Stronger evidence: ForVerdict
The question of whether data centers should be regulated has become urgent as artificial intelligence, cloud computing, and streaming services drive explosive growth in computing demand. Masanet et al. (2020) in Science estimated global data center electricity consumption at roughly 205 TWh in 2018 — about 1% of global electricity — and cautioned that efficiency gains that held consumption flat over the prior decade may not keep pace with AI-driven workload growth. The International Energy Agency projects data center electricity demand could more than double by 2026. Water consumption is an equally pressing concern: Siddik et al. (2021) found U.S. data centers consumed over 7 billion liters of water daily, with many facilities located in drought-prone regions where they compete with agriculture and residential needs. On the opposing side, Shehabi et al. (2020) demonstrated that market forces have been remarkably effective — computing workloads grew six-fold between 2010 and 2018 while energy use rose only 6%, driven by voluntary shifts to hyperscale facilities and virtualization. The Lawrence Berkeley National Laboratory (2016) similarly documented the success of voluntary programs like EPA Energy Star. However, the AI era represents a structural break from these trends. Training a single large language model can consume as much electricity as 100 U.S. homes use in a year, and inference workloads scale with user adoption. Major tech companies including Google and Microsoft have reported that their emissions are rising, not falling, despite renewable energy commitments. Voluntary measures worked when efficiency gains outpaced demand growth, but that balance is shifting. Regulation need not be heavy-handed — transparency requirements for energy and water reporting, minimum efficiency standards, and siting restrictions in water-stressed areas represent proportionate interventions that preserve innovation while addressing genuine externalities. The weight of evidence supports some form of regulatory framework as the scale and environmental footprint of data centers continue to grow.
Ranked by citation count.
The debate
Rachel: Data centers now consume more electricity than many countries. The IEA projects their demand could double by 2026 with AI growth. Google's own emissions report showed a 48% increase since 2019 despite their net-zero pledges. At what point do we stop trusting voluntary commitments and start setting rules?
Kevin: The industry has an extraordinary track record of self-correction. Shehabi's research showed that computing demand grew six-fold in a decade while energy use barely budged. That happened without a single regulation — it happened because efficiency is profitable. Hyperscale operators like Google and Microsoft have every financial incentive to reduce energy costs.
Rachel: That was before generative AI. Training GPT-4 consumed an estimated 50 gigawatt-hours of electricity. Every ChatGPT query uses roughly 10 times the energy of a Google search. The old efficiency playbook doesn't apply when the workloads themselves are fundamentally more energy-intensive.
Kevin: But regulation moves slowly and technology moves fast. By the time you write a rule about PUE standards or cooling requirements, the industry has already leapfrogged to the next generation of chips. Prescriptive regulation risks locking in yesterday's technology. NVIDIA's latest GPUs are already three times more energy-efficient per computation than the prior generation.
Rachel: Nobody is talking about dictating chip design. The case for regulation is about transparency and externalities. Siddik's research found data centers consume over 7 billion liters of water per day in the U.S. alone. Many of these facilities are in Arizona, Texas, and other water-stressed regions. Local communities have no say and often don't even know how much water is being consumed. Basic reporting requirements aren't heavy-handed.
Kevin: I'll grant you transparency — companies should disclose energy and water usage. But the moment you add siting restrictions or efficiency mandates, you slow down the buildout of infrastructure that the entire digital economy depends on. Cloud computing, telemedicine, remote work, autonomous vehicles — all of this requires data center capacity. Regulation that constrains supply will raise costs for everyone.
Rachel: Externalities are real costs too. When a data center in a drought zone draws millions of gallons from the local aquifer, the community pays for that. When a facility runs on a fossil-fuel-heavy grid, the climate costs are externalized. Regulation doesn't mean shutting anything down. It means making companies account for the full cost of their operations.
Kevin: And many already are. Microsoft, Amazon, and Google are the largest corporate purchasers of renewable energy in the world. Apple runs its data centers on 100% renewable power. The market is solving this faster than any regulator could.
Rachel: Corporate renewable energy purchases are often accounting exercises — buying credits rather than actually powering facilities with clean energy in real time. And those commitments haven't stopped total emissions from rising. The voluntary approach had its chance. The scale of what's coming with AI demands a regulatory floor.
Kevin: I just worry that regulation will push investment overseas to countries with weaker standards, making the global problem worse while harming U.S. competitiveness.
Rachel: That's an argument for international coordination, not inaction. The EU has already passed the Energy Efficiency Directive requiring data center reporting. The U.S. can lead or follow, but doing nothing isn't a viable option.
Frequently asked questions
How much electricity do data centers use?
One estimate put global data center use at about 205 TWh in 2018, roughly 1% of global electricity. A 2024 study found U.S. data centers account for over 4% of U.S. electricity, and the International Energy Agency projects data center demand could more than double by 2026.
How much water does a data center use?
U.S. data centers were estimated to consume about 7.1 billion liters of water per day for cooling. A mid-sized facility can use as much water as a small town, while larger ones can require up to 5 million gallons a day, comparable to a city of 50,000 people.
Haven't data centers already become more energy efficient on their own?
Yes, between 2010 and 2018 computing workloads grew six-fold while energy use rose only 6%, driven by hyperscale facilities, virtualization, and voluntary programs. However, AI workloads may outpace these gains, and companies such as Google and Microsoft have reported rising emissions despite renewable energy commitments.
What kind of data center regulation is being proposed?
Proposed measures include transparency requirements for energy and water reporting, minimum efficiency standards, and siting restrictions in water-stressed areas. These are presented as proportionate steps that address environmental impacts without heavy-handed controls.