As quantum computing inches closer to mainstream enterprise reality, a key question is arising: Could the technology shake up how IT leaders make use of AI?
Much has been made about the problems quantum computing will create for traditional encryption methods, but experts say the technology could also take some pressure off the compute of large AI workloads, with quantum chips augmenting the GPUs and CPUs used today.
The timeframe for availability of general-purpose, error-corrected, or fault-tolerant quantum computers remains up for debate, but some quantum-focused companies have roadmaps for quantum systems to hit the market in the next two to four years.
For example, QuEra has projected that cloud access to a 256-qubit system will be available in 2028, and IBM plans to release its 200-qubit Starling system in 2029. Some critics question those product roadmaps, however. Still, Australian company Silicon Quantum Computing (SQC) is already selling quantum-based chips — called QPUs or quantum machine learning processors — trained to handle specific machine learning applications in industries such as finance, pharmaceuticals, energy, and defense.
Quantum-based chips can already speed up the training of AI models significantly, and quantum computers have the potential to run AI workloads more efficiently than traditional GPUs and CPUs, says Matthew Bradley, SQC’s vice president of corporate development.
“Given the increasing costs of AI, and if one is concerned about the broader power consumption and resource intensity of AI, that’s a really material result,” he explains. “It’s not quite as simple as drawing a straight line — if you do X, you can save Y — but it’s possible you can save a certain amount of model time that is resource-beneficial, both in terms of costs of running the model, but also sort of time and efficiency within the organization.”
Other industry insiders see the same potential. In some cases, quantum computers are already being used alongside AI to significantly improve results, notes Murray Thom, vice president of quantum business innovation at quantum provider D-Wave Systems.
For example, Japanese pharmaceutical company Shionogi has used D-Wave’s quantum computing technology and AI in the drug discovery process, resulting in a 10-fold increase in the number of desirable molecules uncovered, he says. “We see early research and proof-of-concept projects indicating great potential for quantum computers to handle certain elements of AI workloads more efficiently than CPUs and GPUs,” Thom adds. “The opportunity is not for quantum computers to replace CPUs and GPUs, but to complement them by handling parts of AI workflows where quantum approaches may offer advantages in performance, solution quality, or energy efficiency.”
As representatives of the quantum industry, Bradley and Thom certainly have strong motivations to tout the benefits of the technology, but other experts also see potential.
Most of the focus so far on the benefits of quantum AI has been on so-called quantum speed-up, when quantum systems process information faster than traditional computers, enabling enterprises to more readily tackle complex calculations and large data sets.
Quantum computers could also explore multiple solutions at the same time, allowing for accelerated decision-making.
Experts say quantum computers won’t replace traditional hardware in the foreseeable future, but running a hybrid architecture combining QPUs, CPUs, and GPUs could accelerate AI workloads and create some energy and compute efficiencies.
“GPUs and CPUs will continue to support AI workloads, and they will keep getting more capable,” says John Licata, futures director and enterprise quantum lead at ServiceNow. “QPUs come in alongside them, on the specific problems where classical compute runs into limits of scale, time, or scope.”
But operators of AI systems need to think creatively about how to use quantum computing alongside AI to get the full benefits, he adds. IT leaders must envision future AI workloads to find a place for quantum AI, he suggests.
“There is real strategic advantage in simulating multiple future pathways to see where GPUs come under stress, especially with billions of agents coming online in the next few years, greater needs for governance and regulatory compliance, and new ways to deliver autonomous decision-making with more confidence,” he says. “The door that opens is hybrid workloads where GPUs, CPUs, and QPUs each do what they do best.”
Quantum solutions still have some disadvantages, however, that make them unlikely replacements for GPUs in AI tasks, other experts say. GPUs are well suited to the dense matrix operations that dominate current machine learning problems, whereas many proposed quantum algorithms come with significant data-, measurement, and error-correction challenges, says Arjun Kudinoor, quantum security advisor at cybersecurity vendor Protegrity.
“Quantum computers could eventually accelerate some AI workloads, but I would not expect them to broadly replace GPUs,” he adds. “Quantum computers still need major advances in error correction, and any benefits will probably be limited to specific subroutines of larger AI workloads.”
Adnan Masood, chief AI architect at IT services vendor UST, also sees significant potential for quantum computing in the coming years but agrees that there are major technical hurdles ahead for the technology to replace GPUs and CPUs for AI tasks.
Still, he expects hybrid quantum and traditional architecture to arrive in the 2030s, with GPU clusters off structured workloads to QPUs and ingesting the output back.
Instead of focusing on the potential for QPUs to replace GPUs or cut energy consumption, IT leaders should focus on the impact on encryption and on quantum-inspired classical algorithms that run on GPUs, such as tensor network compression, Masood says.
Post-quantum cryptography is a non-speculative approach to quantum computing, he adds. “The threat is ‘harvest now, decrypt later’ — an adversary captures encrypted archives and traffic today and decrypts them once a cryptographically relevant machine exists,” he says.
Masood doesn’t recommend that IT leaders start to hire quantum machine learning teams. Every quantum AI vendor pitch should examine whether they provide real hardware or a simulator, whether they deal with classical data or quantum-native data, and what exactly their product is compared to.
IT leaders should also look for evidence of quantum advances, he says, including peer-reviewed end-to-end quantum ML advantages on classical data, and quantum product timelines delivered on schedule.
“Set your escalation triggers now, so you react to evidence rather than press releases,” he adds.
Protegrity’s Kudinoor, also a doctoral student in MIT’s Center for Theoretical Physics, agrees on the encryption threat, with the potential for quantum computers to break RSA- and ECC-based encryption.
Meanwhile, IT leaders should start to pay attention to the quantum AI space, he suggests. “CIOs and other IT leaders should watch the quantum and AI space with cautious optimism, relying on trusted subject matter experts and avoiding hype-driven investments,” he says.