Show me the money
The phrase “quantum advantage” originally referred to a computational advantage, a speed up over the equivalent classical algorithms. Over the past few years, other types of advantages have surfaced. For example, multiple companies have asserted an energy efficiency advantage over high-performance computing (HPC), thus proposing to reduce electricity bills while also being gentler on the environment. In fact, QEI is an organization that is championing energy efficiency in quantum. Now the California Management Review, the Berkeley-Haas’s Premier Management Journal, has published “Practical Quantum Computing is about More Than Just Hardware,” in which researchers from Accenture and MIT propose a concept of “Quantum Economic Advantage.”
Proposal
Imagine a scenario in which a classical algorithm runs faster than a quantum algorithm. Unfortunately, the classical computing resources to do so are expensive. If the quantum algorithm is capable of solving the problem, and if the cost savings are significant enough to offset the speed disadvantage, then there exists an economic advantage with using the quantum algorithm. The article refers to the criteria as: 1) feasibility, and 2) net algorithmic advantage.
Unlike a quantum computational advantage, which refers to classical computers, quantum economic advantage refers to expensive classical computers. The adjective “expensive” makes all the difference.
The stress on “expensive” neglects accuracy, but it shouldn’t. A classical algorithm might solve a problem precisely, whereas a quantum algorithm will approximate it. We can imagine a choice between an expensive exact solution and a relatively inexpensive close approximation and choosing the inexpensive approximation. Speed, accuracy, and cost are all considerations.
Arguments
After the article was published, several inaccuracies were pointed out on social media. One argument wasn’t an inaccuracy, per se, but it pointed out that the article is quite conservative regarding error correction overhead. In other words, the article assumes that significant quantum computational resources will be required to compete with classical algorithms qualitatively.
The estimate used in the article is indeed a bit outdated. Advancements in quantum error correction (QEC) have gradually reduced this overhead. The argument, therefore, is that the “feasibility” criteria will be met sooner than the article suggests. A quantum economic advantage could be realized with smaller quantum computers than the article suggests. In fact, if published roadmaps are achieved, these devices could start becoming available in as little as a few years.
Assumptions
The notion of a quantum economic advantage makes two big assumptions. First, it assumes that high-end classical computing will remain expensive. And, second, it assumes that quantum computing will remain relatively inexpensive. Neither scenario is guaranteed.
Consider, first, that classical computing resources become less expensive over time. They remain costly, overall, but with greater computational power. Every year, the money in your pocket can buy you more classical computing power than it could in years prior.
On the other hand, quantum computer access is relatively inexpensive because they are still in development. We cannot yet use them to solve commercially useful problems. We can imagine, however, that access to a quantum computer powerful enough to discover novel pharmaceuticals and materials for multi-billion-dollar markets is going to be priced a little less reasonably.
Furthermore, the characterization of quantum computers as relatively inexpensive assumes that they are accessed via the cloud. The least expensive quantum computer, to my knowledge, costs roughly a million dollars. To the other extreme, I also know of one that costs tens of millions of dollars. Since many quantum computers will be on-premises, and we see that happening already, any claims of economic advantage have to take into consideration the total cost of ownership beyond these lofty purchase prices. And since many quantum computers will be partnered with HPC resources, those costs might need to be factored in, as well.
Conclusion
The holy grail of quantum computing continues to be the notion of a significant speedup over comparable classical algorithms. But an advantage is an advantage, even if it’s not the one that was originally sought after. If speed is not too critical and approximations are sufficient, then saving money can be a legitimate benefit, assuming, of course, that quantum computers remain priced to realize such a benefit. Here’s hoping they will.
This is a conversation worth continuing, whether as an addendum to the original article or separately. What will it cost to access or own – including potential electric bill savings – commercially useful quantum computers? And in scenarios where there are no computational advantages – no speedups – what would it actually cost to solve these problems with HPC instead? Instead of simply assuming that the former will be cheap and the latter will not, let’s talk about these costs.
