US data center power demand is set to more than quadruple by 2030, and Rice University researchers say the emissions math only closes if the power plants built to feed that growth capture their own carbon, according to Rice University News's coverage of a new study in the journal Energy & Fuels. Hon Chung Lau, an adjunct professor in Rice's Department of Chemical and Biomolecular Engineering, and Steve C. Tsai of Low Carbon Energies modeled what happens when natural gas combined cycle plants built to power data centers route their carbon dioxide underground instead of up a stack. Their finding: up to three-quarters of the emissions load disappears if operators build the storage infrastructure now.
The baseline is ugly. US data center capacity climbs from 40 gigawatts in 2025 to a projected 169 gigawatts in 2030, and if that growth runs on uncaptured fossil generation, annual carbon dioxide emissions jump from 90 million metric tons to 404 million metric tons over the same five years. Lau and Tsai mapped saline aquifer capacity against that curve and found 34 states hold enough underground storage to bank more than a century of emissions. Start injecting in-state in 2025, and by 2030 roughly 299 million metric tons, about 74% of the projected total, never reaches the atmosphere. Pipe the rest to a neighboring state with spare aquifer capacity and the number clears 90%. Lau said natural gas combined cycle plants with capture "will be the best way to provide power to these data centers."
Here is the part the emissions accounting skips. Amine-based capture is the workhorse chemistry behind NGCC carbon capture, and it runs on heat and cooling in roughly equal measure. Solvent regeneration strips CO2 out with steam. The plant then has to cool that solvent back down before it can absorb another batch, and the captured gas still needs multistage compression with intercooling at every stage before it reaches pipeline pressure. Published DOE pilot data on amine retrofits puts the parasitic load at roughly 15 to 30 percent of net plant output, an estimate drawn from the broader CCS literature rather than this specific paper. That parasitic load shows up as a new cooling tower built at the power plant fence line, invisible to any PUE calculation run on the data hall itself.
That math lands hardest on the exact plants this story is about. Operators are already turning to off-grid gas generation to skip interconnection queues, and stacking a capture train onto those behind-the-meter turbines means budgeting for a second thermal system on a site that was only supposed to host one. Some of that burden is recoverable. The same logic that makes absorption chillers viable for onsite generation could, in theory, offset part of a capture plant's cooling demand. The paper's model stops at the carbon math and leaves the thermal side unexamined.
Data center site selection already runs on power and water. This adds geology. The 34 states with a century of aquifer capacity do not fully overlap with the states where grid interconnection is actually available, and an operator chasing both constraints now has to route pipeline access into the site plan too. Cheap land and a fast interconnect mean nothing if the captured CO2 has nowhere to go.
Carbon capture adds a second cooling loop at the plant that powers the data center, complete with its own water permit, cooling towers, and capex line item. The 169 gigawatts of capacity operators are chasing by 2030 comes with a thermal plant attached to every turbine built to clean up after itself.