue matters far beyond one town or one project. Utilities must decide how to allocate capital for transmission, substations and generation. Technology companies must show they can scale responsibly without transferring costs to households. Regulators and local governments, meanwhile, face a balancing act between attracting investment and protecting ratepayers. This makes AI data centers a critical intersection of digital infrastructure policy, utility economics and capital spending trends.

AI data centers are becoming a test of whether the next phase of digital growth can expand without pushing higher costs onto local communities.

Why the power and water debate is intensifying

One argument from data center supporters is that the real weakness lies in the grid itself. After roughly two decades of relatively flat electricity demand, many utilities did not invest aggressively in new infrastructure. The surge in AI-related power demand is now exposing those constraints. From this perspective, large data center customers can actually support the economics of upgrades by adding long-term load and contributing more to fixed system costs.

Operators are also trying to blunt criticism through technology and contractual commitments. Some companies have pledged to cover incremental electricity costs linked to their AI facilities, while others are turning to recycled water for cooling to reduce pressure on freshwater resources. Alternative concepts, including offshore or underwater data center designs, suggest the industry is still exploring ways to add capacity while limiting direct strain on populated areas.

Implications for Investors

For investors, the immediate takeaway is that AI infrastructure remains a major capital expenditure theme, but execution risk is rising. Hyperscalers, utilities, equipment suppliers and engineering firms all stand to benefit from sustained demand for computing capacity. However, permitting delays, community opposition, transmission bottlenecks and environmental restrictions could stretch project timelines and affect returns.

Utility investors should pay close attention to who bears the cost of upgrades. If regulators allow new large-load customers to fund a meaningful share of new transmission, substations and generation, utilities may gain a clearer path to rate-base growth without provoking a political backlash from residential customers. If that cost allocation becomes contentious, utilities in fast-growing data center markets could face sharper scrutiny and slower approvals.

Investors in major technology companies should also watch how operators address water use, energy sourcing and reliability. Companies that secure recycled water access, add renewable or alternative power contracts, and demonstrate cost-sharing commitments may face fewer obstacles in expanding AI capacity. The same is true for businesses exposed to cooling systems, backup power, grid equipment and high-density chip infrastructure, where demand is likely to remain strong even if individual projects are delayed.

Longer term, experimental designs such as underwater data centers remain worth monitoring. Microsoft’s test near the Orkney Islands showed notably lower server failure rates, suggesting there may be operational advantages in sealed environments with stable temperatures and less human interference. Even though the project ended in 2024, the findings hint at a broader industry search for more efficient deployment models.

The next phase of AI growth will depend not just on chips and software, but on whether power, water and local politics can support the physical expansion of AI data centers. Investors should watch utility filings, permitting decisions and hyperscaler spending plans closely as the infrastructure race intensifies.

Ultima Markets