Firgun Ventures Insight

Quantum computing’s advantage in the energy paradigm

Dr. Kris Naudts, Zeynep Koruturk, and Donald Harmitt

Artificial intelligence is reshaping the digital world at breathtaking speed, powered by vast data centres, ultra-fast networks, and ever-denser compute clusters. Each new generation of AI models requires significantly more processing power than the last, and the world's digital infrastructure is already encroaching on the physical and energy limits of what the grid can sustain. According to the World Economic Forum, “a single cutting-edge AI chip can now draw as much electricity as an entire household, and large training runs can consume as much energy as a small city.”

The system-level numbers are equally alarming. The International Energy Agency (IEA) forecast that AI data centres could triple electricity consumption by 2030, and Gartner has predicted that power shortages will restrict 40% of AI data centres by 2027. For a sector that has spent the last decade chasing performance at almost any cost, the binding constraint is no longer chip supply or model architecture, but instead will soon become a matter of watts.

This article looks at how quantum computing fits into that picture, not as a silver bullet nor as a replacement for classical compute, but as a potentially critical piece of the sustainability puzzle for a digital economy running out of energy headroom. The piece will cover where classical computing hits a wall, why quantum systems scale differently, how today's quantum hardware platforms actually compare on power draw, and where the most credible near-term energy wins are likely to emerge.

Why classical compute hits an energy wall

Many of the computational problems that matter most commercially, in machine learning, optimisation, and simulation, have a problem space that grows exponentially with the size of the input. Classical algorithms often face inherent limits in solving these problems efficiently, and the workaround has historically been to throw more silicon at the problem. The computational prowess and power consumption of a supercomputer scale almost linearly with processor count. To double the computational speed, you need roughly twice the processors, and therefore roughly twice the energy. This is the quiet mathematics behind the data centre boom we are witnessing today.

As time complexity grows, energy consumption grows with it, because the processors are running for longer. As the data itself grows in size and complexity, the number of operations required grows faster with it. Large problems also require extensive memory and storage, each with its own power footprint. The Frontier supercomputer at Oak Ridge National Laboratory, one of the most powerful supercomputers in the world, requires 21.1 megawatts to operate. At average US commercial electricity rates (12.75c per kWh in 2024), that translates to an estimated annual energy bill of over US$23 million, for a single machine, despite being regarded as a highly efficient supercomputer. Every major hyperscaler now faces net-zero commitments that are increasingly difficult to reconcile with the compute demands of frontier AI.

A fundamentally different computational substrate

Quantum computers compute using fundamentally different processes from classical machines. Classical computers manipulate bits, which are either zero or one. Quantum computers manipulate qubits, which, thanks to superposition, can exist in a combination of both states at once, and thanks to entanglement, can be correlated with each other in ways that have no classical analogue. The practical consequence is that for the same task, and even for the same running time, the number of underlying operations is dramatically different.

The computational capacity of a quantum machine grows exponentially with qubit count, but critically, this exponential growth does not translate into exponential growth in power consumption. A quantum computer's energy footprint comes from three main places: the quantum processor itself where the qubits live and compute, the cooling infrastructure (largely specific to superconducting systems that need near-absolute-zero temperatures), and the readout and control electronics that operate the system, manage error correction, and handle the overhead of running a computation. None of these areas scale the way adding processors to a supercomputer scales.

Previous modelling efforts in 2023 by quantum computing firm, QuEra, suggests that if a single quantum computer were to handle just 5% of the tasks currently run on a supercomputer, it could yield annual savings of over US$1 million and reduce energy consumption by nearly 10 gigawatt-hours per year. On a mass scale, this begins to make a significant impact on commercial bottom lines within the hundreds of millions bracket, alongside tremendous energy savings, which is particularly relevant given current geopolitical headwinds. Viewed through this lens, quantum computing presents itself as a plausible pathway for sustaining digital progress without a matching rise in electricity demand.

Not every problem is a quantum problem

Not every problem is well-suited to a quantum solution, and energy efficiency is a good illustration of why. Quantum computing can meaningfully reduce power consumption for the right class of problem, but for smaller, more straightforward tasks, or tasks well-structured for classical hardware, the additional overhead of error correction and control electronics tips the balance the other way.

Where quantum becomes genuinely plausible is when the problem class itself is hard in a very specific way. Four algorithmic families in quantum computing help illustrate this point. Shor's algorithm can factor large numbers exponentially faster than the best-known classical methods, with obvious implications for cryptographic workloads. Grover's algorithm delivers a quadratic speedup for unstructured search, relevant for database queries and lookup-heavy applications. The Quantum Approximate Optimisation Algorithm, or QAOA, targets hard combinatorial optimisation problems such as scheduling, routing, and financial portfolio construction. Lastly, quantum simulation methods can model complex quantum systems with a level of efficiency that classical simulation cannot match, which is where the materials science and drug discovery applications come in.

It is important to note that these advantages are not uniformly demonstrated at commercial scale today. Current quantum systems are noisy and require extensive error correction, and for a factoring problem of a one-million-digit number, for instance, today's quantum approach would consume significantly more energy than its classical counterpart because the overheads currently outweigh the theoretical speedup.

With that being said, the direction of travel is highly promising. In Dr. Kris Naudts’ Time to Talk Quantum: Quantum in the Age of AI podcast episode, with Dr. Robert Sutor & Yariv Adan, Dr. Robert Sutor mentioned that there are a number of studies trying to answer the question of “what are the energy needs of a quantum computer vs. a supercomputer?” An example of this is a paper by Florian Meier and Hayata Yamasaki, “Energy-Consumption Advantage of Quantum Computation”, which builds a detailed accounting of everything a computer actually consumes to run a calculation. The detailed account includes not just the processing itself, but the energy needed to control the system and, crucially, to erase and reset its memory between tasks (a physical cost often overlooked in energy comparisons). Using this framework, the authors prove that for certain classes of problem, quantum systems can achieve an exponential energy advantage over classical ones, meaning that as problems grow in complexity, the gap in power consumption widens dramatically in quantum's favour. Perhaps more importantly, the methodology itself gives the field a technology-independent framework for assessing computational sustainability, and hints at a future in which quantum supremacy is defined not by speed alone but by power efficiency.

Hardware platform matters in the long run

Understanding energy efficiency in quantum requires comparing not just algorithms but the hardware platforms that implement them. Whether a platform depends on large dilution refrigerators or runs at or near room temperature is a major determinant of its long-term energy profile, and the differences between leading approaches are already stark. Any comparison of present-day machines has to start with an important caveat, which is that public power figures for quantum hardware are not yet standardised the way they are for chips or supercomputers, and power draw is not the same thing as energy per useful answer. Even so, the differences across hardware modalities are already large enough to matter.

Superconducting quantum computers, the modality favoured by IBM, Google, Rigetti, and IQM among others, rely on circuits that operate only at temperatures close to absolute zero. This entails multi-stage dilution refrigerators, and in current cloud-accessible systems, total power consumption of roughly 25 kilowatts. Most of that electricity goes into refrigeration and supporting equipment, not the qubits themselves. For superconducting machines, the energy required for cooling is significantly larger than that required for computation, a complete reversal of the pattern seen in conventional computing. Designing a sustainable superconducting quantum computer therefore means both improving cooling efficiency and minimising the need for cooling in the first place.

Neutral-atom and trapped-ion platforms tell a different story. Neutral-atom systems, such as QuEra's 256-qubit Aquila, operate at or near room temperature and avoid millikelvin cryogenics entirely. Aquila, one of the largest publicly accessible neutral atom quantum computers in the world, draws less than 7 kilowatts, which is under 0.05% of the Frontier supercomputer's consumption. PASQAL's neutral-atom processor, operating at the scale of hundreds of qubits, consumes around 2.6 kilowatts in total, roughly half of which goes to the lasers and the rest split between electronics and environmental control. OQC's Toshiko Gen1 superconducting system, by contrast, uses 35 kilowatts per system, with power draw expected to reach around 40 kilowatts as it scales towards hundreds of qubits.

Even at today's modest scales, two quantum computers with comparable processor sizes can differ by roughly a factor of 10 or more in power draw, depending on whether they rely on extreme cooling. The picture becomes even more notable at scale, as QuEra projects that a 10,000-qubit neutral-atom machine, a system more powerful than most of today's supercomputers, would still consume less than 10 kilowatts. That is under one-tenth of one percent of the power draw of the most capable classical supercomputers on earth today. In a similar vein, McKinsey has estimated that at scale, the power requirements of trapped-ion and neutral atom systems would be 10,000x less than a supercomputer, demonstrating the immense energy savings expected as quantum computers approach the fault-tolerance era. Taking into account an error factor of 10-100x, this would still yield significant savings in power demand and annual electricity costs.

Where the real energy wins come from

It is tempting to frame quantum computing's energy story purely in terms of the machines themselves, but the more consequential case sits at the system level. In the use cases that matter most, the electricity consumed by a quantum algorithm is trivial compared with the system-level savings it could unlock across the wider economy. Even modest efficiency gains can yield enormous reductions in resource use and emissions.

Take the Haber-Bosch process, which produces the ammonia that fertilises most of the world's crops and is responsible for a meaningful share of global natural gas consumption (IEA estimates that ammonia production accounts for around 2% of total final energy consumption worldwide). Quantum simulation could, in principle, identify more efficient catalysts that reduce the energy intensity of the reaction, a prize with obvious geopolitical resonance at a time when natural gas supplies are shaped by events in the Strait of Hormuz and the broader Middle East. Logistics optimisation is another substantial lever, with quantum-optimised routing promising fuel savings across shipping, aviation, and last-mile delivery. Materials science offers perhaps the deepest pathway: better batteries, more efficient solar cells, improved catalysts and membranes for hydrogen production, and novel chemistries for EV battery packs. Each of these is a multi-gigawatt-scale problem at the economy level.


Four use cases where quantum could meaningfully shift the energy equation

One further point worth keeping in view is that quantum technologies are not designed to replace classical CPUs; they are designed to work alongside them. In a hybrid future where quantum processors handle the specific classes of problem where they have a clear advantage and classical silicon handles everything else, the overall energy footprint of computing could meaningfully decline.

Doing more with less will be paramount into the future

The quantum sector's energy narrative is still developing, but the contours are clear. Hardware platform choice will have compounding implications for operating cost, site flexibility, and carbon intensity as quantum moves from lab to enterprise. Algorithms will matter, but so will data centre economics, cooling supply chains, and the grid itself. At a moment when the world's appetite for compute is outpacing the grid's ability to supply it, the ability to do more with dramatically less power becomes a strategic consideration at mass scale. This framing will need to sit at the centre of how the quantum ecosystem thinks about the hardware platforms, algorithms, and infrastructure bets that will define the next chapter of the industry.

Insights