After two decades of flat demand, electricity use is projected to surge in the United States due primarily to the growth of data centers for artificial intelligence (AI). The number of data centers anticipated to be built and how much electricity they will need are both highly uncertain. What sources of electricity are used to power them has important implications for energy affordability, grid reliability, climate change, and public health. The Union of Concerned Scientists (UCS) used a power sector capacity expansion model developed by the National Renewable Energy Laboratory to analyze different approaches to meeting data center demand growth under varying policies and assumptions. We found that the United States can meet electricity demand from data centers primarily with clean energy, and at the same time, phase down fossil fuel use. But achieving this outcome will require policy intervention. The cost of meeting data center demand between 2026 and 2050 is approximately $900 billion, representing 18 percent of total US wholesale electricity costs. Clean energy policies reduce air pollution and heat-trapping emissions from using fossil fuels, resulting in billions of dollars in health benefits and trillions of dollars of avoided climate damages that outweigh the cost of transitioning to clean energy. In addition to adopting strong clean energy policies, policymakers and regulators should require utilities and data center developers to be more transparent and accountable and to improve long-term planning for meeting data center demand while protecting other customers from cost increases and negative health impacts.