A new analysis indicates that certain designs for fault-tolerant quantum computers could consume far more energy than the world's most powerful supercomputers. Presented at a recent conference, the estimates highlight a wide range of potential power needs, from modest to enormous. This variation stems from different technologies used to build and operate these machines.
Quantum computing holds promise for tackling complex problems beyond the reach of classical supercomputers, such as accelerating drug discovery. However, achieving practical utility requires scaling up to fault-tolerant quantum computers (FTQCs) with thousands of error-corrected qubits, a challenge that involves diverse engineering approaches.
At the Q2B Silicon Valley conference in Santa Clara, California, on December 9, Olivier Ezratty from the Quantum Energy Initiative presented preliminary energy consumption estimates for these future machines. Drawing from public data, company insights, and models, he outlined a spectrum ranging from 100 kilowatts to 200 megawatts.
For context, the leading supercomputer, El Capitan at Lawrence Livermore National Laboratory in California, draws about 20 megawatts—roughly three times the power used by the nearby city of Livermore, home to 88,000 residents. Ezratty's analysis showed that two FTQC designs, scaled to 4,000 logical qubits, could exceed this, with one potentially requiring 200 megawatts. In contrast, three ongoing designs might use less than 1 megawatt, akin to research supercomputers.
These differences arise from qubit technologies. Superconducting qubits, like those from IBM, demand massive refrigeration. Light-based systems need cooling for photon sources and detectors, while trapped-ion or ultracold atom setups rely on energy-intensive lasers and microwaves.
Oliver Dial of IBM anticipates their large-scale FTQC will need under 2 or 3 megawatts, a small fraction compared to hyperscale AI data centers and possibly less if paired with existing supercomputers. QuEra, focusing on ultracold atoms, projects around 100 kilowatts for theirs. Companies like Xanadu, Google Quantum AI, and PsiQuantum did not comment.
Beyond hardware, error-correction electronics and computation runtime add to the energy load. Ezratty calls for industry standards to measure and report footprints, noting efforts in the US and EU. His work, still early, underscores opportunities for optimization: “There are many, many technical options that could work in favour of reducing the energetic footprint.” Such insights could shape the quantum industry's path, favoring efficient designs.