An Economic Analysis of Quantum Computing
Most analysis of quantum computing asks whether a machine can be built. This paper asks what it would cost to build and run one that is commercially useful, and what customers could realistically pay for it. It applies a standard unit-economics analysis to the five main ways of building a quantum computer: superconducting, photonic, trapped-ion, neutral-atom and spins-in-silicon qubits.
Download the full paper (PDF, 30 pages)
Key findings
- Commercial usefulness starts at around 4,000 logical qubits. Drawing on DARPA's Quantum Benchmarking Initiative, the paper takes about 4,000 error-corrected (logical) qubits as the baseline for commercially useful problems. With today's error-correction overheads, that means millions of physical qubits.
- The number of modules drives the cost. A module is one quantum processor plus the cooling and control equipment it needs. How many qubits fit on a single chip decides how many modules a machine needs, and so most of its capital and running costs. Most approaches would need hundreds to thousands of modules at this scale; spins in silicon aims to fit on one chip.
- The cost gap is orders of magnitude. Building a 4,000-logical-qubit machine is estimated at tens of millions to billions of dollars for the other four approaches, against millions for spins in silicon. Annual running costs range from millions to hundreds of millions of dollars, against hundreds of thousands.
- So are size and power. The same machines would need 200 to 20,000 m² of floor space and 4 to 160 MW of power, against about 5 m² and 0.4 MW for spins in silicon. For comparison, the most power-hungry supercomputer, Aurora, uses about 40 MW.
- Most approaches cost more than buyers can pay. Benchmarked against what governments pay for supercomputers and what large companies could spend from their R&D budgets, most approaches fall outside what customers can afford, even if they succeed technically.
- Conclusion: at the time of writing, only single-module approaches, currently spins in silicon, showed a realistic path to commercially viable quantum computing.
The estimates are order-of-magnitude and point-in-time, and use the most optimistic published inputs for each approach. The paper also includes 15 questions to ask any quantum computing hardware company.
Each square's area is proportional to floor space. Source: Table 4 of the paper.
| Superconducting | 20,000 m² | a large data center |
|---|---|---|
| Photonics | 4,000 m² | a small data center |
| Ion traps | 625 m² | a tennis court |
| Neutral atoms | 200 m² | half a basketball court |
| Spins in silicon | 5 m² | a king-size bed |
About the paper
Published by ICM in April 2025; written by Matthew Gould and Kate Prebble, both then at ICM.
Disclosures: at the time of writing, ICM had investments in the quantum computing companies Diraq (spins in silicon) and Q-CTRL. Kate later joined Diraq.