Lately, an increasing amount of discussion in the quantum computing community has been focused on quantifying “quantum energy advantage.” That is, if quantum computers can solve certain problems faster than classical computers, they might also solve these or other problems using less energy. Given seemingly exponential growth in energy usage attributed to data centers, high-performance computing, and training AI models (among others), the idea of a quantum energy advantage seems tantalizing: we can have our cake (big data) and not burn up the planet in the process!
The issue with this is that economics (and human psychology) don’t work that way. Quantum energy advantage is a piece of the story, but ultimately it’s collective human behavior – not the energy cost of specific algorithms – that determines how much total energy is consumed on computing (or any other application).
Simply put: when the cost (in money, time, or any other finite resource) of some action or commodity goes down, demand generally increases until a new equilibrium is established. This phenomenon is known variously throughout the literature as “induced demand” or the “rebound effect.” Depending on the specifics of how the equilibrium is set, the final resource consumption may well be equal to or even higher than the initial consumption!
How might human behavior interact with quantum energy advantage? Consider the case of quantum Bitcoin mining. The number of computers being used at any time to mine Bitcoin (a “proof-of-work” cryptocurrency) is determined almost exclusively by energy costs. If – as a recent paper suggests – quantum computing will make Bitcoin mining more energy-efficient, it will just mean more computers competing for the same number of available Bitcoins. The energy efficiency argument, in this case, is pure greenwashing.
In this article, we will introduce the concept of “induced demand” in a somewhat more familiar circumstance (road construction), and show how the same economic calculus applies to quantum computing. We will conclude with a discussion of how to engage with the topic of quantum energy advantage responsibly.
Induced demand: From road widening to quantum computing
In traffic engineering, it is increasingly known that widening roads does not, in the long term, reduce traffic congestion. Why not? Human behavior! Adding more lanes may temporarily decongest the road, but people who otherwise would have avoided driving at rush hour will be lured back. Within a few years, the same equilibrium level of congestion will be reestablished, if not worse: short-term congestion reduction may shift patterns of development and homebuying, locking in even higher demand moving forward (an example of a broader phenomenon known as “path dependency”).
Analogously, increasing efficiency of classical computers has not led to a reduction in either compute time or energy use. When computing resources become less expensive, we use them for activities (e.g. gaming) that would have seemed frivolous before. There’s no reason to expect that quantum computers will change this calculus.
The energy economics of quantum computing: 3 scenarios
In general, demand for high-performance computing is limited by a combination of physical limitations (some problems are exponentially intractable) and resource costs (e.g. electricity bills). Quantum computers may shift this balance on two fronts. Quantum time advantage will make new classes of calculations tractable. Quantum energy advantage will lower the resource costs of running any particular algorithm. How these “quantum advantages” interact with the economics of computing is a complicated discussion, but we can break applications of quantum computing down into 3 general classes or scenarios.
Scenario 1: Quantum energy advantage actually reduces overall energy consumption. This will be the case wherever computing demand is primarily set by factors other than resource costs. For instance, Google searching already has essentially zero cost to the consumer. If a Grover-esque quantum algorithm reduced the energy intensity of Google searches, induced demand would be minimal because demand is already saturated.
Scenario 2: Quantum energy advantage simply shifts resource utilization around, with the same equilibrium energy usage as before. This scenario will occur whenever demand is set by energy costs in a zero-sum game, such as the Bitcoin mining example discussed earlier.
Scenario 3: Applications where the rebound will likely exceed energy savings. This includes cases where demand is limited by resources only in a positive-sum game (if at all). Quantum portfolio optimization algorithms, even if they use less energy per computation, will likely increase overall energy consumption in the financial sector because the economic incentive is to run ever-more-complicated optimization algorithms until the marginal profit from doing so is zero.
Toward a more sensible analysis of climate impacts of quantum computing
Quantum energy advantage is interesting from a computer science theory and quantum thermodynamics perspective. Yet quantum energy advantage, by itself, means nothing for the planet. It’s how we as humans respond to the availability of quantum computing that sets the overall impact of quantum computing on energy consumption. This point needs to be drilled home in every academic publication or popular science article discussing quantum energy advantage – otherwise we risk being complicit in the cycle of greenwashing and quantum hype that distract from effective climate action now.
Some questions we as researchers can keep in mind when working on quantum energy advantage:
· How are we framing quantum energy advantage in our work? Does our rhetoric run the risk of conflating quantum energy advantage with overall energy consumption or climate impact? Would a term like “per-computation energy advantage” better convey our intentions?
· How is demand for computation in a given sector currently set, and how will quantum computing change this calculus (if at all)? Figuring out whether any given application best maps to Scenario 1, 2, or 3 above will be a useful start.
· What implicit assumptions am I making when it comes to quantifying energy advantage? Are “big O” improvements large enough at a realistic N to compensate for the (potentially huge) constant factor cost of a quantum computer? Have I considered the energy cost of dilution refrigerators, or the overhead associated with error correction?
· What possible policy actions might mitigate the energy used by quantum (or for that matter, classical) computations?
This last bullet is worth re-emphasizing. Per game theory, the only way to disrupt a situation like scenario #3 above – where actors are incentivized to consume ever-increasing resources to chase wealth – is to change the rules of the game.
Recall the highway engineering example. The real solution to congestion is not widening roads. It is either to change the cost structure (for instance by tolling) or to make congestion no longer a barrier to mobility (for instance by building rapid transit). Regarding quantum, a tax on financial transactions might, for instance, change the incentives structure of the financial industry away from hyper-optimized portfolios. Quantum energy advantage will help the planet in scenario #1, but appropriate regulation is the only solution to scenario #3.