This research estimates the associated primary energy consumption for different combinations of dynamic switching energy for storage (J/bit), for processing (J/computation) and global computation intensity (computations/s). Bottom-up extrapolation is used. With ≈10 ronnacomputations/s in 2050, the J/computation should improve ≈36% per year from 2024 to 2050 to ≈50 attoJ/computation to keep computing processing energy flat. Also the dynamic switching energy needs a similar reduction rate to keep the computing processing energy flat. Hence, a paradigm shift is required to reduce the power used by computing. Technology could keep the use of primary energy flat if global computations and stored data grow slower than expected and especially if 50 aJ/computation and 1 nJ/bit can be delivered in 2050.