The rapid expansion of AI data centers is driving technology companies toward natural gas power, but the type of turbines they are choosing could significantly increase their environmental impact. Wired reports that developers increasingly favor simple-cycle gas turbines, which are faster to install but less efficient than the combined-cycle systems commonly used by grid-connected power plants.
Amazon’s planned off-grid power plant in Texas illustrates the scale of the problem. The facility is permitted to emit up to 33 million tons of greenhouse gases annually. Even at half that amount, its emissions would exceed those of 78 average-sized natural gas plants. The project will rely entirely on simple-cycle turbines.
Simple-cycle systems burn gas to generate electricity but release substantial waste heat. Combined-cycle plants capture that heat to produce additional electricity through a steam turbine, generating more power from the same fuel and reducing emissions per unit of electricity.
Amazon is not alone. Data center projects associated with Meta, Microsoft, Google, and xAI are also using or planning simple-cycle turbines. The technology appeals to developers because it can be installed more quickly and cheaply, particularly as turbine equipment faces lengthy supply-chain delays.
AI workloads provide another reason. Training large AI models can create sharp fluctuations in electricity demand. Simple-cycle turbines contain fewer moving parts and can start and stop faster than combined-cycle systems, allowing operators to respond more easily to these changing loads. They also generally avoid the substantial water consumption associated with combined-cycle plants.
The trade-off is long-term efficiency. As governments encourage data centers to generate their own electricity rather than burden public grids, developers could create a parallel network of fossil-fuel plants designed to operate for decades. If AI demand declines, some facilities could become stranded assets. If demand continues growing, however, millions of tons of additional greenhouse gas emissions could become embedded in America’s expanding AI infrastructure.
