AI & ENERGY

Problem

AI and data centers are rapidly accelerating their massive electricity and water consumption, posing urgent challenges for Oregon's infrastructure, environment, and communities.

Electricity

• Globally, data centers use about 415 terawatt-hours annually, with U.S. use reaching 176 terawatt-hours.

• In the U.S., use hit 176 terawatt-hours in 2023 (4–5 percent of national demand).

• Energy demand is projected to double or triple by 2028.

• In Oregon, data centers already use about 11 percent of statewide electricity.

Oregon’s power mix:

• 42 percent hydropower

• 38 percent natural gas

• 15 percent wind

• The remainder is solar, geothermal, and biomass.

Result: Without rules, more data centers = more gas plants, higher rates, grid strain, and pressure on rivers.

Water

• U.S. data centers used ~17 billion gallons per year for cooling.

• The Dalles example: Google consumed so much water that the city tried to hide usage through NDAs and lawsuits.

Model Training + Use

• Training one GPT-scale model required about 1,287 megawatt-hours and 500–550 tons of CO₂.

• Inference — everyday user requests — now consumes most of the energy, estimated at 60 percent of total AI use.

• A single AI query uses multiple times the energy of a standard web search.

Solutions

1. Mandatory transparency

Require large AI companies and data centers to publicly report:

• Annual and peak electricity use

• Energy sources and emissions

• Total water withdrawals and cooling needs

• Environmental impact of large training runs

• Estimated energy per user/inference on public models
  

• Tie the federal incentives to no NDAs with local governments.

Result: Communities can plan — and say no to harmful projects.

2. Clean energy requirements

Condition future data center projects on:

• New renewable generation added to the grid (not just credits)

• Independent emissions verification

• Real decarbonization timelines

Use federal funds to:

• Expand transmission

• Add storage

• Enable flexible load shifting.

Result: AI growth does not cannibalize Oregon’s clean energy progress.

3. Water protections

Establish:

• Federal water efficiency standards for cooling

• Drought and watershed risk disclosures

• Consultation with tribes and affected communities

• Community benefit agreements for federally supported projects

Result: Protect water basins, agriculture, households, and salmon-bearing rivers.

4. Energy-smart AI

Support:

• Efficiency benchmarks and labeling (AI “Energy Star”)

• Public funding only for models with energy budgets and reuse plans

• Research that reduces energy per task

• Smaller, open models that local governments and schools can run without mega-facilities.
  Result: Innovation serves public needs — not just large, compute-hungry corporate systems.

5. Shield communities and workers

Require:

• No rate hikes on households and small businesses to subsidize AI

• Fair property taxation

• Local reinvestment in grid upgrades, housing, and cooling centers

• Union labor and apprenticeships on publicly supported projects

Result: Communities benefit — not just data center shareholders.

Bottom Line

AI can solve real problems — but without rules, it becomes:

• A drain on power and water

• A driver of higher bills

• A fossil-fuel growth engine

• A burden on working families and small communities

To ensure AI supports innovation without sacrificing Oregon’s grid, rivers, or ratepayers, we must regulate to require transparency, clean-energy standards, water safeguards, and labor standards. Oregon can scale AI responsibly — supporting innovation without sacrificing our grid, rivers, or ratepayers.